Open-access Harnessing the power of artificial intelligence in weed science: A multifaceted review

Abstract

Background:  This review article comprehensively discusses the various applications of artificial intelligence (AI) in weed management across different facets related to agriculture.

Objective:  It aims to explore the advances of AI in weed science, emphasizing technologies that involve AI, their potential for the future, and the challenges and opportunities they present for sustainable weed management.

Methods:  The method used was the narrative review, describing, synthesizing, and discussing the topics with the existing literature.

Results:  The article delves into the transformative impact of AI technologies on various successful cases in the weed management process, including weed detection, monitoring, precise herbicide application and herbicide discovery.

Conclusions:  This review synthesizes recent advances and discusses future prospects in the rapidly evolving field of artificial intelligence-assisted weed management, emphasizing the role of computational approaches in improving weed detection, decision-making, and control strategies.

Keywords:
Artificial intelligence; Weed science; Weed management; Sustainable agriculture; Emerging technologies

1. Introduction

The concept of “artificial intelligence” (AI) emerged in the mid-1950s and was formally advanced through the 1956 Dartmouth Conference, where John McCarthy and colleagues proposed that machines could be developed to simulate human learning and intelligence (McCarthy et al., 1955). This landmark proposal contributed to the establishment of AI as a distinct field of research and helped popularize the term. AI has evolved into a multidisciplinary field, serving as a powerful tool for problem-solving that often exceeds human intellectual capabilities (Banerjee et al., 2018). In agricultural science, an important milestone occurred in 1985, when AI was used for the first time to optimize cotton production with the creation of GOSSYM, a sophisticated simulation model integrating variables such as irrigation, fertilization, weed control, cultivation, climate, and other factors (McKinion and Lemmon, 1985).

AI is a rapidly evolving multidisciplinary field that encompasses machine learning (ML), data analysis based on previous experiences, object recognition, natural language processing, computer vision, image and video processing, multimedia applications, and other related technologies (Jha et al., 2019). The implementation process of AI-based tools involves several stages, including data compilation, processing, model training, and optimization (Evert et al., 2017). AI applied to agriculture has been shown to enhance crop protection practices and strengthen production systems. AI can complement the methodologies already available in discovering, predicting and developing new pesticide molecules. ML supports the construction of systems based on logical algorithms, artificial neural networks (ANN), and other techniques. These models or systems use high-tech computerized systems to calculate different parameters of crop health and weed detection (Jha et al., 2019).

AI has brought remarkable improvements in agriculture, including crop production, harvesting, and marketing. Traditional farming practices can be automated using robots, sensors, drones, and algorithms, highlighting the role of AI in improving limited farming practices and land efficiency (Smith, 2020). AI, along with ML, deep learning (DL), and neural networks, are driving paradigm shifts in agriculture. These technologies, combined with big data and the Internet of Things (IoT), are being used to explore, understand, and analyze agricultural processes using large volumes of data (Sood et al., 2022; Wolfert et al., 2017). The automation of traditional agricultural operations has enabled more accurate and timely decision-making. Processes that once required several days of data collection and specialized expertise for interpretation can now be streamlined through automated data analysis (Sood et al., 2022). This approach integrates technologies such as infrared cameras, sensors, and unmanned aerial vehicles (UAVs), improving the efficiency and consistency of data collection while reducing reliance on labor-intensive manual assessments. This analysis uses sensors such as infrared cameras, as well as unmanned aerial vehicles (UAVs), to replace humans in data collection. These tools use algorithms to improve predictions of crop yield, pest occurrence, weather conditions, and irrigation needs (Sood et al., 2022). The algorithms are implemented on mechanical platforms (robots, drones, etc.) for real-time weed identification and management, reducing the use of agrochemicals and the proliferation of herbicide-resistant weeds (Jordan and Jannink, 1997). For example, the use of ANN has resulted in the recognition of cultivated plants and weeds with an accuracy between 84 and 99% (Shah et al., 2021).

There are myriad of possible applications of AI in agriculture and several other fields. Our review article aims to explore the advances of AI in weed science, emphasizing the advanced technologies involved, their potential for the future, and the challenges and opportunities they present for sustainable weed management. It aims to expose cutting-edge technologies and speculate on various future applications of this powerful technology in weed science.

1.1 Setting the stage

To set the stage for our review, it is vital to establish some definitions, terminologies, and jargon that are not common in weed science and are becoming increasingly frequent in scientific papers and general media.

Artificial Intelligence (AI), was defined by John McCarthy, as the science and engineering of creating intelligent machines, especially intelligent computer programs; for this purpose, AI uses computers to simulate human intelligence without having to limit itself to biological observation methods (McCarthy, 2007).

Machine Learning (ML), is a branch of AI and computer science that focuses on using data and algorithms to mimic the way humans learn, gradually improving their accuracy (International Business Machine, 2022).

Deep Learning (DL), is a subset of ML, essentially comprising neural networks with three or more layers, which attempt to simulate the behavior of the human brain (International Business Machine, 2020a).

Algorithm, informally is any well-defined computational procedure that takes some value, or set of values, as input and produces some value, or set of values, as output, i.e., it is a procedure with a set of mathematical instructions or finite rules that must be followed to solve a problem (Yanofsky, 2011).

Artificial Neural Network (ANN), is a processing algorithm whose operation is inspired by the design and functioning of a human brain, with a remarkable capacity for self-organization and adaptive learning (Talaviya et al., 2020).

Convolutional Neural Networks (CNNs), are similar to feedforward networks (they process data in one direction, from the input node to the output, node); these networks take advantage of the principles of linear algebra, in particular matrix multiplication, to identify patterns within an image and/or computer vision (International Business Machine, 2020b

Recurrent Neural Networks (RNNs), are characterized by feedback connections that allow information from previous steps in a sequence to influence subsequent predictions. These learning algorithms are commonly used for sequential and time-series data, where patterns over time are important for predicting future outcomes (International Business Machines, 2020b).

Perceptrons, were the first neural networks to be developed and employ a supervised learning rule; they are inspired by the latest research in neuroscience at the time carried out by McCulloch and Pitts in 1943, who suggested that neurons could implement logical operations (Wallisch et al., 2009).

Multilayer Perceptrons (MLPs), are ANNs composed of multiple layers of interconnected artificial neurons organized into input, hidden, and output layers (Shahamiri et al., 2022).

Supervised Learning, is an ML technique in which a model is trained with labeled data to make predictions or classifications. It involves providing the model with input data and corresponding output labels, which allows it to learn the relationship between the two (Morales and Escalante, 2022).

Unsupervised Learning, is a ML technique in which a model is trained on unlabeled data without specific output labels to discover patterns, relationships, and structures within the data. Unlike supervised learning, there is no predefined correct answer or target variable, and it can help uncover hidden insights and provide a deeper understanding of complex datasets (Morales and Escalante, 2022).

Reinforcement Learning, is a ML technique in which an agent learns to make decisions by interacting with an environment and receiving feedback in the form of rewards or sanctions. It uses trial and error to discover the optimal strategy or policy to achieve a specific goal (Morales and Escalante, 2022).

Generative Adversarial Networks, provide a way to learn deep representations without extensively annotated training data. Their learning is performed by deriving backpropagation signals through a competitive process involving a generator and a discriminator network, that estimate the potential distribution of real data samples and generate new samples from that distribution (Creswell et al., 2018; Wang et al., 2017).

Computer Vision or Artificial Vision (CV), is a branch of AI that enables computers and systems to extract useful information from digital images, videos, and other visual inputs, and to act or make recommendations based on that data from an engineering perspective. Technologically, it involves creating a theoretical and algorithmic basis to achieve autonomous visual understanding by processing an image at the pixel level (Sarker, 2022).

Genetic Algorithms, are a method for solving both constrained and unconstrained optimization problems. These algorithms are based on natural selection, the process that drives biological evolution. At each step, it selects individuals from the current population to be parents and uses them to produce offspring for the next generation. Over successive generations, the population ‘evolves’ toward an optimal solution (Mathworks, 2023).

Semantic Segmentation, is a DL algorithm that associates a label or category with each pixel in an image. It is used to recognize a collection of pixels that form distinct categories (Mathworks, 2024).

The Internet of Things (IoT), also called the Internet of Everything or Industrial Internet, is a new technological paradigm conceived as a global network of machines and devices capable of interacting with each other (Lee and Lee, 2015), integrated with sensors, software, and network connectivity that allows them to collect and share data (International Business Machine, 2016).

Big Data, refers to vast and diverse collections of structured, unstructured, and semi-structured data that continue to grow exponentially over time. These datasets, are so complex that traditional data management systems are unable to store, process, and analyze them (Cloud Google, 2023).

Robotics, is a branch of technology, that deals with the design, construction, and programming of robots. These programmable machines, often integrated with AI, can perform a series of actions autonomously or semi-autonomously. They interact with the physical world through sensors and actuators (Omankwu et al., 2017).

Sensors, are electronic devices composed of sensitive elements, that can measure physical parameters such as light fluctuation using a photoresistor, temperature using a thermistor, or any other fluctuation in the environment. Thus, sensors, enable the measurement of specific physical parameters or the detection of changes in their immediate environment (García et al., 2017).

Actuators, enable actions within themselves or induce actions in other specific objects. Actuators enable robots to control movement and perform specific tasks, such as steering, spraying, or manipulating objects. (García et al., 2017).

Autoencoder (AE), is a type of neural network architecture designed to compress (encode) input data efficiently to its essential characteristics and then reconstruct (decode) the original input from this compressed representation (International Business Machine, 2023).

Bagging, also known as bootstrap aggregation, is an ensemble learning method commonly used to reduce variance within a noisy dataset and yield a more accurate estimate (International Business Machine, 2024).

Central Processing Unit (CPU), known as the ‘brain’ of the computer, a CPU is essential for all modern computer systems. It executes the commands and processes necessary for your computer and operating system (Integrated Electronics, 2024).

Graphics Processing Unit (GPU), a GPU is a processor composed of many smaller and specialized cores. When these cores work together, they offer tremendous performance when a processing task can be divided among many cores simultaneously (or in parallel). This is particularly important in AI tasks (Integrated Electronics, 2024).

This comprehensive list of definitions and terminologies serves as a foundation for understanding the various concepts and technologies discussed in the subsequent sections of this review.

2. Artificial intelligence in weed science: A few applications so far

Weeds are an integral part of the agroecosystem. Their management is complex due to several factors such as space, light, water, soil nutrients, propagation, and size. The rapid and continuous adaptation of weeds can explain the persistent nature of yield losses due to weeds, despite technological advances. Therefore, their management is considered crucial and essential in all cropping systems (Jordan and Jannink, 1997; Murad et al., 2023). Labor shortages and the high costs associated with manual weeding have driven the advancement of automated weed control systems. In many cases, real-time automated weeding is an effective approach to ensuring the sustainability of agricultural production systems (Sow et al., 2024). AI can play an important role in advancing weed management by supporting the development of new solutions to this global challenge. For example, AI-based tools can help farmers detect weed proliferation and generate recommendations for optimal herbicide application. Although weed recognition still faces several challenges due to irregularities in growth patterns and difficulties in early detection, hybrid models that combine DL and traditional image processing are expected to improve weed detection and classification. However, the high cost of AI technologies currently represents a barrier to commercialization (Sow et al., 2024).

Recent advances in sensor technologies, computational capacity, and machine-learning algorithms have substantially improved the ability to detect weeds accurately and efficiently. This progress has enabled the development of timely applications, providing farmers with information to support decision-making, as well as the creation of new autonomous equipment (Kipkorir et al., 2021). Various sensors and techniques including computer vision, spectral evaluated, remote sensing and thermal imaging have been analyzed for weed detection, in addition ML and DL play a crucial role in weed recognition allowing targeted spraying of weeds in infested areas eliminating the need to treat the entire field. Automated weed control systems that use sensors for weed identification and expert systems for herbicide dose determination have gained attention in agriculture because they can reduce environmental pollution and minimize poisoning risks for applicators. These systems also enable localized herbicide applications by targeting areas with high weed proliferation, helping prevent weed populations from exceeding economic thresholds (Sow et al., 2024).

The accessibility and performance of image-based weed recognition tools in crops have evolved due to advancements in three key areas: (1) increased computational power, efficiency, and hardware accessibility; (2) enhanced availability of image data for training complex algorithms; and (3) the development of novel algorithm architectures. The broader technology industry, including entities like Google AI and Meta AI, and the computing community have propelled many of these advancements through collaborative and open-source methods, thereby opening new avenues for weed control (Fernández-Quintanilla et al., 2018).

Despite these technological advancements, the accurate recognition of weeds in crops remains a significant challenge, especially in large-scale production systems. Factors such as the plasticity of weed and crop morphology and the broad environmental variability complicate reliable detection at high speeds (Coleman et al., 2022). The technology industry, with its continuous advancements and the incorporation of AI, is enabling a shift in weed management and control. Moving from merely detecting the presence of a weed in an image to identifying specific weed species based on their morphological characteristics in real time for targeted applications (Coleman et al., 2022)

2.1 AI technologies in weed recognition

Image classification neural networks such as DenseNet, EfficientNetV2, ResNet, RegNet, and VGGNet have successfully discriminated weeds such as dandelion, dallisgrass, purple nutsedge, and white clover in turfgrass, achieving an F-score higher than 98%. This has enabled the creation of autonomous systems for precise herbicide spraying (Jin et al., 2022b). Additionally, neural networks have been used in wheat crops for the detection of broadleaf weeds, achieving an F-Score of up to 99% with AlexNet, DenseNet, ResNet, and VGGNet (Zhuang et al., 2022). Another study mentions that the CenterNet neural network achieves a 95% accuracy in correctly identifying weeds in vegetables (Jin et al., 2021); DetectNet, trained with leaves of Geranium carolinianum, achieved the highest F-Score (0.94) (Sharpe et al., 2019).

Deep Convolutional Neural Networks (DCNN) have emerged as a crucial tool in weed detection, demonstrating an exceptional ability to learn weed characteristics (LeCun et al., 2015). This makes it possible to distinguish between crops and weed species using computer vision (Jin et al., 2022c; Zhou et al., 2020). These tools are typically used for accurate weed detection and localized applications (Chostner, 2017). Among the detectors used to locate plants are the Faster R-CNN methods (He et al., 2017), YOLO (You Only Look Once) (López-Correa et al., 2022), and SSD (Single-Shot MultiBox Detector) (Liu et al., 2016). These methods frame the plants, but they do not provide an exact determination of the weed in the image (Tan et al., 2021). Therefore, Mask R-CNN segmentation networks have been employed to refine this result. However, these networks require a pixelated image, making them more time-consuming (He et al., 2017).

2.2 Computer vision and object recognition

Digital image processing and computer vision are among the most widely researched techniques for weed detection. These methods typically involve the classification of morphology, texture, and pattern characteristics, often including segmentation and transformations to feed the ANN (Liu and Bruch, 2020). Additionally, these techniques involve image preprocessing (such as color space transformation or image resizing), feature extraction (where relevant image attributes are selected), and finally, the application of a classification algorithm that uses these features to identify the weed (Oerke et al., 2010; Wang et al., 2019).

Weed recognition is improving (Coleman et al., 2022). For instance, in the work conducted by Burks, Shearer and Payne (2000), an accuracy of up to 96.7% was achieved in classifying five similar weed species, including velvetleaf (Abutilon theophrasti Medik.), giant foxtail (Setaria faberi Herrm.), common lambsquarters (Chenopodium album L.), large crabgrass (Digitaria sanguinalis [L.] Scop.), and ivyleaf morningglory (Ipomoea hederacea Jacq.), as well as soil class. The classification was based on texture, color, and color-related features using discriminant analysis and a neural network. Drone-acquired images from a soybean crop, analyzed using AlexNet, allowed weeds to be classified as grasses or broadleaves with 77% accuracy, thereby supporting precise herbicide application (Ferreira et al., 2017). Other studies have performed classifications based on the fusion of surface color and leaf edge shapes, as well as methods that propose classifications by leaf differences (Dyrman et al., 2017). Refined techniques incorporate information from the specific reflectance spectra of each plant species (Gao et al., 2018; Im and Jensen, 2008; Peerbhay et al., 2015).

Advances in computer vision have created opportunities for automatically classifying herbicide-induced symptoms in cotton cultivation through digital images and DL. Evaluations of the damage to cotton caused by herbicides, including nicosulfuron, 2,4-D, dicamba, atrazine, isoxaflutole, glyphosate, glufosinate, and paraquat at three simulated drift rates, suggest a promising future for image-based models in diagnosing herbicide symptomatology. Models such as DenseNet121, InceptionV3, and ViT, trained with a dataset of more than 23,000 images, achieved classification accuracies between 74% and 96% (Torres et al., 2023).

2.3 Satellite and drone imagery

Satellite and drone imagery are key tools in precision agriculture, allowing farmers to remotely collect crop data for decision-making. However, these technologies differ in scale, resolution, cost, and operational flexibility. Satellite imagery is useful for large-scale monitoring, allowing large areas of land to be analyzed at relatively low cost and high frequency (Fakhar and Khalid, 2023). Satellites collect data in different spectral bands, allowing farmers to monitor parameters such as vegetation vigor (through indices like NDVI), soil water content, and plant water stress. However, their main limitation is their dependence on weather conditions; clouds can prevent the acquisition of adequate images at a specific time. They are ideal for regular, long-term monitoring and for early detection of problems on large farms. Drones are ideal for spot inspections and obtaining very high-resolution images (Ranjan et al., 2024).

Drones can fly below the clouds and capture detailed images of specific areas on precise dates, which is crucial at key moments in the crop cycle. With the help of multispectral cameras, they can identify problems at the individual plant level, such as the presence of pests or diseases. They are also used for tasks such as precise spraying, aerial seeding in hard-to-reach areas, and the generation of detailed 3D maps of the terrain. These two technologies are complementary tools, with an ideal approach being to use satellite imagery for initial monitoring of large areas. Once satellite imagery identifies an area with anomalies or low yields, a drone can be sent to that specific area for a more detailed, high-resolution inspection, enabling a synergistic approach to optimize resources, save time, and make more efficient decisions for farmers (Ranjan et al., 2024).

Emerging technologies have been incorporated into crop management, enabling the development of tools that utilize satellite or aerial images. These tools can be used to manage weeds by identifying and precisely mapping infestations with an accuracy approaching 99% (López-Granados et al., 2016). Despite these advancements, there are still reported issues related to the type of cameras used, flight altitude, temporal resolution (Peña et al., 2015), as well as crop size and weed types (Barrero et al., 2016). Unmanned Aircraft Systems (UAS) and geospatial information technologies present reliability challenges when used for imaging due to their moderate spatial, spectral, and radiometric resolutions of the acquired images. Given these challenges, research efforts have been mainly focused on the development and evaluation of vegetation indices for the empirical quantitative prediction of vegetation conditions (Singh et al., 2020).

UAVs are increasingly being used in agriculture due to their potential to perform tasks remotely. The incorporation of ML and DL algorithms into these UAV systems allows them to provide accurate and actionable information on crop health and weed presence. However, obstacles remain in terms of data management, algorithmic complexity, and operational limitations under different environmental conditions, drawing attention to innovation and research (Anam et al., 2024).

Using RGB images transmitted by unmanned aerial systems, the determination of the relationship between coverage and weed densities in cotton reached between 89.16% and 83.33%, with kappa values or Cohen’s Kappa index of 0.84; 0.79; and 0.75. Specifically, the highest correlations were observed with r2 values of 0.91 and 0.88 for Palmer amaranth (Amaranthus palmeri S. Watson) and Dinebra panicea (Leptochloa mucronata (Michx.) Kunth) respectively (Sapkota et al., 2020a). Additionally, Italian ryegrass was detected with an accuracy of 95.44% (Sapkota et al., 2020b)

Hyperspectral analysis, which uses spectral reflectance across hundreds of bands of the electromagnetic spectrum, is a technology used for weed detection. This technology can be deployed via satellite remote sensing, airplanes, or drones, and it provides information on the specific vegetation of a location (Im and Jensen, 2008). The detection depends on the growth stage of the weeds (Goel et al., 2003). By using hyperspectral chambers, researchers have been able to determine the effects of damage caused by glyphosate (Zhao et al., 2015) and dicamba on soybeans (Huang et al., 2016). This technology has also been used to differentiate between glyphosate-resistant and glyphosate-sensitive weeds, such as Amaranthus palmeri and Italian ryegrass (Reddy et al., 2014). In addition to using it to evaluate the early detection of herbicide stress and its potential to accelerate herbicide development processes, with an accuracy of 81.5% 1 day after treatment and in only 7 days with high precision (Niu et al., 2023). One of the cases is the application of hyperspectral technology to describe and predict the spectral response of common lambsquarters (Chenopodium album L., CHEAL) to the application of glyphosate, which surpasses the current visual methods in use and could be a tool in the detection of herbicide resistance (Soto et al., 2025). Despite the advantages of hyperspectral imaging technology, some disadvantages can also be found, such as the cost of hyperspectral systems compared to other image processing techniques, the use of high-speed computers for data processing and data storage capacity, noise in the image signal, as well as the difficulty in detecting and identifying the different elements within the Hyperspectral image, except for elements differentiated by absorption characteristics (Bhargava et al., 2024).

Remote sensing for weed detection and mapping currently relies on 2D imagery, the Normalized Difference Vegetation Index (NDVI), and hyperspectral imagery. However, these techniques are limited in their ability to capture information from lower layers of the vegetation canopy. This limitation results from variations in light conditions, leaf and stem overlap, and shadowing, all of which can reduce their effectiveness. To mitigate these challenges, researchers are developing 3D imaging techniques that include the additional dimension of depth (Amean et al., 2021). These methods, such as Structure from Motion (SfM), stereo vision, and Light Detection and Ranging (LIDAR), can reconstruct 3D models of plants and canopies from multiple successive images. This allows for the mapping of the spatial heterogeneity of weeds in response to management tactics (Andújar et al., 2018; Bagavathiannan et al., 2022; Comba et al., 2019).

The techniques mentioned in the previous paragraph offer promising solutions for monitoring and applying herbicides when weeds are uniform or exhibit variability that may interfere with weed control, prevent weed escapes, increase the soil weed seed bank, and lead to the evolution of herbicide resistance (Bagavathiannan et al., 2022; Neve et al., 2011). Three-dimensional imaging can enhance the accuracy of automated control systems by differentiating weeds by size class in realtime, modeling weed-crop competition, and predicting yield loss. However, it’s important to consider the limitations of this method. Factors such as leaf shape and orientation, canopy density, and environmental conditions can affect the efficiency and accuracy of 3D reconstructions. Despite these challenges, continued technological advances are likely to improve the accuracy of 3D modeling. This method provides a more informative description of vegetation architecture than RGB images alone and is expected to be integrated with other technologies to leverage its considerable potential benefits. However, it’s worth noting that there are relatively few studies that use 3D images to detect weeds (Bagavathiannan et al., 2022).

2.4 Sensors for Weed Detection

There exists a significant opportunity in the realm of traditional sensor hardware designs that remains unexplored. This opportunity lies in the development of ML-based smart sensors, which have the potential to drive a multitude of new applications and enable unique sensing capabilities across various areas (Ha et al., 2020). We envision a new generation of computational weed detection systems that reduce data-processing requirements while improving weed detection accuracy and efficiency. These systems would be compact and low-cost sensor implementations designed through iterative analysis of data-driven detection results facilitated by ML. Leveraging ‘big data’ analytics and the IoT, we aim to create robust detection networks that can significantly impact various fields, such as agriculture (Valeri et al., 2020).

The workflow of a ML-enabled smart sensor begins with the acquisition of detection data. This data is used to train a learning model that produces detection results as predictions. The model’s learning is evaluated through a cost function, J(y,y’), which measures the difference between the verified target (standard, y) and the algorithmic output (y’) of the sensor. Following this, the sensor undergoes optimization, which may involve redesigning its detection hardware. Uninformative or less valuable features are then eliminated and replaced with adjustments. The process concludes with the design of a final ML-enabled smart sensor (Valeri et al., 2020).

Compressive sensing leverages a robust mathematical framework often used in computational systems (Duarte et al., 2008; Goldstein and Osher, 2009). It encodes a given signal, x, using a sparse sampling operator or encoder, θ, defined a priori. This process solves the linear equation y=θx. The original signal x can be reconstructed from a smaller number of measurements, compared to what is dictated by the Nyquist sampling theorem. This is under the assumption that the signal can be represented as a sparse vector in a mathematical domain, such as the wavelet domain (Goldstein and Osher, 2009; Valeri et al., 2020).

ML-inspired sensor design aims to systematically prioritize or select potential measurement features based on their statistical value and contribution to accurately predicting the desired sensor output (Saeys et al., 2007). While these tools are powerful, one should be wary of a phenomenon known as the “curse of dimensionality”. This refers to how a data-driven design approach applied to the high-dimensional space of the training data can obscure meaningful correlations with the target detection information (Valeri et al., 2020). Genetic algorithms and wrapper methods, based on standard statistical tests, SVMs, and neural networks, have been used to select an optimal subset of spectral bands for efficient and application-specific detection (Anzanello et al., 2017; Xiaobo et al., 2010). These methods can be implemented in conjunction with fluorescence-based systems for various applications, including sensor-based, non-invasive, and destructible applications, among others (Göröcs et al., 2016).

ML algorithms, when used in conjunction with sensors, can accurately detect and discriminate weeds. Several techniques are currently being employed in sensors for weed management systems. These include NIR spectroscopy sensors (Nevavuori et al., 2019), optical NIR sensors (Hoogi et al., 2016), RGB-NIR sensors (Nevavuori et al., 2019), electrochemical biosensors (Sheng et al., 2019), and hyperspectral and multispectral imaging (Ferreira et al., 2017; Mohanty et al., 2016; Pantazi et al., 2016, 2017). Additionally, RGB images are also used in these systems (Ferentinos, 2018).

2.5 Machine learning and deep learning algorithms for weed recognition

2.5.1 Machine learning algorithms

Among the most important machine learning (ML) algorithms used in weed detection are support vector machines (SVMs), decision trees (DTs), random forests (RFs), and adaptive boosting (AdaBoost) (Bishop, 2006; Ha et al., 2020; Islam et al., 2021; Mitchell, 1997; Tan et al., 2016; Wu and Zhao, 2011; Zhang et al., 2021) SVMs estimate an optimal hyperplane to linearly separate data into two classes (Scherrer et al., 2019). The DT model structures data into a tree-like form, with nodes representing decision points based on statistical criteria, and leaves labeling the classes of resources. RF is an algorithm that uses sets of DTs. It employs bootstrap aggregation (bagging) to generate m (where m<N) standardized and randomized datasets D1, D2, ..., Dm with replenishment. For each dataset, m DTs h1, h2, ..., hm are generated. New instances are tested in each tree, and the class with the highest vote across all m trees is reported as the final value. Adaboost uses a reinforcement technique to build robust learning from weak learning with DTs. This algorithm identifies data points that are incorrectly classified in each learning iteration, increasing their weights for the new interaction while reducing the weights of correctly classified points.

The k-nearest neighbor (KNN) algorithm categorizes an input, denoted as x, based on its nearest k neighbors. It uses the Euclidean distance to calculate probabilities and assigns the input to the class that has the highest number of data points among the k neighbors. On the other hand, the K-means algorithm, which also uses Euclidean distance, divides the data into predefined k clusters. Each cluster groups data with similar characteristics, ensuring that each piece of data belongs to only one cluster. The k centroids of these clusters are set randomly, and each piece of data is assigned to its nearest cluster.

ANNs, which originated from the Perceptron, are essentially multilayer versions of these Perceptrons. To better understand ML versions of CNNs, which model complex nonlinear stochastic relationships and are widely used in intelligent control and weed detection (Kujawa and Niedbała, 2021), it’s important to delve into the structure of ANNs. Within ANNs, predictors are processed through each layer of units, known as neurons. Each neuron generates weights through a mathematical activation function, which are then used as synapses. This process is represented as hiσ=∑i=0NVijxithid, where σ is the activation function, N is the number of input neurons for a given neuron, Vij is the weights of those input neurons, Xi shows the input values for the input neurons, and T is the threshold for activation.

2.5.2 Deep learning algorithms

Deep learning (DL) techniques have increasingly replaced traditional machine learning (ML) approaches for real-time weed detection, leading to the development of new models and an increase in computational power. This shift in techniques has been well-documented (Murad et al., 2023), with the introduction of increasingly advanced GPUs, the processing of images, videos, and other complex data has been significantly accelerated. This advancement has overcome some limitations of weed detectors (Nixon and Aguado, 2019; Tan and Jiang, 2018). Despite this significant progress, it is believed that ML and DL still have much more to contribute than we can currently imagine (Castrignano et al., 2020; Goodfellow et al., 2016). In the beginning, techniques based on digital imaging and computer vision, in combination with ML, achieved resonant accuracy. However, since 2015, DL has achieved significant performance improvements with the advancement of CNNs and their current variants such as SegNet, GoogLeNet, ResNet, DetectNet, and VGGNet (Murad et al., 2023).

There are several gaps in knowledge regarding the use of DL for weed detection. One significant issue is the lack of a large enough dataset for training, validating, and evaluating models in different scenarios (Moazzam et al., 2019). Another challenge is the presence of light, which suggests the need for image processing and segmentation (Li et al., 2019). Further discussions revolve around topics such as plant growth. Semi-supervised learning has been proposed as a solution that can improve yield, as well as address issues related to the positioning in the rows of the crop (Shanmugam et al., 2020) and the density of both weed and crop (Hasan et al., 2021). In addition to these problems, there are other challenges. For instance, once a DL model has been trained, it cannot approach new species. To overcome this limitation, incremental learning has been proposed to extend the existing model without the need for retraining (Hu et al., 2021).

Deep Neural Networks (DNNs) are extensions of ANNs. A CNN is a type of DNN that learns weights and biases from various input images to distinguish and classify objects, such as weeds (Li et al., 2016; Murad et al., 2023). One of the advantages of CNNs is that they do not require manual selection of features. Instead, the network automatically learns from the training data. CNNs autonomously capture the spatial and temporal dependencies of the input image through filters, providing efficient image processing with fewer parameters and reduced processing time. CNNs use a rectified linear activation unit (ReLU) as the activation function, denoted as g(z). This function generates an aggregate value z greater than 0 for its activation; otherwise, g(z) equals the maximum of {0,z}. The linearity of ReLU for z>0 facilitates the descent of the stochastic gradient due to its conservation of input properties (Nwankpa et al., 2018). In addition, CNNs normalize the training images in standardized mini-batches for each layer. This process ensures stable learning and reduces the training times of CNNs (Ioffe and Szegedy, 2015).

The development of DNN architectures began in 1979 with a visual pattern recognition neural network model called Neocognitron (Fukushima, 1980; Fukushima and Miyake, 1982). This model served as the basis for the first multilayer CNN, known as ConvNet, which uses a backpropagation algorithm. This was a significant advancement compared to the unsupervised reinforcement learning scheme of its predecessor, Neocognitron (LeCun et al., 1989). In 1998, Yann LeCun proposed an improved version of ConvNet, called LeNet-5 (Lecun et al., 1998). This was the first concretely usable CNN architecture, featuring a 2-layer structure with 6 and 16 feature maps in the first and second layers, respectively, followed by two fully connected layers. With the arrival of ImageNet (Russakovsky et al., 2015), researchers began proposing methods to reduce the error of incorrectly classified images. These were ranked in descending order of predictive confidence, denoted as P(Yi|Xi), where Xi represents the input test data, and Yi is the label of the classes under consideration. This led to the development of many other architectures, such as AlexNet (Khan et al., 2020), which used ReLU activation on all layers except the last one, which employed softmax activation. GoogleNet, linked to the Inception-V and Inception-ResNet Modules, followed, as did the neural network VGGNet (Simonyan and Zisserman, 2015) and Highway Net.

A major breakthrough in CNN development was the introduction of ResNet, or residual network, by He et al. (2016). ResNet is a deep CNN architecture with up to 152 layers that helps address the vanishing gradient problem, in which gradients become too small during backpropagation to effectively update model weights.

By using residual connections, ResNet allows deeper networks to learn more efficiently and helps prevent deeper architectures from producing higher training errors than shallower ones. These advances have contributed to the widespread use of deep convolutional neural networks (DCNNs) for weed detection and discrimination (Jin et al., 2022a).

Since 2015, there has been a rapid increase in architectural innovations and applications of CNNs, with significant improvements in architecture and application potential. Certain problems, such as the vanishing gradient, have been addressed by connecting auxiliary intermediate layers to learning (Szegedy et al., 2015), implementing multi-layer information control mechanisms with skip connections, and establishing channel connectivity between layers (Huang et al., 2017; Wu et al., 2018). Other advancements have utilized instance-based object segmentation and semantics, location, and scene analysis, etc. Examples of these include the Single Shot Multibox Detector (SSD), Region-based CNN (R-CNN), Faster R-CNN, Mask R-CNN, and Fully Convolutional Neural Network (FCN), which were built on the foundations of ResNet, VGG, Inception, etc. (Cai and Vasconcelos, 2021; Pang et al., 2019).

These advancements were also extended to image labeling (Girshick, 2015; Lin et al., 2017; Long et al., 2015; Vinyals et al., 2017). They involved stacking multiple transformations both in-depth and in parallel for complex problems (Han et al., 2017; Zagoruyko and Komodakis, 2016). Hybrid architectures (Lv et al., 2019; Szegedy et al., 2016; Yamada et al., 2016) and network block architectures (Wang et al., 2017; Woo et al., 2018) have also been explored. Furthermore, approaches such as enhanced input channels, auxiliary classifiers, and transfer learning have been introduced to improve model performance (Khan et al., 2018).

Different improvements have been made in the architecture of CNNs, such as parameter optimization, regularization, and structural redesign. However, the current trend is toward designing lightweight architectures that do not compromise performance, making CNNs applicable for resource-constrained hardware (Khan et al., 2020). In the area of weed detection, there has been an attempt to innovate SegNet (Badrinarayanan et al., 2017), an efficient architecture for semantic pixel segmentation. This architecture is composed of a network of encoders, similar to AEs (Goodfellow et al., 2016). The encoding process involves convolutions using 13 convolutional layers of VGGNet, followed by a 2 × 2 maximum pooling to generate the encoded representation. This increases the resolution of the encoded data with the maximum indexes and uses the softmax function at the end, resulting in a FCN. Another algorithm that utilizes the FCN concept is U-Net (Ronneberger et al., 2015), unlike other algorithms, U-Net does not employ clustering indexes during the decoding phase. Instead, it transfers the resource maps from the encoder to the decoder, achieving better segmentation but at the cost of time and memory. Deeplab’s segmentation algorithms (Chen et al., 2018) in FCNs reduce the spatial resolution of images by using atrous convolutions to generate maps of decoded features with lower computational overhead.

DL has seen significant advancements in recent years, particularly with the introduction of generative adversarial networks, as well as variants of CNNs, AEs, and RNNs. Typically, data for these networks is generated directly at the research site using customized equipment such as drones, robots, semi-professional cameras, or mobile phones (Murad et al., 2023). Researchers use ML and DL algorithms to detect weeds. For DL, CNN are commonly used, although AEs can also be utilized. However, AEs are generally not used separately. The performance measures, such as precision, accuracy, F-Score, recall, and Area Under the Curve (AUC), do not seem to have a standard and are used interchangeably (Murad et al., 2023). Researchers have highlighted the importance of large datasets for better weed detection, which is a challenge (Espejo-Garcia et al., 2021; Murad et al., 2023; Vasileiou et al., 2024). In addition to selecting algorithms and techniques to detect specific types of weeds and/or associated crops, it is also important to standardize the measures of accuracy, precision, recall, F-Score, and AUC-Score (Murad et al., 2023).

Supervised ML requires annotated datasets for model training, enabling algorithms to classify or make predictions on previously unseen, unannotated data. However, generating high-quality annotations is often labor-intensive and time-consuming. A solution to this problem is Few-Shot Learning (FSL), which has recently been gaining attention. FSL reduces the amount of annotated data needed to train an effective model (Wang et al. , 2020). FSL can assist in the development of vision systems for robotic solutions for spraying or selective weed removal. This reduces the cost of acquisition and annotation by weed specialists in sufficiently large datasets (Li and Yang, 2021; Snell et al., 2017). It also optimizes less robust compute architectures GPUs and complex hyperparameter lookups (Bosilj et al., 2020; Fawakherji et al., 2020). FSL methods are categorized into (i) data augmentation, (ii) transfer learning, and (iii) meta-learning. Data augmentation based on transfer learning involves pre-training a model on a larger source dataset before proceeding to training on a small-sized dataset. Meta-learning approaches, or ‘learning how to learn’, aim to develop models that specialize in new tasks using little data (Ghatrehsamani et al., 2023).

Data augmentation using geometric transformations is used as pre-processing strategies to generate multiple variations of training images in terms of rotation, inversion, cropping, resizing, and translation (Güldenring et al., 2021). Generative Adversarial Networks have been used to generate images of weeds and crops, appearing to be a solution to the size of very small datasets (Fawakherji et al., 2021; 2020; Shorten and Khoshgoftaar, 2019). Popular model weights in DL computer vision, pre-trained on large datasets with generic characteristics, are generally used in transfer learning. This is done by fitting the last layer to the target data or using the pre-trained network as a feature extractor and using the features to train other ML or DL models with gradient-boosting (Espejo-Garcia et al., 2020b, 2020a). Thus, transfer learning has been able to make a segmentation model up to 80% more efficient in terms of training when compared to a CNN trained from scratch (Bosil; et al., 2020).

Meta-learning has been proposed to solve the N-way K-shot mechanism (Lu et al., 2023; Padmanabhan et al., 2023), so that the algorithm can learn new classes from K small set of example images (Wang et al., 2020), in some optimization steps (Finn et al., 2017; S. Wang et al., 2022), and by learning metric and data identification from a different perspective being a contrastive learning (Güldenring and Nalpantidis, 2021; Kaya and Bilge, 2019; Li and Yang, 2021; Snell et al., 2017).

2.6 Availability of weed datasets

The limited availability of weed image datasets remains a major constraint for applying deep learning (DL) techniques in weed management. To address this challenge, a multi-state effort is currently underway to develop an open-access repository for sharing crop and weed images, supporting model training, validation, and future research. The goal is to train AI algorithms, promoting the research and development of ‘smart’ precision technologies. These technologies can increase efficiency, reduce costs, and improve the overall sustainability of agricultural practices. This progress would be particularly significant in the field of weed recognition applications. Currently, companies often store data behind closed doors, which slows progress for everyone involved (Kutugata, 2023).

In a review conducted by Murad et al. (2023), it was noted that machine-learning datasets used for weed detection vary widely in size, with the largest datasets containing approximately 11,150 to 36,000 images and smaller datasets containing only 60 to 120 images. The review also highlighted that weed detection work is carried out exclusively on 26 crops, including sugar beet, carrots, and various cereals (wheat, corn, rice, and soybean). These studies cover 34 types of weeds among the different crops. The most frequently mentioned weeds include broadleaf dock (Rumex obtusifolius L.), common lambsquarter (Chenopodium album L.), canadian thistle (Cirsium arvense (L.) Scop.), common chickweed (Stellaria media (L.) Vill.), mayweed chamomile (Anthemis cotula L.), blackgrass (Alopecurus myosuroides Huds.), junglerice (Echinochloa colona (L.) Link), purple nutsedge (Cyperus rotundus L.), shepherds purse (Capsella bursa-pastoris (L.) Medik.), fescue grass (Schedonorus sp.), tumble pigweed (Amaranthus albus L.) and grass weed (Monocot) (Murad et al., 2023).

Weed-AI, a crop weed image repository from the Precision Weed Control Group and Sydney Informatics, allows researchers to collect and upload weed and crop image datasets. The latest version comprises 28,897 annotated images, 18,659 of which are weed images (Sydney Informatics Hub, 2021). The Weed25 dataset, which contains 14,035 images of 25 different weed species at various growth stages, was used to train the DL models YOLOv3, YOLOv5, and Faster R-CNN. These models achieved average detection accuracies of 91.8%; 92.4%; and 92.15% respectively under the same training parameters (P. Wang et al., 2022). The DeepWeeds dataset consists of 17,509 labeled images of eight nationally important weed species native to eight locations in northern Australia. DL models, Inception-v3 and ResNet-50, were trained on this dataset, achieving average classification accuracies of 95.1% and 95.7% respectively. The ResNet-50 model demonstrated real-time performance of 53.4 ms per image (Olsen et al., 2019).

Another notable dataset includes 24,284 images of weeds obtained from 329 species of vascular plants growing on agricultural land in seven regions of the European part of Russia (Tretyakova et al., 2020). Among the sizes of image banks for DL, Coleman et al. (2022), presented a publicly available and open-access databank on different crop production systems, which fits among the sizes of image banks for DL mentioned by Murad et al. (2023). In an experiment conducted for weed detection in wheat with small training datasets of 5,500 negative and 5,500 positive images, or large ones with 11,000 negative and 11,000 positive images, and three training image sizes of 200×200, 300×300, and 400×400 pixels, the authors concluded that CenterNet, Faster R-CNN, TridentNet, VFNet, and YOLOv3 networks were insufficient. However, AlexNet, DenseNet, ResNet, and VGGNet trained with a large image dataset of 200 × 200 pixels were very effective for the detection of broadleaf weed seedlings in wheat (Zhuang et al., 2022).

A common question posed by specialists in weed management is why such large image datasets are needed to train AI models. The answer lies in the complexity of the task to be performed by the model, the settings of the model parameters, and the complexity of the learning algorithm (Goodfellow et al., 2016). Generally, models should exhibit a relationship between decreasing training error and validation error. Large datasets or augmented datasets represent a more complete set of possible data points, thus minimizing the distance between the training and validation set (overfitting) by addressing a solution at the root of the problem: the training dataset. It’s important to note that this is not the only technique developed to reduce overfitting. Other strategies focus on the model architecture itself, with functional solutions such as dropout regularization, batch normalization, transfer learning, and pre-training aiming to extend DL for application to smaller datasets (Shahinfar et al., 2020; Shorten and Khoshgoftaar, 2019).

In order for the algorithm to accurately, robustly, and effectively distinguish between crops and weeds, a diverse and high-quality dataset is needed that includes the varied agricultural environments of the real world, where weeds are found in different stages of growth, lighting conditions, climate and soil types. This data must be processed and standardized to prepare it for training for which Image annotation is performed, which is a critical and laborious process where each image must be labeled to tell the AI what it is seeing, then object detection / semantic segmentation by drawing bounding boxes around each individual plant (weed or crop) and labeling it, data augmentation to increase the size and diversity of the dataset without collecting new images, involving applying various transformations to existing images such as rotation, adjusting brightness and contrast, and cropping, and then moving to normalization where the pixel values of the image are scaled to a standard range (for example, from 0 to 1). This helps the algorithm learn more efficiently and consistently (Chen et al., 2018; Li et al., 2019; Shahinfar et al., 2020; Shorten and Khoshgoftaar, 2019; Lei Tian et al., 1997).

2.7 Weed monitoring uses artificial intelligence

In this section, we briefly address how AI is improving weed monitoring. One of the benefits of AI in weed management, as highlighted by Muchhadiya et al. (2024), is its ability to monitor weed populations and track weed infestation hotspots over time. Vasileiou et al. (2024) described how AI, in conjunction with remote sensing and geospatial technologies, can support the detection, mapping, and monitoring of weed populations. Vasileiou et al. (2024), describe how AI, in conjunction with auxiliary technologies such as IoT devices, cameras, UAVs, and field robots, facilitates efficient weed monitoring. While several methods exist, terrestrial platforms are considered more appropriate for real-time detection during the seedling stage, especially for weeds like Solanum rostratum Dunal. The platform camera’s proximity to the ground allows for a higher pixel area of a single plant in the image, making plant traits more obvious and easier to identify (Q. Wang et al., 2022).

During and after crop production, AI-related strategies can be employed to periodically monitor weeds and prevent further damage (Bhoi et al., 2021). Real-time monitoring and tracking systems accurately determine the severity of current weed infestations. Additionally, prediction models can forecast weed infestations, creating warning systems based on environmental factors. Integrated ecophysiological decision support systems help implement management and control measures specific to weeds (Abd Manaf et al., 2024). Customizing monitoring methods based on specific growth stages and vulnerability of both weeds and crops is essential. Furthermore, AI systems must possess a comprehensive understanding of the particular crop under consideration to ensure crop health and productivity. Accurate weed identification relies on this knowledge (Saqib et al., 2023).

Precision and timeliness are fundamental to the efficiency of AI-based weed management systems (Mishra and Gautam, 2021). AI-operated systems can monitor weeds in the fields, operating in real time or asynchronously, constantly detecting weed presence. Immediate intervention minimizes the period during which weeds can proliferate and compete with agricultural crops (Thanh Le et al., 2021). Combining AI with imaging technology enables efficient, real-time weed monitoring in the field. Detecting and responding to weed problems efficiently, personalized with multi-sensor technologies, allows for strategically based weed control. Precise spatial and temporal information about weeds minimizes herbicide use, reduces costs, and protects the environment (Wang et al., 2019).

An automatic method for generating crop and weed density maps, using an algorithm that significantly correlates manual and automatic evaluation, describes an approach for evaluating crop and weed areas in cereal crop images (Hague et al., 2006). Q. Wang et al. (2022), designed a method using the proposed neural network YOLO_CBAM, achieving a precision of 0.9465 and a recovery of 0.9017 for real-time recognition. This detection speed and effectiveness can be applied to real-time field detection and monitoring of seedlings of the invasive weed Solanum rostratum Dunal.

2.8 Non-chemical actions that use artificial intelligence and computer vision in the control and monitoring of weeds

In this section, we generally address the contribution of AI to non-chemical weed management. Weeding is one of the most tedious and labor-intensive tasks in agriculture, and chemical weed control has historically been, and continues to be, the most widely used option for managing weeds. However, because of the growing demand for more sustainable agricultural practices, non-chemical approaches, such as mechanical weeding, robotic weeding, and other automated weed control technologies, are being increasingly promoted (Fountas et al., 2022).

AI-based predictive modeling uses historical data on climate, soil conditions, and weed occurrence to help farmers identify areas at risk of future weed outbreaks. This information allows farmers to plan preventive measures and optimize control strategies, while AI-powered robots equipped with computer vision can navigate fields, identify weeds, and remove them using mechanical tools. Thus, mechanical weeding robots are being prototyped and developed for various row (Åstrand and Baerveldt, 2002; Nørremark et al., 2008; Quan et al., 2022), such as rice (Oryza sativa L.) (Raja et al., 2020a), corn (Zea mays L.) (Cordill and Grift, 2011), peach (Prunus persica (L.) Batsch.) (Luo et al., 2020), tomato (Solanum lycopersicum L.) and lettuce (Lactuca sativa L.) (Raja et al., 2020b). Some robotic solutions are commercially available, for example Dino from Naio technologies and the Farming GT- Weeding robot from Farming Revolution (Fountas et al., 2022).

A computer vision guide has been used to address the problem of mechanically eliminating weeds within rows of transplanted vegetables and salads (Tillett et al., 2008), however every day these machines seek to improve their efficiency in the field (Cordill and Grift, 2011). Physical management is also being strongly driven by AI and computer vision with a view more specifically based on directed energy (Cao et al., 2024) and thermal lasers (Carbon Robotics, 2024; CSIC-UPM, 2023; Delbert, 2021).

2.9 Application of precision herbicides in weed control using AI

There is growing interest in optimizing herbicide applications through AI-based weed detection. This section briefly discusses how AI technologies are improving herbicide application efficiency. Among these approaches, variable-rate spraying is considered one of the most advanced and efficient methods for site-specific weed management (Patil et al., 2023; Ukaegbu et al., 2022). Precision application technologies have been linked to computer vision for several decades, particularly in the context of robotic and automated weed control (Slaughter et al., 2008). Robotic technology may help reduce agriculture’s dependency on herbicides while improving sustainability and reducing environmental impacts. In this context, Slaughter et al. (2008) reviewed four core technologies for robotic weed control: guidance, detection and identification, precision in-row weed control, and mapping. Currently, there are few related works in this area. However, with the increase in computational capacity and reduction of hardware cost, research in this area is expected to increase. Typically, CV and image processing are applied together to obtain information about weeds and apply the spray rate, mechanical or electrical weeding (Astrand and Baerveldt, 2002; Guan et al., 2015; Tillett et al., 2002).

Giles and Slaughter (1997), developed a precision system that involved machine vision and a nozzle that adjusted the spray pattern’s width and the nozzle’s orientation angle concerning the direction. Tian (2002), Tian, Steward and Tang (2000), designed a similar system that additionally uses an environment-adaptive image segmentation (EASA) algorithm (Lei. Tian et al., 1997). This system evaluates the infesting grain using a decision algorithm, thus obtaining information for spraying set for reducing herbicide use (L F Tian et al., 1999). Another approach using CV for micro-spray was shown by Lee, Slaughter and Giles (1999), which was later modified by Lamm; Slaughter and Giles (2002), using an algorithm called “ERODE” achieved 88% weed-recognition accuracy.

A project to develop a smart sprayer uses a DL system with the YOLO model (Partel et al., 2019), achieving herbicide savings of up to 42% in simulated experiments. Similarly Bawden et al. (2017), presented a system with an efficiency between 92 and 96%, demonstrating that this is a robust system to selectively weed or mechanically eliminate (Meshram et al., 2022). Hu et al. (2022), presented the software and hardware design of an automatic system for microvolume herbicide spraying using a mobile robot for early-stage weed control. Evaluations on soil surfaces with artificial objectives showed lateral errors of herbicide spraying at subcentimeter levels, and the proposed assignment algorithm demonstrates that it can provide good coverage within the regions within the row. The image recognition algorithms are the core of Variable Rate Spraying (Bawden et al. , 2017), but these types of systems need advanced hardware to improve their performance and accuracy in weed control (Partel et al., 2019).

Weed mapping for site-specific herbicide application has been reported to reduce herbicide use by 23% to 89% (Berge et al. , 2012; Christensen et al., 2003; Gerhards and Oebel, 2006). By directing spray applications only to areas where weeds are present, site-specific systems can avoid treating 59% to 80% of the field area when weeds are absent. Systems based on the concepts of spot application, robotic weed removal, weed recognition with localized spray, sensor-based electrical weed control are currently being developed (Dyrmann et al., 2016; Gutjahr et al., 2012; Kunz et al., 2015; Peteinatos et al., 2020; Reiser et al., 2019; Tillett et al., 2002). Other developments include UAV mapping and weed spot spraying (Mink et al., 2018; Rasmussen et al., 2013; Xia et al., 2022), weed control with targeted direct herbicide injection (Pohl et al., 2019; Ruigrok et al., 2020). An autonomous multi-camera robotic system proposed by Wu et al. (2020), provides flexibility, with mechanical and chemical weeding tools fully operational and capable of performing selective mechanical and chemical in-line weeding in different terrain conditions and crop growth stages.

It is evident that the integration of AI systems (RTK-GPS) and machine vision guidance systems are adapted to row control technologies, however, these technologies still face signal interference and satellite dependence for equipment positioning, and sensors must improve their accuracy for weed identification, in which mechanical and physical methods should be integrated to strengthen robotics for weed control. However, due to the development costs of robotic systems, the integration of technology (sensors, ML, DL, multispectral (MS) and hyperspectral (HS) imaging technologies, LiDAR, Image Processing techniques, etc.) is limited to give more versatility to these systems for autonomous navigation and avoid dependence on the precision of a single sensor. Accurate plant identification especially in weed management faces several challenges related to sensor accuracy and environmental conditions. For example, RGB sensors, combined with machine learning (ML) and deep learning (DL), are widely used for weed identification and classification. However, their performance can be affected by fluctuations in ambient light conditions when operating in outdoor environments. In addition, RGB sensors have a limited spectral range and lack red-edge (RE) and near-infrared (NIR) bands. These limitations can be partly overcome by multispectral (MS) sensors, which offer greater sensitivity for vegetation analysis. However, the higher cost of MS sensors can limit their integration into robotic weed control systems. Even so, MS sensors also have limitations, particularly in providing detailed spectral information and distinguishing subtle differences among weed species and crops (Upadhyay et al., 2024).

HS sensors, compared to previous ones, are considerably more expensive, larger, and heavier. They offer a solution to sensitivity problems. However, real-time data processing is more complex due to the large number of bands, and the problem of susceptibility to variations in spectral signatures due to light exposure and atmospheric conditions persists. Thus, general challenges in weed and crop detection include similarity between plants in shape, color, and image textures; changing environmental conditions that affect the performance of weed detection systems; factors such as weed density, occlusion, overlap, shading from plant foliage, sun angle, different stages of plant growth, varying lighting conditions in the field; and interference from radar, laser, and other electronic systems with the sensors used on robots (Upadhyay et al., 2024).

3. Successful implementations of AI in global weed management and integration with traditional farming practices

Integrated weed management and the adoption of new technologies remain major concerns for farmers. However, translating these technologies into practical, farmer-accessible solutions is challenging. Although the extensive research literature may suggest that numerous tools are available for implementation, commercially accessible options remain relatively limited. Nevertheless, several successful cases involving different technologies have been reported, as presented by Silverberg (2023).

John Deere has developed a new high-tech sprayer called See & Spray Ultimate, which is available for corn, soybeans, and cotton. It can reduce herbicide use by up to two-thirds and is controlled by an AI software system equipped with 36 cameras and weed identification software. Its algorithm, trained with more than 300,000 images in the John Deere database.

The Smart Spraying solution from Bosch BASF Smart Farming together with AGCO Corporation commercializes the Smart Spraying technology in the Fendt Rogator sprayers with a weed weeder equipped with a camera and sophisticated sensors, detects weeds in milliseconds, allows the generation of digital maps with the areas with the highest weed density to increase operational efficiency, savings of herbicides and targeted spraying during the day and night, this technology is currently available for corn, soybeans, cotton, canola, sunflower and sugar beet crops (BASF, 2023).

The Israeli company Greeneye Technology has developed an AI-powered precision weeding system that can be integrated with existing sprayers. The system is reported to operate at speeds of up to 25 km h−1 and to enable targeted herbicide application in both pre- and postemergence weed control. According to Greeneye Technology (2023), this technology can reduce herbicide use by up to 90% and can be adapted to different row crops.

Precision AI is embedding AI technology into drones to achieve this breakthrough. The company has used its database of more than two million images and uses images of 15,000 plant species to train computer algorithms to distinguish staple crops such as corn, wheat, and soybeans (Liu, 2023).

LaserWeeder, developed by Carbon Robotics, is an automated weeding technology that eliminates the need for herbicide spraying by using high-powered thermal lasers to kill weeds. The system combines cameras and artificial intelligence (AI) to identify weeds and then targets them with lasers. According to the company, LaserWeeder can eliminate up to 100,000 weeds per hour when pulled by a tractor. This technology has been used in specialty crops, including potato, onion, and garlic fields, particularly in U.S. states such as California, Washington, and Idaho. The robot LaserWeeder™ is autonomous, and its tractor attachment version can cover up to eight hectares per day with an efficiency of up to 99%. It is based on technology with eight high-power carbon dioxide lasers, computer vision, and AI systems to eliminate weeds. It is equipped with 12 high-resolution cameras and moves in the terrain at about 8 km h−1 (Carbon Robotics, 2024; Delbert, 2021).

Garford Farm Machinery has three models in production: one for cereals, one for sugar beets, and one for vegetables. Among these, according to them, Robocrop offers the main benefit of a camera-guided hoe system that observes the crop rows and sends the images, allowing the operated tractor to move much faster than would be possible with a manual hoe. It is presented as an option for mechanical weed control. An innovation of this product is Robocrop InRow Weeder, which has a camera-controlled rotor for the reaction of weeds between plants and rows. It has a greater application in vegetables with weeding of up to 98.5% of the soil surface. Another product already linked to the spot sprayer is the Spot Robocrop, which with the help of specially developed nozzles reduces the cost of chemical control. It is available for potatoes grown on carrots, turnips, onions, or leeks. In addition, the Robocrop Side Shift System has been implemented in sprayers with a strip and hood (Garford, 2023; Tillett et al., 2008).

Ferrari Grow Tech’s Remoweed high-precision automated weeders removers use an AI vision system consisting of three models: the Remoweed, which removes weeds between rows and plants; and the Remolite, which detects images from a single row of plants or multiple rows. The wide reading angle of the camera is ideal for multiple rows of information processed by the vision system and is used to control the side-scrolling hydraulic frame to guide the machine hoes as close to the crop as possible. The Scorpion consists of an articulated structure capable of lateral transfer, is equipped with an infrared optical bar capable of identifying, during progress, the presence or absence of cultivated plants, and is capable of eliminating weeds up to one centimeter away from the cultivated plant (Ferrari Ggrow Tech, 2024).

Poulsen Engineering has the Robovator, which is equipped with a unique plant detection camera. It performs automatic weeding in the row, is adjustable, and can be controlled by your smartphone, laptop, or by the display itself. Its function is aimed at maintaining weed control since its efficiency has been reduced in high infestation and is not recommended in the presence of high weed densities (Poulsen Engineering, 2024).

Dino, developed by Naϊo Technology, is an electric agricultural robot designed to perform autonomous mechanical weeding. It uses a camera system to detect crops in the row and sends a signal to position the tools in the tool holder in order to weed the area between rows without damaging the plants. It is guided by global positioning system (GPS), Real Time Kinematic (RTK), and camera-guided tools. So far, it has been tested in salad, celery, parsley, cabbage, fennel, and leek crops. It has an efficiency of up to 7 acres of vegetables in 7 hours. Currently, this robot is a tool under a mechanical weeding service contract offered by Naϊo Technology (2021).

Oz, another technology of Naϊo Technology, consumes little energy and works completely autonomously thanks to its RTK and GPS guidance system. It is guided by a vision system by cameras and lasers. In 2020, this system served the service of 1,080 hectares of weeded fields. Its battery lasts approximately 8 hours depending on use and has an efficiency of 1,000 m2 hour−1. Its size is 130 cm long, 47 cm wide, and 88 cm high, which allows it to adapt to crops to take care of manual weeding (Naϊo Technology, 2023a).

Orio, developed by Naϊo Technology, was initially designed for weeding garlic. It is power-driven and equipped with RTK, GPS, a safety bubble created by 3 lidars, and an auto-guidance system with about 100,000 hours of experience. It follows the working lines and performs half turns with centimeter precision. Thanks to the Treffler tine harrow, the mechanical garlic weeder meets expectations. It is able to cover several hectares per day, is light, and has the ability to move sideways. It can transport cultivators, hoes, harrows, and even precision planters according to their characteristics (Naïo Technology, 2023b).

Ted, another autonomous robot from Naïo Technology, is equipped with GPS and RTK technology. It is presented as a perfect mechanical control alternative to reduce herbicide treatments around vineyards. It has an autonomy of 8 hours per day, and its size (400 cm long, 190 cm wide, and 240 to 275 cm high) allows it to work at speeds of up to 4.5 km/h. It has a battery autonomy of up to 8 hours depending on the conditions of use and an efficiency of 5 ha/day (Naïo Technology, 2024).

The Weedcer prototype, developed by Bosch Deepfield Robotics, uses Weedcer CV software based on decision rules to estimate the proportions of young weed leaves and cereal leaves in high-resolution RGB images for the application of herbicides on cereals at specific sites. This robot achieved significant herbicide savings between 18 and 27% (Berge et al., 2012).

BoniRob is a robot that has been integrated into multipurpose agricultural tasks by combining innovative agricultural engineering with web-based communication technologies. It proposes to use a mechanical robotic weed control system that integrates a human user as a remote worker in the process. It has an application of morphological and spectral phenotyping based on cameras to detect green areas. The spray nozzles are controlled so that herbicides are only applied in areas that actually contain weeds, thus reducing the amount of herbicides applied per hectare. This robot has managed to apply the weed treatment within the rows of carrots in BBCH scales of 10 to 20 (growth stages) in which weed control interventions were performed manually (Bangert et al., 2013).

The Ladybird, developed at the Australian Centre for Field Robotics (ACFR) at the University of Sydney, is powered by rechargeable batteries. These batteries are charged through curved solar panels located on each side of the robot. The robot is adaptable to different crop heights and can move at about five kilometers per hour. Autonomous navigation is achieved using a combined GPS/INS (inertial navigation system). As the robot moves, a LIDAR and a high-resolution spherical digital video camera capture environmental features. This allows the robot to identify and avoid obstacles and detect crop rows. To classify and analyze the crop, imaging sensors capture RGB images of the crops. A HS imaging camera (400-900 nm) captures IR and UV data, which provides data for weed control through fingerprint information. This data is used by a spray end-effector coupled to a Universal Robots UR5 six-axis robotic arm. When the ML algorithm identifies a weed, its coordinates are transferred to the robotic system. The system can then position itself directly over the weed to apply a controlled amount of herbicide (Underwood et al., 2017, 2015).

RIPPA (Robot for Intelligent Perception and Precision Application) developed by the ACFR Group in 2016, allows for precision applications and intelligent sensing. It runs on solar-charged batteries and is equipped with an intelligent precision variable injection applicator. Like the Ladybird technology, it delivers controlled and localized amounts of herbicide. It can operate autonomously in crop rows with 4 cm accuracy using satellite-based location corrections (Bogue, 2016; Engineers Australia, 2016).

WeLASER is an autonomous robot that performs weed control with a mobile high-power thulium-doped fiber laser. It has an advanced AI system for laser positioning in the meristem. Its perception system is equipped with vision hardware for weed and crop discrimination. It is controlled by the Agri-DSS IoT system in the cloud. This project is currently concluding its first phase of development, which has resulted in the development of different innovations with multiple applications (CSIC-UPM, 2023).

Robotti (Agrointelli) is an autonomous implement carrier that is very suitable for precise and accurate plant establishment and care tasks in row crops and beds for horticultural production. Using the Schmotzer implement (a hoe with multipurpose parallelograms), it effectively removes weeds between rows. It has an adjustable angle of inclination to combat aggressive weeds in sugar beet crops, beans, barley, pumpkins, and corn. With the Kult row weeder system, it is possible to mechanically control weeds close to the crop thanks to the precise recognition of each plant. It is designed for celery, broccoli, and cabbage crops. In addition, there is also the SpitFire band sprayer customized by Robotti, which is currently working with Agrointelli vision equipment for spot spraying on corn (Agrointelli, 2024).

EcoRobotix Generation 1 is a self-contained smart sprayer designed for ultra-localized treatments of row crops, pastures, and lawns. With over 95% efficiency in identifying and treating weeds, it is powered by solar panels and equipped with on-board cameras, RTK GPS, and sensors. These features allow it to identify crops and plot its course of travel. It is estimated to cover up to three hectares of land per day, using 30% less herbicide than traditional methods (Agtecher, 2020).

ARA developed by Ecorobotix, is a non-autonomous robot created for the precise application of herbicides. It is versatile and can be used in a large number of crops such as field vegetables, large crops, lawns, and many others. It integrates a navigation system, sensors, a processor, a color camera, two arms, solar panels, and self-charging batteries. It identifies an invasive plant in real time through the captured images. The processor estimates a position just above the plant and moves one of its arms to this position to spray a micro dose of herbicide. Its efficiency is four hectares per hour at a maximum speed of 7 km/h (Ecorobotix, 2024).

AVO is a self-contained solution, powered by interchangeable solar-powered batteries. It increases productivity and crop yields while reducing the environmental footprint. It has an efficiency of up to 10 hectares per day and is equipped with Lidar and ultrasonic sensors for obstacle detection and navigation. It follows a pre-programmed route using its GPS. It features reduced spray drift pollution through the use of specialized panels, nozzles, and adjuvants, and reduces herbicide use by up to 95% (Agtecher, 2023).

Myce-Vigne Autonomous is an autonomous robot supplied by Wall-YE. It operates for weeding, cutting, mowing, etc. It has a visual and GPS navigation system, solar panels to extend its autonomy in the field, and contains tools for different agricultural tasks in different crops (Wall-YE, 2020).

Myce-Agriculture is another autonomous robot with the technology of the company Wall-YE. It can be controlled from a smartphone and tackles small plants in flat fields. It is functional for planting, spraying, weeding, and harvesting. It has a GPS and visual navigation system guided by its cameras, detects weeds and pulls them, and has been used in vegetables such as carrots and other crops (Wall-YE, 2020).

Ibex2 is a prototype autonomous weed control robot under development in collaboration with Ibex Automation Ltd., the University of Lincoln, and the University of Leeds. It integrates proximity sensors, cameras, and a GNSS system into an autonomous precision weed detection and spraying system (Ibexautomation, 2020).

Tertill a prototype designed for gardening by Tertill (formerly Franklin Robotics), is equipped with sensors that allow it to recognize weeds and cut them with small scissors. It is powered by solar energy and is waterproof, making it suitable for both gardens and farmland. It has demonstrated a high efficiency between 60 and 75% in weed control (Sanchez and Gallandt, 2021).

TerraSentia unveiled by EarthSense Inc., is a robust autonomous multimodal robot that moves through the field using GPS and RTK, sending thousands of LIDAR laser pulses to scan its environment. This robot is capable of detailing the characteristics of the crop with its sensors and cameras, in addition to identifying weeds and calculating the amount of herbicide to be sprayed. It operates for 10 consecutive hours on one battery charge and does not operate overnight. TerraSentia is among the smallest agricultural robots available today. At 12.5 inches wide and about the same height, the 30-pound robot fits well between rows, and has already proven itself in a wide variety of crops including corn, soybeans, sorghum, cotton, wheat, tomatoes, strawberries, citrus crops, apple orchards, almond farms, and vineyards (Earthsense, 2021).

Adigo, the robot presented by Utstumo et al. (2018), uses machine vision and GPS location to navigate and operate in vegetable fields autonomously. It was designed with the specific task of herbicide application Drop on Demand (DoD). The DoD system is adjustable to account for differences in cultivation methods, number of crop rows, track width, and crop row height. The robot effectively controlled all weeds in the field trial with a tenfold reduction in herbicide use.

Machine vision-based AgBotII for weed detection and classification identifies crops and weeds using LBP and covariance features in its image processing stage. The weeding system combines chemical and mechanical methods for weed destruction using three different tools: an arrow-shaped hoe, a serrated tool, and a cutting tool. In field tests of the weed detection and classification system, it showed 92.3% accuracy on a variety of weed species. The heterogeneous weed management system can selectively apply a mechanical or chemical control method based on the weed species, demonstrating the potential for targeted robotic weed management (Bawden et al., 2017).

Hortibot a GPS-guided autonomous weed eradicator, uses Agrocom Vision’s row detection system. It includes easy adaptation of the tool holder to field conditions in terms of row spacing and plot size. It can recognize 25 different types of weeds, eliminating them with its modular tools for precision spraying and mechanical weed removal. Preliminary results show the potential to significantly reduce costs and still meet established operational requirements as it shows weed density reduction between 40 and 90% (Sørensen et al., 2007).

Tom is a weeding robot that uses HS cameras for weed detection. It records weed emergence data, detection, and plant re-finding, making herbicide applications more efficient. It is configurable for row and different height crops. It reduces herbicide applications by approximately 77%. Tom V4’s navigation system uses a Swiftnav GPS with 2 GNSS receivers to provide RTK position with an accuracy of 1 cm. Tom uses 4 rechargeable batteries (1.56 kwh) and has a data storage capacity of 4 TB (removable drive). Its recognition rate is 2.2 ha/h and a ground speed of 1.5 m/s. Tom uses Wilma, an AI advisory engine that absorbs billions of Tom’s data points and converts them into intelligence per plant (Small robot company, 2023a).

Dick another robot from the Small robot company, is in field trials. It has a robotic arm in which it can offer any technology for weed control. It will eliminate individual weeds using Rootwave weed elimination technology that uses electricity and AI. It has great performance and minimal use of chemicals when performing precision applications. This robot is intended in certain aspects for precise herbicide-resistant weed management. It will use the intelligence per plant processed by Wilma to control only weeds such as Blackgrass (Small robot company, 2023b).

Agribot a prototype robot developed by Chang; Xie; Chung (2021), is powered by energy. It has an RGB digital camera with 4K Ultra HD resolution and automatic focus. It also has proximity switches and is equipped with a weeding tool claw rake type. It moves at a speed of between 10 and 15 cm/s. The average efficiency results are 88.6% with an average detection precision of 90.7%.

The Laser weeding a prototype developed by Xiong et al. (2017), uses CV. It had a camera and laser pointers to detect crops and weeds and two steerable gimbals that controlled the laser pointers to target the weeds and control the platform in real time, performing continuous weeding. In tests, it showed a high accuracy rate of 97%.

The proposed prototype field robot by Sujaritha et al. (2017), uses a Raspberry Pi microcontroller and suitable input/output subsystems such as cameras, small light sources, and electric motors. The prototype was proposed for sugarcane cultivation in which it detects weeds with an accuracy of 92.9% in a processing time of 0.02 s.

A prototype robot developed by van Evert et al. (2011), for the detection and elimination of broadleaf weeds, its navigation system is executed by following a predefined route using a GPS of centimeter precision. The detection of weeds is performed by a camera and then destroyed by a cutting device. It achieves a rate of 93% weed elimination when they are in the form of a rosette.

The weeding robot proposed by Sori et al. (2018), navigates autonomously while eliminating weeds by shaking the soil to inhibit weed growth. It has capacitive touch sensors to detect rice plants and an azimuth sensor that is used when turning. It is designed to work in irrigated rice and has shown in experiments to be effective in controlling weeds, promoting the growth of rice seedlings.

Shah, Nasika and Otterpohl (2021), designed a high-level weeding robot to identify weeds at early stages. It is powered by solar panel energy which provides a renewable source of energy for robot movement. It has an RGB camera, sensors, and a single board computer. Accuracy values in the range of 88 to 98% were observed on the ground. A completely unknown dataset on plants and weeds, predictions in the range of 84% to 99% were observed in the case of weeds.

Although many robotic prototypes for weed control have been developed, challenges remain in converting them into commercially available products that achieve adequate farmer adoption. Current prototypes demonstrate the potential of robotic systems as practical tools for field-scale weed management. Ongoing advances are increasingly focused on localized weed control, including site-specific applications and non-chemical methods such as mechanical and physical control, with the goal of reducing development costs and improving weeding efficiency.

4. Herbicide discovery through AI

4.1 Advances in the use of artificial intelligence in computational drug discovery

Recent advances have led to remarkable progress in computational drug discovery and ligand optimization. AI is now capable of describing the properties of ligands and their interactions with biological targets. The increase in computational power and storage capacities allows for the processing of large amounts of data in shorter times than could have been accomplished decades ago (Muller et al., 2022; Orsolic et al., 2021; Silva-Mendonça et al., 2023; von Lilienfeld and Burke, 2020).

Nowadays, there is an opportunity to apply ML and DL in all phases of discovery, such as target validation, biomarker identification, diagnostics, and data analysis in screening trials of potential substances (Figure 1). Usually, the herbicide development program requires the identification of targets. Currently, it is one of the steps that require a large amount of genetic data, transcriptomic, proteomic, and metabolomic profiles, the same ones that are growing thanks to modern biology. The availability of this data in public databases presents new opportunities to be explored to identify and validate early new targets, in line with the multidimensional AI data analysis methods that are increasingly being applied (Jiménez-Luna et al., 2021).

Figure 1
Simplified scheme of the basic stages of using ML and DL for AI herbicide discovery. Adapted with modifications from Chen et al. (2023).

Another important point to consider is that although there is a large amount of data, not all of it has the characteristics of quality and accuracy. The generation of high-quality data for training and generating new AI models can be expensive, which requires a coordinated consortium involving companies, academic and political institutions that allow these approaches to generate a standard in the data. Currently, the data are not comprehensive and their applications are specific, data curation is key to providing reusable and reliable data (Jiménez-Luna et al., 2021). Datasets of combinations of substances, such as in the field of pharmaceuticals, seem to be a solution (Liu et al., 2024), but this solution in the area of herbicides may be a challenge that has only recently begun, improving the performance of ML models that predict synergistic effects.

4.2 Artificial intelligence implementations in computational herbicides or pesticides discovery

Currently, BASF, Bayer, Corteva, FMC, and Syngenta are the five leading global corporations that maintain herbicide discovery programs. This task is also in the hands of 19 small companies and may continue to grow in response to the increase in herbicide resistance and the rise in AI tools for herbicide research (Powles, 2022).

Among these, Agrematch has developed Agresense, an algorithmic system with AI, biology, and chemistry concepts for the development of molecules with the desired interaction between compounds and target organisms (Muller et al., 2022). This platform has a dataset of approximately ~1.2B, where the executed instances can last from seconds to days (Barker et al., 2023). Currently, Agrematch announced positive results codenamed AS9057, a natural herbicide discovered by AI, for the post-emergence control of weeds, including those of the genera Amaranthus and Chenopodium. The team expects that this discovery will include a novel mechanism of action, making their discovery highly significant, as this system has the ability to predict whether compounds belong to a class of the Herbicide Resistance Action Committee (HRAC) (Agrematch, 2022; Barker et al., 2023). Another case where this system is involved is in the molecule 4-chloro-2-penthenamides as a novel inhibitor of protoporphyrinogen oxidase (PPO), a well-known herbicide target site discovered by Dayan and Duke (2010).

Another company is Moa-technology that recently, at the 64th Annual Conference of the Weed Science Society of America (WSSA), revealed advances in this area, with field tests for herbicides of new modes of action (MOA),

Moa-technology examined more than 750,000 chemical and natural compounds using the Galaxy platform, identifying more than 60 new MOA, and were characterized in greenhouses. Moa’s proprietary state-of-the-art biotechnology platforms examine a vast diversity of natural and synthetic chemicals, combining a constellation of powerful genetic, biochemical, OMICs, and bioinformatics tools to rapidly identify the target protein, and enable new molecules, accelerating safety predictions and optimization in off-the-shelf agricultural herbicides (Jacq, 2024).

AgPlenus is a company with a slightly different approach: the company applies AI to predict plant enzymes that might be a good target and, through X-ray crystallography, looks for pockets in that protein where a molecule could bind. Then, it runs billions of computer simulations to predict molecules that are likely to inhibit its activity, being potential herbicides. Then, through computational chemistry AI and engineering of the molecules, it improves and designs herbicide candidates (Blois, 2022). This company has identified two APTH1 target proteins and another confidential one in which it works in partnership with Corteva Agriscience (AgPlenus, 2022). We can see that years ago, the cost of tooling and lack of computational capacity made these types of research unfeasible.

Enko is a start-up company focused on identifying promising bioactive molecules for agricultural applications. Its discovery approach involves screening a pool of approximately 140 billion compounds, supplied by Bayer and Syngenta, by incubating them with target enzymes to identify bound ligands for further evaluation. After incubation, the enzyme is washed, allowing unbound compounds to be removed while bound compounds remain attached. The DNA tags associated with the remaining compounds are then amplified to identify the ligands that bind to the target enzymes (Enko, 2021). BASF also states that it uses a similar approach because it is quick to achieve, but determining activity is another task, and bringing it to a real application is another challenge (Blois, 2022).

FortePhest, a start-up with investment from BASF (Hackmann, 2021), is inspired by the search for herbicides in natural products and has refined the productm-tyrosine, a non-coding amino acid produced by the grass Allelopathic fescue. This molecule appears to be a mimetic because it incorporates itself by mistake into a protein, causing deformations sufficient to cause the death of the plant (Blois, 2022).

The company OerthBio hopes to use proteolysis-targeted chimera (PROTAC) technology to make PROTAC molecules signal target proteins to be destroyed by the cell’s proteolysis system that is not critical targets but can have an effect, which also opens up many new targets. This is a different approach to existing herbicides that inhibit enzymatic activity, to affect the plant, since the process is found in all plant and animal cells that break down old or badly folded proteins. So, this company has achieved prominence in the discovery of two new MOA (WC-001 and WC-002) to combat weed resistance and provide greater selectivity, with improved safety profiles (OerthBio, 2024). It’s not surprising that this company is also using an AI tool to support its discoveries in the future (Blois, 2022).

The start-up Projini Agchem has focused its findings on how to disrupt the interactions between proteins. This technology relies on AI to predict where two proteins engage and then tries to design molecules that fit into that space and disrupt the interaction. The company has stated that it has already demonstrated the proof of concept and that it has a herbicide candidate with this approach different from classic enzyme inhibitors (Blois, 2022; Projini, 2024). Prospective for Advances in AI Technologies

The prospects of AI in weed management seem to complement the methodologies already available by automating them and enhancing them with sensors and algorithms capable of learning in real time and making accurate decisions. It promotes transfer learning along with the FSL and meta-learning to achieve this. While there are advancements, the effectiveness of these systems depends on integrating image recognition algorithms with dedicated advanced hardware to enhance performance and accuracy in weed control.

We recognize that AI is still far from fully answering the question posed by Duke and Dayan (2022) in their article titled “The Search for New Mechanisms of Action of Herbicides: Is There a ‘Holy Grail’?” However, AI technologies are increasingly being promoted as tools to identify biomarkers and targets for use in screening assays. Transcriptomic, proteomic, metabolomic, phenotypic, and weed imaging molecular tools are expected to further boost herbicide discovery. It is also evident that the availability of standard datasets from the areas that provide information for AI promises more reach for algorithms in the future. Significant advances have led to the discovery of new molecules, new herbicide targets, and new techniques, methods of screening and searching for targets have been found.

AI is expected to expand the availability of practical tools for farmers in the coming years. Successful applications already reported in agriculture highlight the potential of these technologies to improve operational efficiency, accuracy, and decision-making. The aim is to use AI technologies to make spraying technologies available that can vary the mixture of herbicides in real-time according to the distribution of weed species. Prospects indicate a continuous evolution in integrating advanced technologies, such as AI, ML, and computer vision, to optimize weed management and control. An increase in agricultural efficiency, reduced herbicide use, development of new herbicide MOA, and increased sustainability are expected as technological innovations continue to shape the agricultural landscape. The focus on automation, accuracy, and customization of farming practices is a trend that is likely to be driven by constant advancements in research and development of AI in weed management.

It is essential for weed specialists to support farmers as they adapt to new technologies. On-farm research is crucial because it helps farmers develop the understanding and skills needed to use these technologies effectively, while also allowing specialists to identify the specific challenges farmers face during adoption. This approach can support the development of more effective solutions adapted to farmers’ needs and help prevent these technologies from falling into a “trough of disillusionment” caused by misuse or unrealistic expectations. Ultimately, such efforts can promote the sustainability, productivity, and practical adoption of these tools in agriculture.

5. Shaping the future of agriculture through AI-based weed management

Intelligent control systems with robots and automated machinery in tasks such as weeding and site-specific weed management have demonstrated their potential, addressing challenges ranging from weed identification, high development costs of precise, real-time application systems, and the integration of agricultural practices (Upadhyay et al., 2024). In the future, AI technologies for plant-specific management are expected to evolve toward more accurate DL and computer vision systems, autonomous agricultural robots, integration of biotechnology and AI into resistant crops, advanced predictive models, sophisticated sensor networks, and closer collaboration between different agricultural crop systems. These advances promise a more efficient, sustainable, and autonomous management of specific plants in agriculture. Based on our review of the management of harmful plants and weeds, several advances in AI technologies are expected in the future:

  1. Increased Accuracy and Personalization: Improved AI algorithms are anticipated for more accurate identification of specific plants, allowing for more specific treatments, and further reducing reliance on herbicides.

  2. Integration of Multiple Technologies: The combination of AI with drones, robots, and advanced sensing systems could result in integrated solutions that more efficiently and sustainably address plant-specific management.

  3. Developing New Modes of Action: Continued research supported by AI could lead to the discovery and development of herbicides with innovative MOA, improving herbicide resistance and reducing environmental impact.

  4. Large-scale Automation: Agricultural automation could become more widespread as autonomous machines equipped with computer vision and machine learning (ML) technologies are increasingly used to detect and treat specific plants in real time.

  5. Increased Collaboration Between Industries: Collaboration between the agricultural and technology industries could intensify, generating more advanced solutions tailored to the specific needs of farmers.

  6. Focus on Quality Data: Given the importance of data in ML, the continued emphasis on collecting and curating high-quality data is expected to improve the effectiveness of AI models.

  7. Development of Low-Cost Systems: An effort is anticipated to develop more accessible AI solutions to enable widespread adoption, especially in smaller farming communities or regions with limited resources.

  8. Increase the Public Global Database for Machine-Learning Training: Expanding publicly available global databases for machine-learning training would democratize knowledge and promote collaboration, transparency, and innovation in weed management. Addressing privacy, data ownership, and intellectual-property concerns is essential for effective data sharing. Larger and more diverse datasets could improve weed detection, mechanical control, and weed-emergence prediction, while supporting preventive management strategies that reduce reliance on herbicides and help mitigate herbicide resistance and environmental impacts.

6. Conclusions

The lack of automated weed classification methods was once a major barrier to the development of mechanized and precision weed management. Advances in AI have helped bridge this gap by enabling more efficient weed detection, classification, and decision-making, allowing weed management to move beyond approaches traditionally limited by manual assessment.

Although there are advances in these technologies for the discovery and development of pesticides and robotic systems capable of accurately locating and identifying weeds, their use is still limited. It is necessary to deepen research for the development of tools that guarantee the handling of large volumes of data and tasks with greater precision. This aims to reduce the costs of this technology and consequently generate a reduction of energy, water, and agrochemicals used in agriculture.

AI contributes significantly to the overall management of weeds in agriculture, offering remarkable advances in detection, control, and decision-making. These are essential to address the challenges faced by the agricultural sector, such as weed resistance to herbicides. Nevertheless, it is still necessary for farmers to become familiar with these new technologies.

Acknowledgements

Declaration of Generative AI and AI-assisted Technologies in the Writing Process: During the preparation of this work, the authors used Microsoft Copilot, powered by OpenAI’s GPT-4 model, and Grammarly to improve readability, grammar, and language clarity. After using these tools, the authors reviewed and edited the text as needed and take full responsibility for the content of the publication.

Data availability

The data availability policy does not apply.

References

  • Abd Manaf MNH, Juraimi AS, Motmainna M, Che’Ya NN, Mat Su AS, Mohd Roslim MH et al. Detection of sedge weeds infestation in wetland rice cultivation using hyperspectral images and artificial intelligence: a review. Pertanika J Sci Technol. 2024;32(3):1317-34.https://doi.org/10.47836/pjst.32.318
    » https://doi.org/10.47836/pjst.32.318
  • AgPlenus. AgPlenus pipeline. Park Rehovot: AgPlenus; 2022[access Mar 13, 2026]. Available from: https://www.agplenus.com/products
    » https://www.agplenus.com/products
  • Agrematch. Agrematch announces trials success for AI-discovered natural herbicide. AgReads. Oct 18, 2022.
  • Agrointelli. Robotti. Aarhus: Agrointelli; 2024[access Mar 13, 2026]. Available from: https://agrointelli.com/robotti/
    » https://agrointelli.com/robotti/
  • Agtecher. AVO by Ecorobotix. Agtecher O local da Agri Tech. 2023[access Mar 13, 2026]. Available from: https://agtecher.com/product/avo-by-ecorobotix/
    » https://agtecher.com/product/avo-by-ecorobotix/
  • Agtecher. ecorobotix Generation 1 Autonomous Weeding Robot. 2020[access Mar 13, 2026]. Available from: https://agtecher.com/product/ecorobotixs-autonomous-weed-killer/
    » https://agtecher.com/product/ecorobotixs-autonomous-weed-killer/
  • Amean ZM, Low T, Hancock N. Automatic leaf segmentation and overlapping leaf separation using stereo vision. Array 2021;12:1-17. Available from: https://doi.org/10.1016/j.array.2021.100099
    » https://doi.org/10.1016/j.array.2021.100099
  • Anam I, Arafat N, Hafiz MS, Jim JR, Kabir MM, Mridha MF. A systematic review of UAV and AI integration for targeted disease detection, weed management, and pest control in precision agriculture. Smart Agric Technol. 2024;9:1-24. Available from: https://doi.org/10.1016/j.atech.2024.100647
    » https://doi.org/10.1016/j.atech.2024.100647
  • Andújar D, Calle M, Fernández-Quintanilla C, Ribeiro Á, Dorado J. Three-dimensional modeling of weed plants using low-cost photogrammetry. Sensors (Basel). 2018;18(4) Available from: https://doi.org/10.3390/s18041077
    » https://doi.org/10.3390/s18041077
  • Anzanello MJ, Yamashita G, Marcelo M, Fogliatto FS, Ortiz RS, Mariotti K et al. A genetic algorithm-based framework for wavelength selection on sample categorization. Drug Test Anal. 2017;9(8) :1172-81. Available from: https://doi.org/10.1002/dta.2138
    » https://doi.org/10.1002/dta.2138
  • Åstrand B, Baerveldt A-J. An agricultural mobile robot with vision-based perception for mechanical weed control. Autonom Robots. 2002;13:21-35. Available from: https://doi.org/10.1023/A:1015674004201
    » https://doi.org/10.1023/A:1015674004201
  • Badrinarayanan V, Kendall A, Cipolla R. SegNet: A Deep Convolutional Encoder-Decoder Architecture for Image Segmentation. IEEE Trans Pattern Anal Mach Intell. 2017;39(12) :2481-95. Available from: https://doi.org/10.1109/TPAMI.2016.2644615
    » https://doi.org/10.1109/TPAMI.2016.2644615
  • Bagavathiannan MV, Dobbs AM, Ginn D, Leon RG, Mirsky SB, Reberg-Horton CS et al. New directions in weed management and research using 3D imaging. Weed Sci. 2022;70(6) :641-7. Available from: https://doi.org/10.1017/wsc.2022.56
    » https://doi.org/10.1017/wsc.2022.56
  • Banerjee G, Sarkar U, Das S, Ghosh I. Artificial intelligence in agriculture: a literature survey. Int J Sci Res Comput Sci Eng Inf Technol. 2018;3(3) :1-8.
  • Bangert W, Kielhorn A, Rahe F, Albert A, Biber P, Grzonka S et al. Field-robot-based agriculture: “RemoteFarming. 1” and “BoniRob-Apps.” VDI Ber. 2013;2193:1-2.
  • Barker AL, Geva Y, Simonovsky E, Shemesh N, Phillip Y, Shub I et al. Protoporphyrinogen oxidase inhibitors discovered by Artificial Intelligence platform. Adv Weed Sci. 2023;41:1-11. Available from: https://doi.org/10.51694/AdvWeedSci/2023;41:00011
    » https://doi.org/10.51694/AdvWeedSci/2023;41:00011
  • Barrero O, Rojas D, Gonzalez C, Perdomo S. Weed detection in rice fields using aerial images and neural networks. In: 2016 XXI Symposium on Signal Processing, Images and Artificial Vision (STSIVA). IEEE; 2016. p. 1–4.
  • BASF. Bosch BASF Smart Farming and AGCO announce joint development and commercialization of smart spraying capabilities. Trade News. 2023[access Mar 13, 2026]. Available from: https://www.basf.com/global/en/media/news-re-leases/2023/04/p-23-159.html
    » https://www.basf.com/global/en/media/news-re-leases/2023/04/p-23-159.html
  • Bawden O, Kulk J, Russell R, McCool C, English A, Dayoub F et al. Robot for weed species plant-specific management. J Field Robot. 2017;34(6) :1179–99. Available from: https://doi.org/10.1002/rob.21727
    » https://doi.org/10.1002/rob.21727
  • Berge TW, Goldberg S, Kaspersen K, Netland J. Towards machine vision based site-specific weed management in cereals. Comput Electron Agric. 2012;81:79-86. Available from: https://doi.org/10.1016/j.compag.2011.11.004
    » https://doi.org/10.1016/j.compag.2011.11.004
  • Bhargava A, Sachdeva A, Sharma K, Alsharif MH, Uthansakul P, Uthansakul M. Hyperspectral imaging and its applications: a review. Heliyon. 2024;10(14):1-15. Available from: https://doi.org/10.1016/j.heliyon.2024.e33208
    » https://doi.org/10.1016/j.heliyon.2024.e33208
  • Bhoi SK, Jena KK, Panda SK, Long HV, Kumar R, Subbulakshmi P et al. An internet of things assisted unmanned aerial vehicle based artificial intelligence model for rice pest detection. Microprocess Microsyst. 2021;80:103607. Available from: https://doi.org/10.1016/j.micpro.2020.103607
    » https://doi.org/10.1016/j.micpro.2020.103607
  • Bishop CM. Pattern Recognition and Machine Learning. New York: Springer; 2006.
  • Blois M. The herbicide well runs dry. C&EN Glob Enterp. 2022;100(31) :22–8.
  • Bogue R. Robots poised to revolutionise agriculture. Ind Robot. 2016;43(5):450–6. Available from: https://doi.org/10.1108/IR-05-2016-0142
    » https://doi.org/10.1108/IR-05-2016-0142
  • Bosilj P, Aptoula E, Duckett T, Cielniak G. Transfer learning between crop types for semantic segmentation of crops versus weeds in precision agriculture. J Field Robot. 2020;37(1):7–19. Available from: https://doi.org/10.1002/rob.21869
    » https://doi.org/10.1002/rob.21869
  • Burks TF, Shearer SA, Payne FA. Classification of weed species using color texture features and discriminant analysis. Trans ASAE. 2000;43(2):441x8. Available from: https://doi.org/10.13031/2013.2723
    » https://doi.org/10.13031/2013.2723
  • Cai Z, Vasconcelos N. Cascade R-CNN: High quality object detection and instance segmentation. IEEE Trans Pattern Anal Mach Intell. 2021;43(5):1483–98. Available from: https://doi.org/10.1109/TPAMI.2019.2956516
    » https://doi.org/10.1109/TPAMI.2019.2956516
  • Cao D, Zhang H, Dhillon R. An organic weed control prototype using directed energy and deep learning. ArXiv. May 31, 2024[access Mar 13, 2026]. Available from: https://arxiv.org/abs/2405.21056
    » https://arxiv.org/abs/2405.21056
  • Carbon Robotics. Precision Laserthinning Using Laserweeder™: precision Weed Control for Large-Scale Specialty Crops. Carbon Robotics. 2024[access Mar 13, 2026]. Available in: https://carbon-robotics.com/
    » https://carbon-robotics.com/
  • Castrignano A, Buttafuoco G, Khosla R, Mouazen A, Moshou D, Naud O. Agricultural internet of things and decision support for precision smart farming. Amsterdam: Elsevier Science; 2020.
  • Chang CL, Xie BX, Chung SC. Mechanical control with a deep learning method for precise weeding on a farm. Agriculture. 2021;11(11):1-21. Available from: https://doi.org/10.3390/agriculture11111049
    » https://doi.org/10.3390/agriculture11111049
  • Chen LC, Papandreou G, Kokkinos I, Murphy K, Yuille AL. DeepLab: semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected CRFs. IEEE Trans Pattern Anal Mach Intell. 2018;40(4) :834-48. Available from: https://doi.org/10.1109/TPAMI.2017.2699184
    » https://doi.org/10.1109/TPAMI.2017.2699184
  • Chen W, Liu X, Zhang S, Chen S. Artificial intelligence for drug discovery: resources, methods, and applications. Mol Ther Nucleic Acids. 2023;31:691-702. Available from: https://doi.org/10.1016/j.omtn.2023.02.019
    » https://doi.org/10.1016/j.omtn.2023.02.019
  • Chostner B. See & spray: the next generation of weed control. Resource Mag. 2017;24(4) :4–5.
  • Christensen S, Heisel T, Walter AM, Graglia E. A decision algorithm for patch spraying. Weed Res. 2003;43(4):276–284. Available from: https://doi.org/10.1046/j.1365-3180.2003.00344.x
    » https://doi.org/10.1046/j.1365-3180.2003.00344.x
  • Cloud Google. What is Big Data? Google Cloud. 2023[access Mar 13, 2026]. Available from: https://cloud.google.com/learn/what-is-big-data
    » https://cloud.google.com/learn/what-is-big-data
  • Coleman GRY, Bender A, Hu K, Sharpe SM, Schumann AW, Wang Z et al. Weed detection to weed recognition: reviewing 50 years of research to identify constraints and opportunities for large-scale cropping systems. Weed Technol. 2022;36(6):741–57. Available from: https://doi.org/10.1017/wet.2022.84
    » https://doi.org/10.1017/wet.2022.84
  • Comba L, Biglia A, Aimonino DR, Barge P, Tortia C, Gay P. 2D and 3D data fusion for crop monitoring in precision agriculture. In: 2019 IEEE International Workshop on Metrology for Agriculture and Forestry (Metro-AgriFor). IEEE; 2019. p. 62–7.
  • Cordill C, Grift TE. Design and testing of an intra-row mechanical weeding machine for corn. Biosyst Eng. 2011;110(3):247-52. Available from: https://doi.org/10.1016/j.biosystemseng.2011.07.007
    » https://doi.org/10.1016/j.biosystemseng.2011.07.007
  • Creswell A, White T, Dumoulin V, Arulkumaran K, Sengupta B, Bharath AA. Generative adversarial networks: an overview. IEEE Signal Process Mag. 2018;35(1):53–65. Available from: https://doi.org/10.1109/MSP.2017.2765202
    » https://doi.org/10.1109/MSP.2017.2765202
  • CSIC-UPM. Eco-Innovative weeding with laser: sustainable weed management in agriculture with laser-based autonomous tools. Welaser. 2023[access Mar 13, 2026]. Available from: https://welaser-project.eu/
    » https://welaser-project.eu/
  • Dayan FE, Duke SO. Chapter 81 – Protoporphyrinogen Oxidase-Inhibiting Herbicides. In: Krieger R, editor. Hay’s handbook of pesticide toxicology. 3rd ed. New York: Academic Press; 2010. p. 1733–51.
  • Delbert C. This badass robot uses lasers to slay 100,000 weeds per hour. Popular Mechanics. 2021[access Mar 13, 2026]. Available from: https://www.popularmechanics.com/technology/robots/a36331690/autonomous-farming-robot-kills-weeds/
    » https://www.popularmechanics.com/technology/robots/a36331690/autonomous-farming-robot-kills-weeds/
  • Duarte MF, Davenport MA, Takhar D, Laska JN, Sun T, Kelly KF et al. Single-pixel imaging via compressive sampling. IEEE Signal Process Mag. 2008;25(2):83–91. Available from: https://doi.org/10.1109/MSP.2007.914730
    » https://doi.org/10.1109/MSP.2007.914730
  • Duke SO, Dayan FE. The search for new herbicide mechanisms of action: is there a ‘holy grail’? Pest Manag Sci. 2022;78(4):1303–13. Available from: https://doi.org/10.1002/ps.6726
    » https://doi.org/10.1002/ps.6726
  • Dyrmann M, Jørgensen RN, Midtiby HS. RoboWeedSupport: detection of weed locations in leaf occluded cereal crops using a fully convolutional neural network. Adv Anim Biosci. 2017;8(2):842–7. Available from: https://doi.org/10.1017/S2040470017000206
    » https://doi.org/10.1017/S2040470017000206
  • Dyrmann M, Karstoft H, Midtiby HS. Plant species classification using deep convolutional neural network. Biosyst Eng. 2016;151:72-80. Available from: https://doi.org/10.1016/j.biosystemseng.2016.08.024
    » https://doi.org/10.1016/j.biosystemseng.2016.08.024
  • Earthsense. TerraSentia robot. Earthsense News. 2021[access Mar 13, 2026]. Available from: https://www.earthsense.co/news/2019/10/12/earthsense-a-launch-partner-for-usdas-part-nerships-for-data-innovation
    » https://www.earthsense.co/news/2019/10/12/earthsense-a-launch-partner-for-usdas-part-nerships-for-data-innovation
  • Ecorobotix. ARA, the most precise sprayer on the market. Yverdon-les-Bains: Ecorobotix; 2024[access Mar 13, 2026]. Available from: https://ecorobotix.com/en/ara/
    » https://ecorobotix.com/en/ara/
  • Engineers Australia. Sydney Uni unleashes RIPPA the weed killer. Barton: Engineers Australia; 2016[access Mar 13, 2026]. Available from: https://portal.engineersaustralia.org.au/news/sydney-uni-unleash-es-rippa-weed-killer
    » https://portal.engineersaustralia.org.au/news/sydney-uni-unleash-es-rippa-weed-killer
  • Enko. The Traditional Way – The Enko Way. 2021[access Mar 13, 2026]. Available from: https://www.enko.ag/
    » https://www.enko.ag/
  • Espejo-Garcia B, Mylonas N, Athanasakos L, Fountas S, Vasilakoglou I. Towards weeds identification assistance through transfer learning. Comput Electron Agric. 2020;171. Available from: https://doi.org/10.1016/j.compag.2020.105306
    » https://doi.org/10.1016/j.compag.2020.105306
  • Espejo-Garcia B, Mylonas n, Athanasakos L, Fountas S. Improving weeds identification with a repository of agricultural pre-trained deep neural networks. Comput Electron Agric. 2020;175:105593. Available from: https://doi.org/10.1016/j.compag.2020.105593
    » https://doi.org/10.1016/j.compag.2020.105593
  • Espejo-Garcia B, Mylonas N, Athanasakos L, Vali E, Fountas S. Combining generative adversarial networks and agricultural transfer learning for weeds identification. Biosyst Eng. 2021;204:79-89. Available from: https://doi.org/10.1016/j.biosystemseng.2021.01.014
    » https://doi.org/10.1016/j.biosystemseng.2021.01.014
  • Evert FK van, Fountas S, Jakovetic D, Crnojevic V, Travlos I, Kempenaar C. Big Data for weed control and crop protection. Weed Res. 2017;57(4):218–33. Available from: https://doi.org/10.1111/wre.12255
    » https://doi.org/10.1111/wre.12255
  • Fakhar M, Khalid M. Satellites to agricultural fields: the role of remote sensing in precision agriculture. Biol Agric Sci Res J. 2023;2023(14):1-10. Available from: https://doi.org/10.54112/basrj.v2023i1.14
    » https://doi.org/10.54112/basrj.v2023i1.14
  • Fawakherji M, Potena C, Pretto A, Bloisi DD, Nardi D. Multi-spectral image synthesis for crop/weed segmentation in precision farming. Rob Auton Syst. 2021;146. Available from: https://doi.org/10.1016/j.robot.2021.103861
    » https://doi.org/10.1016/j.robot.2021.103861
  • Fawakherji M, Potena C, Prevedello I, Pretto A, Bloisi DD, Nardi D. Data augmentation using GANs for crop/weed segmentation in precision farming. In: 2020 IEEE Conference on Control Technology and Applications (CCTA). IEEE; 2020. p. 279–84.
  • Ferentinos KP. Deep learning models for plant disease detection and diagnosis. Comput Electron Agric. 2018;145:311–8. Available from: https://doi.org/10.1016/j.compag.2018.01.009
    » https://doi.org/10.1016/j.compag.2018.01.009
  • Fernández-Quintanilla C, Peña JM, Andújar D, Dorado J, Ribeiro A, López-Granados F. Is the current state of the art of weed monitoring suitable for site-specific weed management in arable crops? Weed Res. 2018;58:259–72. Available from: https://doi.org/10.1111/wre.12307
    » https://doi.org/10.1111/wre.12307
  • Ferrari Ggrow Tech. Choose Most suitable Prod. Ferrari Ggrow Tech; 2024[access Mar 13, 2026]. Available from: https://ferrarigrowtech.com/en/automated-weeders/28-remoweed.html
    » https://ferrarigrowtech.com/en/automated-weeders/28-remoweed.html
  • Ferreira AS, Matte Freitas D, Gonçalves da Silva G, Pistori H, Theophilo Folhes M. Weed detection in soybean crops using ConvNets. Comput Electron Agric. 2017;143:314–24. Available from: https://doi.org/10.1016/j.compag.2017.10.027
    » https://doi.org/10.1016/j.compag.2017.10.027
  • Finn C, Abbeel P, Levine S. Model-agnostic meta-learning for fast adaptation of deep networks. In: International Conference on Machine Learning. PMLR; 2017. p. 1126–35.
  • Fountas S, Malounas I, Athanasakos L, Avgoustakis I, Espejo-Garcia B. AI-Assisted vision for agricultural robots. AgriEngineering. 2022;4(3):674–94. Available from: https://doi.org/10.3390/agriengineering4030043
    » https://doi.org/10.3390/agriengineering4030043
  • Fukushima K, Miyake S. Neocognitron: a self-organizing neural network model for a mechanism of visual pattern recognition. In: Amari S, Arbib MA, editors. Competition and cooperation in neural nets. Berlin: Springer Berlin Heidelberg; 1982. p. 267–85.
  • Fukushima K. Neocognitron: a self-organizing neural network model for a mechanism of pattern recognition unaffected by shift in position. Biol Cybern. 1980;36:193–202. Available from: https://doi.org/10.1007/BF00344251
    » https://doi.org/10.1007/BF00344251
  • Gao J, Nuyttens D, Lootens P, He Y, Pieters JG. Recognising weeds in a maize crop using a random forest machine-learning algorithm and near-infrared snapshot mosaic hyperspectral imagery. Biosyst Eng. 2018;170:39–50. Available from: https://doi.org/10.1016/j.biosystemseng.2018.03.006
    » https://doi.org/10.1016/j.biosystemseng.2018.03.006
  • García CG, Meana-Llorián D, Lovelle JMC. A review about Smart Objects, Sensors, and Actuators. Int J Interact Multimed Artif Intell. 2017;4(3):7–10. Available from: https://doi.org/10.9781/ijimai.2017.431
    » https://doi.org/10.9781/ijimai.2017.431
  • Garford Corp. Garford farm machinery products. Peterborough: Garford; 2023[access Mar 13, 2026]. Available from: https://garford.com/products/
    » https://garford.com/products/
  • Gerhards R, Oebel H. Practical experiences with a system for site-specific weed control in arable crops using real-time image analysis and GPS-controlled patch spraying. Weed Res. 2006;46:185–93. Available from: https://doi.org/10.1111/j.1365-3180.2006.00504.x
    » https://doi.org/10.1111/j.1365-3180.2006.00504.x
  • Ghatrehsamani S, Jha G, Dutta W, Molaei f, Nazrul F, Fortin M et al. Artificial Intelligence tools and techniques to combat herbicide resistant weeds: a review. Sustainability. 2023;15(3):1–18. Available from: https://doi.org/10.3390/su15031843
    » https://doi.org/10.3390/su15031843
  • Giles DK, Slaughter DC. Precision band spraying with machine-vision guidance and adjustable yaw nozzles. Trans ASAE. 1997;40(1) :29–36. Available from: https://doi.org/10.13031/2013.21240
    » https://doi.org/10.13031/2013.21240
  • Girshick RB. Fast R-CNN. CoRR. 2015;abs/1504.0.
  • Goel P, Prasher S, Landry J, Patel R, Bonnell R, Viau A et al. Potential of airborne hyperspectral remote sensing to detect nitrogen deficiency and weed infestation in corn. Comput Electron Agric. 2003;38:99–124. Available from: https://doi.org/10.1016/S0168-1699(02)00138-2
    » https://doi.org/10.1016/S0168-1699(02)00138-2
  • Goldstein T, Osher S. The split Bregman method for L1-regularized problems. SIAM J Imaging Sci. 2009;2(2):323–43. Available from: https://doi.org/10.1137/080725891
    » https://doi.org/10.1137/080725891
  • Goodfellow I, Bengio Y, Courville A. Deep learning (adaptive computation and machine learning series). Cambridge: MIT Press; 2016.
  • Göröcs Z, Rivenson Y, Ceylan Koydemir H, Tseng D, Troy TL, Demas V et al. Quantitative fluorescence sensing through highly autofluorescent, scattering, and absorbing media using mobile microscopy. ACS Nano. 2016;10:8989-99. Available from: https://doi.org/10.1021/acsna-no.6b05129
    » https://doi.org/10.1021/acsna-no.6b05129
  • Greeneye Technology. Introducing Greeneye’s selective spraying system, Designed for seamless integration and maximum profitability. Greeneye Technol; 2023[access Mar 13, 2026]. Available from: https://greeneye.ag/technology/
    » https://greeneye.ag/technology/
  • Guan Y, Chen D, He K, Liu Y, Li L. Review on research and application of variable rate spray in agriculture. In: 2015 IEEE 10th Conference on Industrial Electronics and Applications (ICIEA). IEEE; 2015. p. 1575–80.
  • Gϋldenring R, Boukas E, Ravn O, Nalpantidis L. Few-leaf learning: Weed segmentation in grasslands. In: 2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE; 2021. p. 3248–54.
  • Gϋldenring R, Nalpantidis L. Self-supervised contrastive learning on agricultural images. Comput Electron Agric. 2021;191. Available from: https://doi.org/10.1016/j.compag.2021.106510
    » https://doi.org/10.1016/j.compag.2021.106510
  • Gutjahr C, Sökefeld M, Gerhards R. Evaluation of two patch spraying systems in winter wheat and maize. Weed Res. 2012;52(6) :510–19. Available from: https://doi.org/10.1111/j.1365-3180.2012.00943.x
    » https://doi.org/10.1111/j.1365-3180.2012.00943.x
  • Ha N, Xu K, Ren G, Mitchell A, Ou JZ. Machine learning-enabled smart sensor systems. Adv Intell Syst. 2020;2(9). Available from: https://doi.org/10.1002/aisy.202000063
    » https://doi.org/10.1002/aisy.202000063
  • Hackmann C. FortePhest Ltd. Ludwigshafen: BASF; 2021[access Mar 13, 2026]. Available from: https://www.basf.com/global/en/who-we-are/organization/group-companies/BASF_Venture-Capital/portfolio/fortephest.html
    » https://www.basf.com/global/en/who-we-are/organization/group-companies/BASF_Venture-Capital/portfolio/fortephest.html
  • Hague T, Tillett ND, Wheeler H. Automated crop and weed monitoring in widely spaced cereals. Precis Agric. 2006;7:21-32. Available from: https://doi.org/10.1007/s11119-005-6787-1
    » https://doi.org/10.1007/s11119-005-6787-1
  • Han D, Kim J, Kim J. Deep pyramidal residual networks. In: 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). IEEE; 2017. p. 6307–15.
  • Hasan ASMM, Sohel F, Diepeveen D, Laga H, Jones MGK. A survey of deep learning techniques for weed detection from images. Comput Electron Agric. 2021;184. Available from: https://doi.org/10.1016/j.compag.2021.106067
    » https://doi.org/10.1016/j.compag.2021.106067
  • He K, Gkioxari G, Dollar P, Girshick RB. Mask R-CNN. In: 2017 IEEE International Conference on Computer Vision (ICCV). IEEE; 2017. p. 2980–8.
  • He K, Zhang X, Ren S, Sun J. Deep Residual Learning for Image Recognition. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 2016.
  • Hoogi A, Subramaniam A, Veerapaneni R, Rubin DL. Adaptive estimation of active contour parameters using convolutional neural networks and texture analysis. IEEE Trans Med Imaging. 2016;36:781-91. Available from: https://doi.org/10.1109/TMI.2016.2628084
    » https://doi.org/10.1109/TMI.2016.2628084
  • Hu C, Xie S, Song D, Thomasson JA, IV RGH, Bagavathiannan M. Algorithm and system development for robotic micro-volume herbicide spray towards precision weed management. IEEE Robot Autom Lett. 2022;7(4):11633–40. Available from: https://doi.org/10.1109/LRA.2022.3191240
    » https://doi.org/10.1109/LRA.2022.3191240
  • Hu K, Wang Z, Coleman G, Bender A, Yao T, Zeng S et al. Deep learning techniques for in-crop weed identification: a review. 2021.
  • Huang G, Liu Z, Van Der Maaten L, Weinberger KQ. Densely connected convolutional networks. In: 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). IEEE; 2017. p. 2261–9.
  • Huang Y, Lee MA, Thomson SJ, Reddy KN. Ground-based hyperspectral remote sensing for weed management in crop production. Int J Agric Biol Eng. 2016;9(2):98–109.
  • Ibexautomation. Ibex2. IBEX. 2020[access Mar 13, 2026]. Available from: http://www.ibexautomation.co.uk/
    » http://www.ibexautomation.co.uk/
  • International Business Machine – IBM. What are neural networks? Top Think [Internet]. 2020b[access Mar 13, 2026]. Available from: https://www.ibm.com/topics/neural-networks
    » https://www.ibm.com/topics/neural-networks
  • Im J, Jensen JR. Hyperspectral remote sensing of vegetation: geography compass. Boca Raton: CRC Press; 2008.
  • Integrated Electronics – INTEL. CPU vs. GPU: What’s the difference? Docs-processors. Santa Clara: Intel; 2024[access Mar 13, 2026]. Available from: https://www.intel.com/content/www/us/en/products/docs/processors/cpu-vs-gpu.html
    » https://www.intel.com/content/www/us/en/products/docs/processors/cpu-vs-gpu.html
  • International Business Machine – IBM. What is an autoencoder? Top Think. 2023[access Mar 13, 2026]. Available from: https://www.ibm.com/topics/autoencoder
    » https://www.ibm.com/topics/autoencoder
  • International Business Machine – IBM. What is Bagging? Top Think. 2024[access Mar 13, 2026]. Available from: https://www.ibm.com/topics/bagging
    » https://www.ibm.com/topics/bagging
  • International Business Machine – IBM. What is deep learning? Top Think 2020[access Mar 13, 2026]. Available from: https://www.ibm.com/topics/deep-learning
    » https://www.ibm.com/topics/deep-learning
  • International Business Machine – IBM. What is machine learning? TopThink. 2022[access Mar 13, 2026]. Available from: https://www.ibm.com/topics/machine-learning
    » https://www.ibm.com/topics/machine-learning
  • International Business Machine – IBM. What is the internet of things (IoT)? Top Think. 2016[access Mar 13, 2026]. Available from: https://www.ibm.com/topics/internet-of-things
    » https://www.ibm.com/topics/internet-of-things
  • Ioffe S, Szegedy C. Batch normalization: accelerating deep network training by reducing internal covariate shift. CoRR. 2015.
  • Islam N, Rashid MM, Wibowo S, Xu C-Y, Morshed A, Wasimi SA et al. Early weed detection using image processing and machine learning techniques in an Australian chilli farm. Agriculture. 2021;11(5) :1-13. Available from: https://doi.org/10.3390/agriculture11050387
    » https://doi.org/10.3390/agriculture11050387
  • Jacq X. Moa reveals data on next-generation herbicides with new modes of action and progress towards field trials: galaxy, a high-content imaging platf. Enabling Nov. Herbic. Discov. Jan 25, 2024[access Mar 13, 2026]. Available from: https://www.moa-technology.com/post/moa-reveals-data-on-next-generation-herbicides-with-new-modes-of-action-and-progress-to-wards-field-trials
    » https://www.moa-technology.com/post/moa-reveals-data-on-next-generation-herbicides-with-new-modes-of-action-and-progress-to-wards-field-trials
  • Jha K, Doshi A, Patel P, Shah M. A comprehensive review on automation in agriculture using artificial intelligence. Artif Intell Agric. 2019;2:1-12. Available from: https://doi.org/10.1016/j.aiia.2019.05.004
    » https://doi.org/10.1016/j.aiia.2019.05.004
  • Jiménez-Luna J, Grisoni F, Weskamp N, Schneider G. Artificial intelligence in drug discovery: recent advances and future perspectives. Expert Opin Drug Discov. 2021;16:949-59. Available from: https://doi.org/10.1080/17460441.2021.1909567
    » https://doi.org/10.1080/17460441.2021.1909567
  • Jin X, Bagavathiannan M, Maity A, Chen Y, Yu J. Deep learning for detecting herbicide weed control spectrum in turfgrass. Plant Methods. 2022a;18:94. Available from: https://doi.org/10.1186/s13007-022-00929-4
    » https://doi.org/10.1186/s13007-022-00929-4
  • Jin X, Bagavathiannan M, McCullough PE, Chen Y, Yu J. A deep learning-based method for classification, detection, and localization of weeds in turfgrass. Pest Manag Sci. 2022B;78:4809-21. Available from: https://doi.org/10.1002/ps.7102
    » https://doi.org/10.1002/ps.7102
  • Jin X, Che J, Chen Y. Weed identification using deep learning and image processing in vegetable plantation. IEEE Access. 2021;9:10940-50. Available from: https://doi.org/10.1109/ACCESS.2021.3050296
    » https://doi.org/10.1109/ACCESS.2021.3050296
  • Jin X, Liu T, Chen Y, Yu J. Deep learning-based weed detection in turf: a review. Agronomy. 2022c;12(2):3051. Available from: https://doi.org/10.3390/agronomy12123051
    » https://doi.org/10.3390/agronomy12123051
  • Jordan NR, Jannink JL. Assessing the practical importance of weed evolution: a research agenda. Weed Res. 1997;37:237–46.
  • Kaya M, Bilge HŞ. Deep metric learning: A survey. Symmetry (Basel). 2019;11:1066.
  • Khan A, Sohail A, Ali A. A new channel boosted convolution neural network using transfer learning. CoRR. Apr 2018. Available from: https://doi.org/10.48550/arXiv.1804.08528
    » https://doi.org/10.48550/arXiv.1804.08528
  • Khan A, Sohail A, Zahoora U, Qureshi AS. A survey of the recent architectures of deep convolutional neural networks. Artif Intell Rev. 2020;53:5455–516. Available from: https://doi.org/10.1007/s10462-020-09825-6
    » https://doi.org/10.1007/s10462-020-09825-6
  • Kipkorir SM, Fredrick AM, Maake F. Adoption of artificial intelligence in agriculture in the developing nations: a review. J Lang Technol Entrep Africa. 2021;12.
  • Kujawa S, Niedbała G. Artificial neural networks in agriculture. Agriculture. 2021;11:497. Available from: https://doi.org/10.3390/agriculture11060497
    » https://doi.org/10.3390/agriculture11060497
  • Kunz C, Weber JF, Gerhards R. Benefits of Precision Farming Technologies for Mechanical Weed Control in Soybean and Sugar Beet: comparison of precision hoeing with conventional mechanical weed control. Agronomy. 2015;5(2):130–42. Available from: https://doi.org/10.3390/agronomy5020130
    » https://doi.org/10.3390/agronomy5020130
  • Kutugata M. How a national image repository can transform agriculture. Grow-iwm. 2023[access Mar 13, 2026]. Available from: https://growiwm.org/author/grow-iwm/
    » https://growiwm.org/author/grow-iwm/
  • Lamm RD, Slaughter DC, Giles DK. Precision weed control system for cotton. Trans ASAE. 2002;45(1):231. Available from: https://doi.org/10.13031/2013.7861
    » https://doi.org/10.13031/2013.7861
  • LeCun Y, Bengio Y, Hinton G. Deep learning. Nature. 2015;521(7553):436-44. Available from: https://doi.org/10.1038/nature14539
    » https://doi.org/10.1038/nature14539
  • LeCun Y, Boser B, Denker JS, Henderson D, Howard RE, Hubbard W et al. Backpropagation applied to handwritten zip code recognition. Neural Comput. 1989;1(4):541–51. Available from: https://doi.org/10.1162/neco.1989.1.4.541
    » https://doi.org/10.1162/neco.1989.1.4.541
  • Lecun Y, Bottou L, Bengio Y, Haffner P. Gradient-based learning applied to document recognition. Proc IEEE. 1998;86(11):2278–324. Available from: https://doi.org/10.1109/5.726791
    » https://doi.org/10.1109/5.726791
  • Lee I, Lee K. The Internet of Things (IoT): applications, investments, and challenges for enterprises. Bus Horiz. 2015;58(4):431–40. Available from: https://doi.org/10.1016/j.bushor.2015.03.008
    » https://doi.org/10.1016/j.bushor.2015.03.008
  • Lee WS, Slaughter DC, Giles DK. Robotic weed control system for tomatoes. Precis Agric. 1999;1(1):95–113. Available from: https://doi.org/10.1023/A:1009977903204
    » https://doi.org/10.1023/A:1009977903204
  • Li L, Fan Y, Huang X, Tian L. Real-time UAV weed scout for selective weed control by adaptive robust control and machine learning algorithm. In: 2016 ASABE International Meeting. American Society of Agricultural and Biological Engineers; 2016.
  • Li N, Zhang X, Zhang C, Ge L, He Y, Wu X. Review of machine-vision-based plant detection technologies for robotic weeding. In: 2019 IEEE International Conference on Robotics and Biomimetics (ROBIO). IEEE; 2019. p. 2370–7.
  • Li Y, Yang J. Meta-learning baselines and database for few-shot classification in agriculture. Comput Electron Agric. 2021;182. Available from: https://doi.org/10.1016/j.compag.2021.106055
    » https://doi.org/10.1016/j.compag.2021.106055
  • Lin T-Y, Dollar P, Girshick R, He K, Hariharan B, Belongie S. Feature pyramid networks for object detection. In: 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). IEEE; 2017. p. 936–44.
  • Liu B, Bruch R. Weed detection for selective spraying: a review. Curr Robot Reports. 2020;1:19-26. Available from: https://doi.org/10.1007/s43154-020-00001-w
    » https://doi.org/10.1007/s43154-020-00001-w
  • Liu C. AI weed-killing drones are coming for the mega farms. Bloom Green Tech. 2023[access Mar 13, 2026]. Available from: https://www.bloomberg.com/news/features/2023-04-19/ai-drone-cuts-chemi-cal-use-in-big-farm-weeding
    » https://www.bloomberg.com/news/features/2023-04-19/ai-drone-cuts-chemi-cal-use-in-big-farm-weeding
  • Liu M, Srivastava G, Ramanujam J, Brylinski M. Augmented drug combination dataset to improve the performance of machine learning models predicting synergistic anticancer effects. Sci Rep. 2024;14. Available from: https://doi.org/10.1038/s41598-024-51940-9
    » https://doi.org/10.1038/s41598-024-51940-9
  • Liu W, Anguelov D, Erhan D, Szegedy C, Reed S, Fu C-Y et al. Ssd: Single shot multibox detector. In: Computer Vision-ECCV 2016: 14th European Conference, Amsterdam, The Netherlands, October 11–14, 2016, Proceedings, Part I 14. Springer; 2016. p. 21–37.
  • Long J, Shelhamer E, Darrell T. Fully convolutional networks for semantic segmentation. In: 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). IEEE; 2015. p. 3431–40.
  • López-Correa JM, Moreno H, Ribeiro A, Andújar D. Intelligent weed management based on object detection neural networks in tomato crops. Agronomy. 2022;12(12):2953. Available from: https://doi.org/10.3390/agronomy12122953
    » https://doi.org/10.3390/agronomy12122953
  • López-Granados F, Torres-Sánchez J, Serrano-Pérez A, Castro AI, Mesas-Carrascosa FJ, Pena JM. Early season weed mapping in sunflower using UAV technology: variability of herbicide treatment maps against weed thresholds. Precis Agric. 2016;17(2):183–99. Available from: https://doi.org/10.1007/s11119-015-9415-8
    » https://doi.org/10.1007/s11119-015-9415-8
  • Lu J, Gong P, Ye J, Zhang J, Zhang C. A survey on machine learning from few samples. Pattern Recognit. 2023;139. Available from: https://doi.org/10.1016/j.patcog.2023.109480
    » https://doi.org/10.1016/j.patcog.2023.109480
  • Luo J, You Y, Wang D, Sun X, Lv J, Ma W et al. Peach tree detection for weeding robot based on Faster-RCNN. In: 2020 ASABE Annual International Virtual Meeting. St. Joseph (MI) : American Society of Agricultural and Biological Engineers; 2020.
  • Lv E, Wang X, Cheng Y, Yu Q. Deep ensemble network based on multi-path fusion. Art Intell Rev. 2019;52(1) :151-68. Available from: https://doi.org/10.1007/s10462-019-09708-5
    » https://doi.org/10.1007/s10462-019-09708-5
  • Mathworks. What is semantic segmentation? Natick: Mathworks; 2024[access Mar 13, 2026]. Available from: https://www.math-works.com/solutions/image-video-processing/semantic-segmentation.html
    » https://www.math-works.com/solutions/image-video-processing/semantic-segmentation.html
  • Mathworks. What is the genetic algorithm? Natick: Mathworks; 2023[access Mar 13, 2026]. Available from: https://www.mathworks.com/help/gads/what-is-the-genetic-algorithm.html
    » https://www.mathworks.com/help/gads/what-is-the-genetic-algorithm.html
  • McCarthy J, Minskl ML, Rochester N, Shannon CE. A Proposal for the Dartmouth summer research project on artificial intelligence. 1955.
  • McCarthy J. What is artificial intelligence? Stanford: Stanford University; 2007[access Mar 13, 2026]. Available from: http://jmc.stanford.edu/articles/whatisai.html
    » http://jmc.stanford.edu/articles/whatisai.html
  • McKinion JM, Lemmon HE. Expert systems for agriculture. Comput Electron Agric. 1985;1(1):31–40. Available from: https://doi.org/10.1016/0168-1699(85)90004-3
    » https://doi.org/10.1016/0168-1699(85)90004-3
  • Meshram AT, Vanalkar AV, Kalambe KB, Badar AM. Pesticide spraying robot for precision agriculture: a categorical literature review and future trends. J Field Robot. 2022;39(2):153–71. Available from: https://doi.org/10.1002/rob.22043
    » https://doi.org/10.1002/rob.22043
  • Mink R, Dutta A, Peteinatos GG, Sökefeld M, Engels JJ, Hahn M et al. Multi-temporal site-specific weed control of Cirsium arvense (L.) Scop. and Rumex crispus L. in maize and sugar beet using unmanned aerial vehicle based mapping. Agriculture. 2018;8(5):1–14. Available from: https://doi.org/10.3390/agriculture8050065
    » https://doi.org/10.3390/agriculture8050065
  • Mishra AM, Gautam V. Weed species identification in different crops using precision weed management: a review. In: International Scientific Indian Conference (ISIC); 2021. p. 180–94.
  • Mitchell TM. Machine learning. New York: McGraw-Hill; 1997.
  • Moazzam SI, Khan US, Tiwana MI, Iqbal J, Qureshi WS, Shah SI. A review of application of deep learning for weeds and crops classification in agriculture. In: 2019 International Conference on Robotics and Automation in Industry (ICRAI). IEEE; 2019. p. 1–6.
  • Mohanty SP, Hughes D, Salathé M. Inference of plant diseases from leaf images through deep learning. Front Plant Sci. 2016;7.
  • Morales EF, Escalante HJ. Chapter 6 – A brief introduction to supervised, unsupervised, and reinforcement learning. In: Torres-García AA, editor. Biosignal processing and classification using computational learning and intelligence. Oxford: Academic Press; 2022. p. 111–29.
  • Muchhadiya R, Kumawat PD, Gohil B, Sakarvadia H. Precision weed management using artificial intelligence tools and techniques for sustainable agriculture. Int J Plant Soil Sci. 2024;6:606-11.
  • Muller C, Rabal O, Diaz Gonzalez C. Artificial Intelligence, Machine Learning, and Deep Learning in Real-Life Drug Design Cases. In: Heifetz A, editor. Artificial Intelligence in Drug Design. New York: Springer; 2022. p. 383–407.
  • Murad NY, Mahmood T, Forkan ARM, Morshed A, Jayaraman PP et al. Weed detection using deep learning: a systematic literature review. Sensors. 2023;23(7):3670. Available from: https://doi.org/10.3390/s23073670
    » https://doi.org/10.3390/s23073670
  • Naϊo Technology. Dino cultivator helps california vegetable growers cut costs. Escalquens: Naϊo Technology; 2021[access Mar 13, 2026]. Available from: https://www.naio-technologies.com/en/news/dino-cultivator-helps-california-vegetable-growers-cut-costs/
    » https://www.naio-technologies.com/en/news/dino-cultivator-helps-california-vegetable-growers-cut-costs/
  • Naϊo Technology. Orio: A Robot to weed garlic, but not only! Escalquens: Naϊo Technology; 2023[access Mar 13, 2026]. Available from: https://www.naio-technologies.com/en/news/a-robot-to-weed-garlic-but-not-only/
    » https://www.naio-technologies.com/en/news/a-robot-to-weed-garlic-but-not-only/
  • Naϊo Technology. Oz – L’assistant Agricole pour les tâches chronophages pénibles. Escalquens: Naϊo Technology; 2023[access Mar 13, 2026]. Available from: https://www.naio-technologies.com/oz/
    » https://www.naio-technologies.com/oz/
  • Naϊo Technology. Ted: Technical Specifications. Escalquens: Naϊo Technology; 2024[access Mar 13, 2026]. Available from: https://www.naio-technologies.com/en/ted/
    » https://www.naio-technologies.com/en/ted/
  • Nevavuori P, Narra N, Lipping T. Crop yield prediction with deep convolutional neural networks. Comput Electron Agric. 2019;163:104859. Available from: https://doi.org/10.1016/j.compag.2019.104859
    » https://doi.org/10.1016/j.compag.2019.104859
  • Neve P, Norsworthy JK, Smith KL, Zelaya IA. Modelling evolution and management of glyphosate resistance in Amaranthus palmeri. Weed Res. 2011;51(2):99–112. Available from: https://doi.org/10.1111/j.1365-3180.2010.00838.x
    » https://doi.org/10.1111/j.1365-3180.2010.00838.x
  • Niu Z, Rehman T, Young J, Johnson WG, Yokoo T, Young B et al. Hyperspectral analysis for discriminating herbicide site of action: a novel approach for accelerating herbicide research. Sensors. 2023;23(23):1-17. Available from: https://doi.org/10.3390/s23239300
    » https://doi.org/10.3390/s23239300
  • Nixon M, Aguado A. Feature extraction and image processing for computer vision. 4th ed. London: Academic Press; 2019.
  • Norremark M, Griepentrog HW, Nielsen J, Sogaard HT. The development and assessment of the accuracy of an autonomous GPS-based system for intra-row mechanical weed control in row crops. Biosyst Eng. 2008;101(4) :396-410. Available from: https://doi.org/10.1016/j.biosystemseng.2008.09.007
    » https://doi.org/10.1016/j.biosystemseng.2008.09.007
  • Nwankpa C, Ijomah W, Gachagan A, Marshall S. Activation functions: comparison of trends in practice and research for deep learning. CoRR. 2018[access Mar 13, 2026]. Available from: https://arxiv.org/abs/1811.03378
    » https://arxiv.org/abs/1811.03378
  • Oerke EC, Gerhards R, Menz G, Sikora RA, editors. Precision crop protection: the challenge and use of heterogeneity. Dordrecht: Springer; 2010.
  • Olsen A, Konovalov DA, Philippa B, Ridd P, Wood JC, Johns J et al. Deep-Weeds: a multiclass weed species image dataset for deep learning. Sci Rep. 2019;9(1) :1-12. Available from: https://doi.org/10.1038/s41598-018-38343-3
    » https://doi.org/10.1038/s41598-018-38343-3
  • Omankwu OC, Nwagu CK, Inyiama H. Robotics and artificial intelligence: differences and similarities. Int J Comput Sci Inf Secur. 2017;15(1):15-21.
  • Oršolić D, Pehar V, Šmuc T, Stepanić V. Comprehensive machine learning based study of the chemical space of herbicides. Sci Rep. 2021;11(1):1-12.
  • Padmanabhan DC, Gowda S, Arani E, Zonooz B. LSFSL: leveraging shape information in few-shot learning. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR); 2023. p. 4970–9.
  • Pang J, Chen K, Shi J, Feng H, Ouyang W, Lin D. Libra R-CNN: Towards balanced learning for object detection. CoRR. 2019[access Mar 13, 2026]. Available from: https://arxiv.org/abs/1904.02701
    » https://arxiv.org/abs/1904.02701
  • Pantazi XE, Moshou D, Bravo C. Active learning system for weed species recognition based on hyperspectral sensing. Biosyst Eng. 2016;146:193–202. Available from: https://doi.org/10.1016/j.biosyste-mseng.2016.01.014
    » https://doi.org/10.1016/j.biosyste-mseng.2016.01.014
  • Pantazi XE, Tamouridou AA, Alexandridis TK, Lagopodi AL, Kashefi J, Moshou D. Evaluation of hierarchical self-organising maps for weed mapping using UAS multispectral imagery. Comput Electron Agric. 2017;139:224–30. Available from: https://doi.org/10.1016/j.compag.2017.05.026
    » https://doi.org/10.1016/j.compag.2017.05.026
  • Partel V, Charan Kakarla S, Ampatzidis Y. Development and evaluation of a low-cost and smart technology for precision weed management utilizing artificial intelligence. Comput Electron Agric. 2019;157:339-50. Available from: https://doi.org/10.1016/j.compag.2018.12.048
    » https://doi.org/10.1016/j.compag.2018.12.048
  • Patil SS, Patil YM, Patil SB. Development of cost-effective precision spraying techniques using sensor technology. In: 2023 2nd International Conference for Innovation in Technology (INOCON). IEEE; 2023. p. 1–4.
  • Peerbhay KY, Mutanga O, Ismail R. Random forests unsupervised classification: the detection and mapping of solanum mauritianum infestations in plantation forestry using hyperspectral data. IEEE J Sel Top Appl Earth Obs Remote Sens. 2015;8(6):3107–22. Available from: https://doi.org/10.1109/JSTARS.2015.2396577
    » https://doi.org/10.1109/JSTARS.2015.2396577
  • Peña J, Torres-Sánchez J, Serrano-Pérez A, Castro A, López-Granados F. Quantifying efficacy and limits of unmanned aerial vehicle (UAV) technology for weed seedling detection as affected by sensor resolution. Sensors. 2015;15(3):5609-26. Available from: https://doi.org/10.3390/s150305609
    » https://doi.org/10.3390/s150305609
  • Peteinatos GG, Reichel P, Karouta J, Andújar D, Gerhards R. Weed identification in maize, sunflower, and potatoes with the aid of convolutional neural networks. Remote Sens. 2020;12(24):4185. Available from: https://doi.org/10.3390/rs12244185
    » https://doi.org/10.3390/rs12244185
  • Pohl JP, Rautmann D, Nordmeyer H, von Hõrsten D. [Direct Injection in precision crop protection: site-specific application of plant protection products]. Gesunde Pflanz. 2019;71(1):51–5. [German]. Available from: https://doi.org/10.1007/s10343-019-00452-y
    » https://doi.org/10.1007/s10343-019-00452-y
  • Poulsen Engineering. Steering system Optical sensor: robovator-optical guidance. Hvals0: Poulsen Engineering; 2024[access Mar 13, 2026]. Available from: https://www.visionweeding.com/steering-system-optical-sensor/
    » https://www.visionweeding.com/steering-system-optical-sensor/
  • Powles S. Herbicide discovery through innovation and diversity. Adv Weed Sci. 2022;40(spe1):1–4. Available from: https://doi.org/10.51694/AdvWeedSci/2022;40:seventy-five012
    » https://doi.org/10.51694/AdvWeedSci/2022;40:seventy-five012
  • Projini. Current development stage: market & pipeline. Tel Aviv: Projini; 2024[access Mar 13, 2026]. Available from: https://www.projini.com/market-and-pipeline
    » https://www.projini.com/market-and-pipeline
  • Quan L, Jiang W, Li H, Li H, Wang Q, Chen L. Intelligent intra-row robotic weeding system combining deep learning technology with a targeted weeding mode. Biosyst Eng. 2022;216:13-31. Available from: https://doi.org/10.1016/j.biosystemseng.2022.01.019
    » https://doi.org/10.1016/j.biosystemseng.2022.01.019
  • Raja R, Nguyen TT, Slaughter DC, Fennimore SA. Real-time robotic weed knife control system for tomato and lettuce based on geometric appearance of plant labels. Biosyst Eng. 2020;194:152-64.
  • Raja R, Nguyen TT, Slaughter DC, Fennimore SA. Real-time weed-crop classification and localisation technique for robotic weed control in lettuce. Biosyst Eng. 2020;192:257-74. Available from: https://doi.org/10.1016/j.biosystemseng.2020.02.002
    » https://doi.org/10.1016/j.biosystemseng.2020.02.002
  • Ranjan A, Singh S, Pathak S. Precision farming: uses drones, satellite imagery, and other technologies to increase efficiency and resource utilization. Int J Recent Innov Acad Res. 2024;8(2):45-53.
  • Rasmussen J, Nielsen J, Garcia-Ruiz F, Christensen S, Streibig JC. Potential uses of small unmanned aircraft systems (UAS) in weed research. Weed Res. 2013;53(4):242–8. Available from: https://doi.org/10.1111/wre.12026
    » https://doi.org/10.1111/wre.12026
  • Reddy KN, Huang Y, Lee MA, Nandula VK, Fletcher RS, Thomson SJ et al. Glyphosate-resistant and glyphosate-susceptible Palmer amaranth (Amaranthus palmeri S. Wats.): hyperspectral reflectance properties of plants and potential for classification. Pest Manag Sci. 2014;70(12):1910–7. Available from: https://doi.org/10.1002/ps.3755
    » https://doi.org/10.1002/ps.3755
  • Reiser D, Sehsah ES, Bumann O, Morhard J, Griepentrog HW. Development of an autonomous electric robot implement for intra-row weeding in vineyards. Agriculture. 2019;9(1):1-12. Available from: https://doi.org/10.3390/agriculture9010018
    » https://doi.org/10.3390/agriculture9010018
  • Ronneberger O, Fischer P, Brox T. U-Net: convolutional networks for biomedical image segmentation. CoRR. 2015[access Mar 13, 2026]. Available from: https://arxiv.org/abs/1505.04597
    » https://arxiv.org/abs/1505.04597
  • Ruigrok T, van Henten E, Booij J, Van Boheemen K, Kootstra G. Application-specific evaluation of a weed-detection algorithm for plant-specific spraying. Sensors. 2020;20(24):7262.
  • Russakovsky O, Deng J, Su H, Krause J, Satheesh S, Ma S et al. ImageNet large scale visual recognition challenge. Int J Comput Vis. 2015;115(3):211–52. Available from: https://doi.org/10.1007/s11263-015-0816-y
    » https://doi.org/10.1007/s11263-015-0816-y
  • Saeys Y, Inza I, Larranaga P. A review of feature selection techniques in bioinformatics. Bioinformatics. 2007;23(19):2507–17. Available from: https://doi.org/10.1093/bioinformatics/btm344
    » https://doi.org/10.1093/bioinformatics/btm344
  • Sanchez J, Gallandt ER. Functionality and efficacy of Franklin Robotics’ Tertill™ robotic weeder. Weed Technol. 2021;35(1):166–70. Available from: https://doi.org/10.1017/wet.2020.94
    » https://doi.org/10.1017/wet.2020.94
  • Sapkota B, Singh V, Cope D, Valasek J, Bagavathiannan M. Mapping and estimating weeds in cotton using unmanned aerial systems-borne imagery. AgriEngineering. 2020;2(2):350–66. Available from: https://doi.org/10.3390/agriengineering2020024
    » https://doi.org/10.3390/agriengineering2020024
  • Sapkota B, Singh V, Neely C, Rajan N, Bagavathiannan M. Detection of italian ryegrass in wheat and prediction of competitive interactions using remote-sensing and machine-learning techniques. Remote Sens. 2020;12(18):1–20. Available from: https://doi.org/10.3390/rs12182977
    » https://doi.org/10.3390/rs12182977
  • Saqib MA, Aqib M, Tahir MN, Hafeez Y. Towards deep learning based smart farming for intelligent weeds management in crops. Front Plant Sci. 2023;14:1211235. Available from: https://doi.org/10.3389/fpls.2023.1211235
    » https://doi.org/10.3389/fpls.2023.1211235
  • Sarker IH. AI-based modeling: techniques, applications and research issues towards automation, intelligent and smart systems. SN Comput Sci. 2022;3(2):158. Available from: https://doi.org/10.1007/s42979-022-01043-x
    » https://doi.org/10.1007/s42979-022-01043-x
  • Scherrer B, Sheppard J, Jha P, Shaw JA. Hyperspectral imaging and neural networks to classify herbicide-resistant weeds. J Appl Remote Sens. 2019;13(4):044520. Available from: https://doi.org/10.1117/1.JRS.13.044516
    » https://doi.org/10.1117/1.JRS.13.044516
  • Shah TM, Nasika DPB, Otterpohl R. Plant and weed identifier robot as an agroecological tool using artificial neural networks for image identification. Agriculture. 2021;11(3):222. Available from: https://doi.org/10.3390/agriculture11030222
    » https://doi.org/10.3390/agriculture11030222
  • Shahamiri SR, Thabtah F, Abdelhamid N. A new classification system for autism based on machine learning of artificial intelligence. Technol Health Care. 2022;30(3):605–22. Available from: https://doi.org/10.3233/THC-213032
    » https://doi.org/10.3233/THC-213032
  • Shahinfar S, Meek P, Falzon G. “How many images do I need?” Understanding how sample size per class affects deep learning model performance metrics for balanced designs in autonomous wildlife monitoring. Ecol Inform. 2020;57:101085. Available from: https://doi.org/10.1016/j.ecoinf.2020.101085
    » https://doi.org/10.1016/j.ecoinf.2020.101085
  • Shanmugam S, Assunção E, Mesquita R, Veiros A, Gaspar PD. Automated weed detection systems: a review. KnE Eng. 2020;2020:51–64. Available from: https://doi.org/10.18502/keg.v5i6.7046
    » https://doi.org/10.18502/keg.v5i6.7046
  • Sharpe SM, Schumann AW, Boyd NS. Detection of Carolina Geranium (Geranium carolinianum) growing in competition with strawberry using convolutional neural networks. Weed Sci. 2019;67(2):239–45. Available from: https://doi.org/10.1017/wsc.2018.66
    » https://doi.org/10.1017/wsc.2018.66
  • Sheng Y, Qian W, Huang J, Wu B, Yang J, Xue T et al. Electrochemical detection combined with machine learning for intelligent sensing of maleic hydrazide by using carboxylated PEDOT modified with copper nanoparticles. Microchim Acta. 2019;186(12):1–12. Available from: https://doi.org/10.1007/s00604-019-3652-x
    » https://doi.org/10.1007/s00604-019-3652-x
  • Shorten C, Khoshgoftaar TM. A survey on image data augmentation for deep learning. J Big Data. 2019;6:60. Available from: https://doi.org/10.1186/s40537-019-0197-0
    » https://doi.org/10.1186/s40537-019-0197-0
  • Silva-Mendonça S, Vitória ARS, Lima TW, Galvão-Filho AR, Andrade CH. Exploring new horizons: empowering computer-assisted drug design with few-shot learning. Artif Intell Life Sci. 2023;4:100086. Available from: https://doi.org/10.1016/j.ailsci.2023.100086
    » https://doi.org/10.1016/j.ailsci.2023.100086
  • Silverberg D. Lasers, drones and AI: The future of weeding. London: BBC News; 2023[access Mar 13, 2026]. Available from: https://www.bbc.com/news/business-64742513
    » https://www.bbc.com/news/business-64742513
  • Simonyan K, Zisserman A. Very deep convolutional networks for large-scale image recognition. arXiv [Preprint]. 2015. Available from: https://doi.org/10.48550/arXiv.1409.1556
    » https://doi.org/10.48550/arXiv.1409.1556
  • Singh V, Rana A, Bishop M, Filippi AM, Cope D, Rajan N et al. Chapter Three – Unmanned aircraft systems for precision weed detection and management: Prospects and challenges. In: Sparks DL, editor. Advances in agronomy. Oxford: Academic Press; 2020. p. 93–134.
  • Slaughter DC, Giles DK, Downey D. Autonomous robotic weed control systems: a review. Comput Electron Agric. 2008;61(1):63–78. Available from: https://doi.org/10.1016/j.compag.2007.05.008
    » https://doi.org/10.1016/j.compag.2007.05.008
  • Small Robot Company. Small Robot Co launches Tom V4: next generation monitoring robot. London: Medium; 2023[access Mar 13, 2026]. Available from: https://medium.com/smallrobotcompany/small-robot-co-launches-tom-v4-next-generation-monitoring-robot-56b2f224d247
    » https://medium.com/smallrobotcompany/small-robot-co-launches-tom-v4-next-generation-monitoring-robot-56b2f224d247
  • Small Robot Company. Small Robot Company announce non-chemical weeding zapping with rootwave. London: Medium; 2023a[access Mar 13, 2026]. Available from: https://medium.com/smallrobotcompany/small-robot-company-announce-non-chemical-weeding-zapping-with-rootwave-499f40a9dce4
    » https://medium.com/smallrobotcompany/small-robot-company-announce-non-chemical-weeding-zapping-with-rootwave-499f40a9dce4
  • Smith MJ. Getting value from artificial intelligence in agriculture. Anim Prod Sci. 2020;60(1) :46-54. Available from: https://doi.org/10.1071/AN18522
    » https://doi.org/10.1071/AN18522
  • Snell J, Swersky K, Zemel R. Prototypical networks for few-shot learning. In: Koyejo S, Mohamed S, Agarwal A, Belgrave D, Cho K, Oh A, editors. Advances in neural information processing systems (NIPS). Red Hook: Curran Associates; 2017. p. 4077–87.
  • Sood A, Sharma RK, Bhardwaj AK. Artificial intelligence research in agriculture: a review. Online Inf Rev. 2022;46(6):1054–75. Available from: https://doi.org/10.1108/OIR-10-2020-0448
    » https://doi.org/10.1108/OIR-10-2020-0448
  • Sørensen CG, Nørremark M, Jørgensen RN, Jensen K, Jensen LA, Maagaard J. Hortibot: feasibility study of a plant nursing robot performing weeding operations – part iv. In: 2007 ASABE Annual International Meeting. St. Joseph: American Society of Agricultural and Biological Engineers; 2007.
  • Sori H, Inoue H, Hatta H, Ando Y. Effect for a paddy weeding robot in wet rice culture. J Robot Mechatronics. 2018;30(2):198–205. Available from: https://doi.org/10.20965/jrm.2018.p0198
    » https://doi.org/10.20965/jrm.2018.p0198
  • Soto M, Poncet AM, Roma-Burgos N, France OW, Velasquez JC, Ashworth AJ et al. Hyperspectral indicators and characterization of glyphosate-induced stress in common lambsquarters (Chenopodium album L.). Smart Agric Technol. 2025;11:1–14. Available from: https://doi.org/10.1016/j.atech.2025.100890
    » https://doi.org/10.1016/j.atech.2025.100890
  • Sow S, Ranjan S, Seleiman MF, Alkharabsheh HM, Kumar M, Kumar N et al. Artificial intelligence for maximizing agricultural input use efficiency: exploring nutrient, water and weed management strategies. Phyton. 2024;93(7):1569–98. Available from: https://doi.org/10.32604/phyton.2024.052241
    » https://doi.org/10.32604/phyton.2024.052241
  • Sujaritha M, Annadurai S, Satheeshkumar J, Kowshik Sharan S, Mahesh L. Weed detecting robot in sugarcane fields using fuzzy real time classifier. Comput Electron Agric. 2017;134:160–71. Available from: https://doi.org/10.1016/j.compag.2017.01.008
    » https://doi.org/10.1016/j.compag.2017.01.008
  • Sydney Informatics Hub. Weed-AI: A repository of Weed Images in Crops. Sydney: University of Sydney; 2021[access Mar 13, 2026]. Available from: https://weed-ai.sydney.edu.au/
    » https://weed-ai.sydney.edu.au/
  • Szegedy C, Ioffe S, Vanhoucke V. Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learning. arXiv. 2016. Available from: https://doi.org/10.48550/arXiv.1602.07261
    » https://doi.org/10.48550/arXiv.1602.07261
  • Szegedy C, Liu W, Jia Y, Sermanet P, Reed S, Anguelov D et al. Going deeper with convolutions. In: 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). IEEE; 2015. p. 1–9.
  • Talaviya T, Shah D, Patel N, Yagnik H, Shah M. Implementation of artificial intelligence in agriculture for optimisation of irrigation and application of pesticides and herbicides. Artif Intell Agric. 2020;4:58–73. Available from: https://doi.org/10.1016/j.aiia.2020.04.002
    » https://doi.org/10.1016/j.aiia.2020.04.002
  • Tan L, Huangfu T, Wu L, Chen W. Comparison of YOLO v3, Faster R-CNN, and SSD for Real-Time Pill Identification. BMC Med Inform Decis Mak. 2021;21:322. Available from: https://doi.org/10.21203/rs.3.rs-668895/v1
    » https://doi.org/10.21203/rs.3.rs-668895/v1
  • Tan L, Jiang J. Chapter 13 – Image Processing Basics. In: Tan L, Jiang J, editors. Digital signal processing: fundamentals and applications. 3rd ed. London: Academic Press; 2018. p. 649–726.
  • Tan PN, Steinbach M, Karpatne A, Kumar V. Introduction to data mining. 2nd ed. Upper Saddle River: Pearson Education; 2016.
  • Thanh Le VN, Truong G, Alameh K. Detecting weeds from crops under complex field environments based on Faster RCNN. In: 2020 IEEE Eighth International Conference on Communications and Electronics (ICCE). IEEE; 2021. p. 350-5.
  • Tian L, Slaughter DC, Norris RF. Outdoor field machine vision identification of tomato seedlings for automated weed control. Trans ASAE. 1997;40(6) :1761–8.
  • Tian L. Development of a sensor-based precision herbicide application system. Comput Electron Agric. 2002;36(2):133–49.
  • Tian LF, Reid JF, Hummel JW. Development of a precision sprayer for site-specific weed management. Trans ASAE. 1999;42(4):893–900.
  • Tian LF, Steward BL, Tang L. Smart sprayer project: sensor-based selective herbicide application system. In: DeShazer JA, Meyer GE, editors. Precision agriculture and biological quality. Bellingham: SPIE; 2000. p. 73–80.
  • Tillett ND, Hague T, Grundy AC, Dedousis AP. Mechanical within-row weed control for transplanted crops using computer vision. Biosyst Eng. 2008;99(2):171–8.
  • Tillett ND, Hague T, Miles SJ. Inter-row vision guidance for mechanical weed control in sugar beet. Comput Electron Agric. 2002;33(3):163–77.
  • Torres U, Johnson J, Sapkota B, Nolte S, Bagavathiannan MV. Deep learning-based classification of herbicide symptomology in cotton. In: ASA, CSSA, SSSA International Annual Meeting; 2023 Oct 29-Nov 1; St. Louis. Madison: ASA-CSSA-SSSA; 2023.
  • Tretyakova A, Grudanov N, Kondratkov P, Baranova O, Luneva N, Mysnik Y et al. A database of weed plants in the European part of Russia. Biodivers Data J. 2020;8:e56113.
  • Ukaegbu U, Mathipa L, Malapane M, Tartibu LK, Olayode IO. Deep Learning for Smart Plant Weed Applications Employing an Unmanned Aerial Vehicle. In: 2022 IEEE 13th International Conference on Mechanical and Intelligent Manufacturing Technologies (ICMIMT). IEEE; 2022. p. 321–5.
  • Underwood J, Wendel A, Schofield B, McMurray L, Kimber R. Efficient in-field plant phenomics for row-crops with an autonomous ground vehicle. J Field Robot. 2017;34(6):1061–83. Available from: https://doi.org/10.1002/rob.21728
    » https://doi.org/10.1002/rob.21728
  • Underwood JP, Calleija M, Taylor Z, Hung C, Nieto JI, Fitch RC et al. Real-time target detection and steerable spray for vegetable crops. In: 2015 IEEE International Conference on Robotics and Automation (ICRA). IEEE; 2015. p. 3106-12.
  • Upadhyay A, Zhang Y, Koparan C, Rai N, Howatt K, Bajwa S et al. Advances in ground robotic technologies for site-specific weed management in precision agriculture: a review. Comput Electron Agric. 2024;225:109363. Available from: https://doi.org/10.1016/j.compag.2024.109363
    » https://doi.org/10.1016/j.compag.2024.109363
  • Valeri JA, Collins KM, Ramesh P, Alcantar MA, Lepe BA, Lu TK et al. Sequence-to-function deep learning frameworks for engineered riboregulators. Nat Commun. 2020;11(1):5058. Available from: https://doi.org/10.1038/s41467-020-18676-2
    » https://doi.org/10.1038/s41467-020-18676-2
  • van Evert FK, Samsom J, Polder G, Vijn M, van Dooren H, Lamaker A et al. A robot to detect and control broad-leaved dock (Rumex obtusifolius L.) in grassland. J Field Robot. 2011;28(2):264–77. Available from: https://doi.org/10.1002/rob.20377
    » https://doi.org/10.1002/rob.20377
  • Vasileiou M, Kyrgiakos LS, Kleisiari C, Kleftodimos G, Vlontzos G, Belhouchette H et al. Transforming weed management in sustainable agriculture with artificial intelligence: a systematic literature review towards weed identification and deep learning. Crop Prot. 2024;176:1-21. Available from: https://doi.org/10.1016/j.cropro.2023.106522
    » https://doi.org/10.1016/j.cropro.2023.106522
  • Vinyals O, Toshev A, Bengio S, Erhan D. Show and tell: lessons learned from the 2015 MSCOCO image captioning challenge. IEEE Trans Pattern Anal Mach Intell. 2017;39(4):652-63. Available from: https://doi.org/10.1109/TPAMI.2016.2587640
    » https://doi.org/10.1109/TPAMI.2016.2587640
  • von Lilienfeld OA, Burke K. Retrospective on a decade of machine learning for chemical discovery. Nat Commun. 2020;11:4895. Available from: https://doi.org/10.1038/s41467-020-18556-9
    » https://doi.org/10.1038/s41467-020-18556-9
  • Wall-YE. Wall-YE softwares & Robots. Macon: Wall-YE; 2020[access Mar 13, 2026]. Available from: http://www.wall-ye.com/index-2.html
    » http://www.wall-ye.com/index-2.html
  • Wallisch P, Lusignan M, Benayoun M, Baker TI, Dickey AS, Hatsopoulos NG. Neural decoding part ii: continuous variables. In: Wallisch P, Lusignan M, Benayoun M, Baker TI, Dickey AS, Hatsopoulos NG, editores. MATLAB for neuroscientists. 2nd ed. London: Academic Press; 2009. p. 211–8.
  • Wang A, Zhang W, Wei X. A review on weed detection using ground-based machine vision and image processing techniques. Comput Electron Agric. 2019;158:226-40.
  • Wang K, Gou C, Duan Y, Lin Y, Zheng X, Wang FY. Generative adversarial networks: introduction and outlook. IEEE/CAA J Autom Sin. 2017;4(4):588-98.
  • Wang P, Tang Y, Luo F, Wang L, Li C, Niu Q et al. Weed25: a deep learning dataset for weed identification. Front Plant Sci. 2022;13:1053329. Available from: https://doi.org/10.3389/fpls.2022.1053329
    » https://doi.org/10.3389/fpls.2022.1053329
  • Wang Q, Cheng M, Huang S, Cai Z, Zhang J, Yuan H. A deep learning approach incorporating YOLO v5 and attention mechanisms for field real-time detection of the invasive weed Solanum rostratum Dunal seedlings. Comput Electron Agric. 2022;199. Available from: https://doi.org/10.1016/j.compag.2022.107194
    » https://doi.org/10.1016/j.compag.2022.107194
  • Wang S, Han Y, Chen J, He X, Zhang Z, Liu X et al. Weed density extraction based on few-shot learning through UAV remote sensing RGB and multispectral images in ecological irrigation area. Front Plant Sci. 2022;12. Available from: https://doi.org/10.3389/fpls.2021.735230
    » https://doi.org/10.3389/fpls.2021.735230
  • Wang Y, Yao Q, Kwok JT, Ni LM. Generalizing from a few examples: a survey on few-shot learning. ACM Comput Surv. 2020;53(3):1-34. Available from: https://doi.org/10.1145/3386252
    » https://doi.org/10.1145/3386252
  • Wolfert S, Ge L, Verdouw C, Bogaardt MJ. Big data in smart farming: a review. Agric Syst. 2017;153:69-80. Available from: https://doi.org/10.1016/j.agsy.2017.01.023
    » https://doi.org/10.1016/j.agsy.2017.01.023
  • Woo S, Park J, Lee JY, Kweon IS. CBAM: Convolutional block attention module. In: Ferrari V, Hebert M, Sminchisescu C, Weiss Y, editors. Computer Vision – ECCV 2018. Cham: Springer; 2018. p. 3-19. (Lecture Notes in Computer Science; vol. 11211).
  • Wu P, Zhao H. Some analysis and research of the AdaBoost algorithm. In: International Conference on Intelligent Computing and Information Science. Berlin: Springer; 2011. p. 1-5.
  • Wu S, Zhong S, Liu Y. Deep residual learning for image steganalysis. Multimed Tools Appl. 2018;77(9):10437-53. Available from: https://doi.org/10.1007/s11042-017-4440-4
    » https://doi.org/10.1007/s11042-017-4440-4
  • Wu X, Aravecchia S, Lottes P, Stachniss C, Pradalier C. Robotic weed control using automated weed and crop classification. J Field Robot. 2020;37(2):322-40. Available from: https://doi.org/10.1002/rob.21938
    » https://doi.org/10.1002/rob.21938
  • Xia F, Quan L, Lou Z, Sun D, Li H, Lv X. Identification and comprehensive evaluation of resistant weeds using unmanned aerial vehicle-based multispectral imagery. Front Plant Sci. 2022;13:1-15. Available from: https://doi.org/10.3389/fpls.2022.938604
    » https://doi.org/10.3389/fpls.2022.938604
  • Xiaobo Z, Jiewen Z, Povey MJW, Holmes M, Hanpin M. Variables selection methods in near-infrared spectroscopy. Anal Chim Acta. 2010;667(1-2):14-32. Available from: https://doi.org/10.1016/j.aca.2010.03.048
    » https://doi.org/10.1016/j.aca.2010.03.048
  • Xiong Y, Ge Y, Liang Y, Blackmore S. Development of a prototype robot and fast path-planning algorithm for static laser weeding. Comput Electron Agric. 2017;142:494-503. Available from: https://doi.org/10.1016/j.compag.2017.11.023
    » https://doi.org/10.1016/j.compag.2017.11.023
  • Yamada Y, Iwamura M, Kise K. Deep pyramidal residual networks with separated stochastic depth. arXiv. 2016. Available from: https://doi.org/10.48550/arXiv.1612.01230
    » https://doi.org/10.48550/arXiv.1612.01230
  • Yanofsky NS. Towards a definition of an algorithm. J Log Comput. 2011;21(2):253-86. Available from: https://doi.org/10.1093/log-com/exq016
    » https://doi.org/10.1093/log-com/exq016
  • Zagoruyko S, Komodakis N. Wide residual networks. arXiv. 2016. Available from: https://doi.org/10.48550/arXiv.1605.07146
    » https://doi.org/10.48550/arXiv.1605.07146
  • Zhang T, Huang Y, Reddy KN, Yang P, Zhao X, Zhang J. Using machine learning and hyperspectral images to assess damages to corn plant caused by glyphosate and to evaluate recoverability. Agronomy. 2021;11(3):1–17. Available from: https://doi.org/10.3390/agronomy11030583
    » https://doi.org/10.3390/agronomy11030583
  • Zhao F, Guo Y, Huang Y, Reddy KN, Zhao Y, Molin WT. Detection of the onset of glyphosate-induced soybean plant injury through chlorophyll fluorescence signal extraction and measurement. J Appl Remote Sens. 2015;9(1). Available from: https://doi.org/10.1117/1.JRS.9.097098
    » https://doi.org/10.1117/1.JRS.9.097098
  • Zhou H, Zhuang Z, Liu Y, Liu Y, Zhang X. Defect classification of green plums based on deep learning. Sensors. 2020;20(23):1-15. Available from: https://doi.org/10.3390/s20236993
    » https://doi.org/10.3390/s20236993
  • Zhuang J, Li X, Bagavathiannan M, Jin X, Yang J, Meng W et al. Evaluation of different deep convolutional neural networks for detection of broadleaf weed seedlings in wheat. Pest Manag Sci. 2022;78(2) :521-9. Available from: https://doi.org/10.1002/ps.6656
    » https://doi.org/10.1002/ps.6656

* Corresponding author:

<luis.avila@msstate.edu>

Editor in Chief:

Carol Ann Mallory-Smith

Associate Editor:

Aldo Merotto Junior

Conflict of Interest:

The authors declare that there is no conflict of interest regarding the publication of this manuscript.

Publication Dates

  • Publication in this collection
    24 Aug 2026
  • Date of issue
    2026

History

  • Received
    11 July 2024
  • Accepted
    19 Feb 2026
location_on
Sociedade Brasileira da Ciência das Plantas Daninhas - SBCPD Rua Santa Catarina, 50, sala 1302 , 86010-470 - Londrina - Paraná / Brasil , +55 (51) 3308-6006 - Londrina - PR - Brazil
E-mail: sbcpd@sbcpd.org
rss_feed Stay informed of issues for this journal through your RSS reader
Go to top