A fundamental step in image processing for machine vision is the scene segmentation and one of the most used methods is the threshold, especially when one desires to cluster the pixels in two classes. The image threshold affects the number of pixels in each class and the object shape and dimension. An automatic thresholding method not only could avoid operator's influence but also could speed up the threshold value definition in the field where light variation influences the pixels values. In this work, two automatic thresholding methods were evaluated for use in a machine vision system that identifies damaged corn plants by the armyworms. The images used were of damaged and not damaged corn plants in three stages. Three groups of 10 plants were collected to take the images of each group under different light intensities. The images processed with the excess green index were thresholded with the automatic methods and compared with the manual thresholding methods. The results of both automatic methods were good, with a mean accuracy higher than 99%, showing the potential of the system for identifying the damaged corn plants.
image processing; machine vision; precision farming
