Predição do conteúdo relativo de clorofila na copa da planta de chá usando algoritmo GRNN otimizado e imagens multiespectrais RPA

Zhou, Qingyan; Zhang, Jincheng; Wei, Tangwei; Xing, Wen; Wang, Jing; Zhang, Youhua

RESUMO

Para avaliar de forma rápida e precisa o crescimento de plantas de chá, este estudo visa encontrar uma nova maneira de prever o conteúdo de clorofila em dosséis de plantas de chá usando aprendizado de máquina. Utilizando aeronaves pilotadas remotamente equipadas com câmeras multiespectrais, imagens de áreas de plantações de chá são capturadas e a reflectância de quatro bandas espectrais é extraída, levando ao cálculo de índices de vegetação. Simultaneamente, o conteúdo relativo de clorofila nos dosséis das plantas de chá foi coletado no terreno usando um detector. Quatro modelos, nomeadamente Floresta Aleatória (RF), rede neural de retropropagação (BPNN), rede de função de base radial (RBFN) e Rede Neural de Regressão Geral (GRNN), foram construídos para prever o conteúdo relativo de clorofila nos dosséis das plantas de chá. Subsequentemente, variáveis importantes de sensoriamento remoto foram identificadas através de filtragem RF, seguidas por uma comparação do desempenho preditivo dos modelos de aprendizado de máquina sob diferentes condições de entrada. Por último, ao integrar o Algoritmo de Busca de Pardal (SSA) para otimizar o fator de suavização no GRNN, o estudo explora o impacto dos algoritmos de otimização no desempenho preditivo do modelo GRNN. Experimentos indicam que, dentro dos modelos de aprendizado de máquina estabelecidos, o GRNN demonstra a maior precisão preditiva. Ao classificar a importância das variáveis de sensoriamento remoto através de RF, 18 variáveis significativas de sensoriamento remoto foram selecionadas, o que melhorou a precisão preditiva dos modelos de aprendizado de máquina. A otimização do fator de suavização do GRNN através do algoritmo SSA pode melhorar significativamente a precisão preditiva do modelo GRNN. Com base em uma série de experimentos, o modelo de predição estabelecido RFSSA-GRNN demonstra um bom desempenho preditivo, com um alcançando 0,84.

Termos para indexação:
Parâmetros fisiológicos da folha; índice de vegetação; triagem de recursos; algoritmo de otimização inteligente.

[1] Editor de seção:

Renato Paiva

Vegetation Index	Formula	Source
Chlorophyll Index-green (CI-GREEN)	$\frac{N I R}{G} - 1$	(Gitelson, Gritz, & Merzlyak, 2003)
Chlorophyll Index-red (CI-RED)	$\frac{N I R}{R} - 1$	(Clevers, Kooistra, & van den Brande, 2017Clevers, J., Kooistra, L., & Van Den Brande, M. (2017). Using sentinel-2 data for retrieving lai and leaf and canopy chlorophyll content of a potato crop. Remote Sensing, 9(5):405.)
Chlorophyll Index-red edge (CI-REDEDGE)	$\frac{N I R}{R E} - 1$	(Gitelson, Gritz, & Merzlyak, 2003Gitelson, A. A., Gritz, Y., & Merzlyak, M. N. (2003). Relationships between leaf chlorophyll content and spectral reflectance and algorithms for non-destructive chlorophyll assessment in higher plant leaves. Journal of Plant Physiology, 160(3):271-282.)
Content Validity Index (CVI)	$\frac{N I R}{G} \times \frac{R}{G}$	(Vincini, Frazzi, & D’Alessio, 2008Vincini, M., Frazzi, E., & D’alessio, P. (2008). A broad-band leaf chlorophyll vegetation index at the canopy scale. Precision Agriculture, 9(5):303-319.)
Dynamic Valgus (DVI)	$N I R - R$	(Rouse et al., 1974Rouse, J. W. et al. (1974). Monitoring vegetation systems in the great plains with erts. NASA. Goddard Space Flight Center 3d ERTS-1 Symptoms, 351(1):309.)
DVI-REG	$N I R - R E$	(Rouse et al., 1974Rouse, J. W. et al. (1974). Monitoring vegetation systems in the great plains with erts. NASA. Goddard Space Flight Center 3d ERTS-1 Symptoms, 351(1):309.)
Enhanced Vegetation Index 2 (EVI2)	$2.5 \times \frac{(N I R - R)}{(N I R + 2.4 R + 1)}$	(Jiang et al., 2008Jiang, Z. et al. (2008). Development of a two-band enhanced vegetation index without a blue band. Remote Sensing of Environment, 112(10):3833-3845.)
Enhanced Vegetation Index2 -2 (EVI2-2)	$2.4 \times \frac{(N I R - R)}{(N I R + R + 1)}$	(Jiang et al., 2008Jiang, Z. et al. (2008). Development of a two-band enhanced vegetation index without a blue band. Remote Sensing of Environment, 112(10):3833-3845.)
Green Normalized Difference Vegetation Index (GNDVI)	$\frac{(N I R - G)}{(N I R + G)}$	(Gitelson, Gritz, & Merzlyak, 2003Gitelson, A. A., Gritz, Y., & Merzlyak, M. N. (2003). Relationships between leaf chlorophyll content and spectral reflectance and algorithms for non-destructive chlorophyll assessment in higher plant leaves. Journal of Plant Physiology, 160(3):271-282.)
Green Optimized Soil Adjusted Vegetation Index (GOSAVI)	$1.16 \times \frac{(N I R - G)}{(N I R + G + 0.16)}$	(Liu, Pattey, & Jego, 2012Liu, J., Pattey, E., & Jego, G. (2012). Assessment of vegetation indices for regional crop green lai estimation from landsat images over multiple growing seasons. Remote Sensing of Environment, 123:347-358.)
Green Red Vegetation Index (GRVI)	$\frac{(G - R)}{(G + R)}$	(Zhang et al., 2018Zhang, L. G. et al. (2018). Density weighted connectivity of grass pixels in image frames for biomass estimation. Expert Systems With Applications, 101:213-227.)
Leaf Chlorophyll Index (LCI)	$\frac{(N I R - R E)}{(N I R - R)}$	(Xiao et al., 2014Xiao, Y. F. et al. (2014). Sensitivity analysis of vegetation reflectance to biochemical and biophysical variables at leaf, canopy, and regional scales. IEEE Transactions on Geoscience and Remote Sensing, 52(7):4014-4024.)
Modified Chlorophyll Absorption in Reflectance Index (MCARI)	$(R E - R) - 0.2 \times (R E - G) \times \frac{R E}{R}$	(Daughtry et al., 2000Daughtry, C. S. T. et al. (2000). Estimating corn leaf chlorophyll concentration from leaf and canopy reflectance. Remote Sensing of Environment, 74(2):229-239.)
Modified Chlorophyll Absorption in Reflectance Index 1 (MCARI1)	$1.2 \times [2.5 (N I R - R) - 1.3 (N I R - G)]$	(Haboudane et al., 2004Haboudane, D. et al. (2004). Hyperspectral vegetation indices and novel algorithms for predicting green lai of crop canopies: Modeling and validation in the context of precision agriculture. Remote Sensing of Environment, 90(3):337-352.)
Modified Chlorophyll Absorption in Reflectance Index 2(MCARI2)	$\frac{(3.75 (N I R - R) - 1.95 (N I R - G))}{\sqrt{{(2 N I R + 1)}^{2} - 6 (N I R - 5 \sqrt{R}) - 0.5}}$	(Haboudane et al., 2004Haboudane, D. et al. (2004). Hyperspectral vegetation indices and novel algorithms for predicting green lai of crop canopies: Modeling and validation in the context of precision agriculture. Remote Sensing of Environment, 90(3):337-352.)
Modified Non-Linear Index (MNLI)	$\frac{(1.5 N I R^{2} - 1.5 G)}{(N I R^{2} + R + 0.5)}$	(Gong et al., 2003Gong, P. et al. (2003). Estimation of forest leaf area index using vegetation indices derived from hyperion hyperspectral data. IEEE Transactions on Geoscience and Remote Sensing, 41(6):1355-1362.)
Modified Simple Ratio (MSR)	$\frac{(N I R / R - 1)}{\sqrt{N I R / R + 1}}$	(Chen, 1996Chen, J. M. (1996). Evaluation of vegetation indices and a modified simple ratio for boreal applications. Canadian Journal of Remote Sensing, 22(3):229-242.)
Modified Simple Ratio-Rededge (MSR-REDEDGE)	$\frac{(N I R / R E - 1)}{\sqrt{N I R / R E + 1}}$	(Chen, 1996Chen, J. M. (1996). Evaluation of vegetation indices and a modified simple ratio for boreal applications. Canadian Journal of Remote Sensing, 22(3):229-242.)
Normalized Difference Red Edge (NDRE)	$\frac{(N I R - R E)}{(N I R + R E)}$	(Gitelson, & Merzlyak, 1997Gitelson, A. A., Merzlyak, M. N. (1997). Remote estimation of chlorophyll content in higher plant leaves. International Journal of Remote Sensing, 18(12):2691-2697.)
Normalized Difference Red/Green Redness Index (NDREI)	$\frac{(R E - G)}{(R E + G)}$	(Hassan et al., 2018Hassan, M. A. et al. (2018). Time-series multispectral indices from unmanned aerial vehicle imagery reveal senescence rate in bread wheat. Remote Sensing, 10(6):809.)
Normalized Area Vegetation Index (NAVI)	$1 - \frac{R}{N I R}$	(Carmona, Rivas, & Fonnegra, 2015Carmona, F., Rivas, R., & Fonnegra, D. C. (2015). Vegetation index to estimate chlorophyll content from multispectral remote sensing data. European Journal of Remote Sensing, 48(1):319-326.)
Normalized Difference Vegetation Index (NDVI)	$\frac{(N I R - R)}{(N I R + R)}$	(Rouse et al., 1974Rouse, J. W. et al. (1974). Monitoring vegetation systems in the great plains with erts. NASA. Goddard Space Flight Center 3d ERTS-1 Symptoms, 351(1):309.)
Optimized Soil Adjusted Vegetation Red Index (OSAVI-RED))	$1.6 \times \frac{(N I R - G)}{(N I R + R + 0.16)}$	(Rondeaux, Steven, & Baret, 1996Rondeaux, G., Steven, M., & Baret, F. (1996). Optimization of soil-adjusted vegetation indices. Remote Sensing of Environment, 55(2):95-107.)
Optimized Soil Adjusted Vegetation Green Index (OSAVI-GREE)	$1.6 \times \frac{(N I R - G)}{(N I R + G + 0.16)}$	(Rondeaux, Steven, & Baret, 1996Rondeaux, G., Steven, M., & Baret, F. (1996). Optimization of soil-adjusted vegetation indices. Remote Sensing of Environment, 55(2):95-107.)
Optimized Soil Adjusted Vegetation Rededge Index (OSAVI-RE)	$1.6 \times \frac{(N I R - R E)}{(N I R + R E + 0.16)}$	(Rondeaux, Steven, & Baret, 1996Rondeaux, G., Steven, M., & Baret, F. (1996). Optimization of soil-adjusted vegetation indices. Remote Sensing of Environment, 55(2):95-107.)
Renormalized Difference Vegetation Index (RDVI)	$\frac{(N I R - R)}{\sqrt{N I R + R}}$	(Roujean, & Breon, 1995Roujean, J.-L., & Breon, F.-M. (1995). Estimating par absorbed by vegetation from bidirectional reflectance measurements. Remote Sensing of Environment, 51(3):375-384.)
Renormalized Difference Vegetation red Index (RDVI-REG)	$\frac{(N I R - R E)}{\sqrt{N I R + R E}}$	(Roujean, & Breon, 1995Roujean, J.-L., & Breon, F.-M. (1995). Estimating par absorbed by vegetation from bidirectional reflectance measurements. Remote Sensing of Environment, 51(3):375-384.)
Red-Edge Triangulated Vegetation Index (RTVI-Core)	$100 \times (N I R - R E) - 10 \times (N I R - G)$	(Walsh et al., 2018Walsh, O. S. et al. (2018). Assessment of uav based vegetation indices for nitrogen concentration estimation in spring wheat. Advances in Remote Sensing, 7(2):19.)
Ratio Vegetation Index (RVI)	$\frac{N I R}{R}$	(Rouse et al., 1974Rouse, J. W. et al. (1974). Monitoring vegetation systems in the great plains with erts. NASA. Goddard Space Flight Center 3d ERTS-1 Symptoms, 351(1):309.)
Soil-Adjusted Vegetation Index (SAVI)	$1.5 \times \frac{(N I R - R)}{(N I R + R + 0.5)}$	(Huete, 1988Huete, A. R. (1988). A soil-adjusted vegetation index savi. Remote Sensing of Environment, 25(3):295-310.)
Soil-Adjusted Vegetation green Index (SAVI-GREEN)	$1.5 \times \frac{(N I R - G)}{(N I R + G + 0.5)}$	(Verrelst et al., 2008Verrelst, J. et al. (2008). Angular sensitivity analysis of vegetation indices derived from chris/proba data. Remote Sensing of Environment, 112(5):2341-2353.)
Simplified Canopy Chlorophyll Content Index (S-CCCI)	$\frac{N D R E}{N D V I}$	(Raper, & Varco, 2015Raper, T. B., & Varco, J. J. (2015). Canopy-scale wavelength and vegetative index sensitivities to cotton growth parameters and nitrogen status. Precision Agriculture, 16(1):62-76.)
Simple Ratio Index (SR-REG)	$\frac{N I R}{R E}$	(Walsh et al., 2018Walsh, O. S. et al. (2018). Assessment of uav based vegetation indices for nitrogen concentration estimation in spring wheat. Advances in Remote Sensing, 7(2):19.)
Transformed Chlorophyll Absorption in Reflectance Index (TCARI)	$3 \times [(R E - R) - 0.2 (R E - G) \times (\frac{R E}{R})]$	(Haboudane et al., 2002Haboudane, D. et al. (2002). Integrated narrow-band vegetation indices for prediction of crop chlorophyll content for application to precision agriculture. Remote Sensing of Environment, 81(2-3):416-426.)
TCARI/OSAVI	$\frac{T C A R I}{O S A V I}$	(Haboudane et al., 2002Haboudane, D. et al. (2002). Integrated narrow-band vegetation indices for prediction of crop chlorophyll content for application to precision agriculture. Remote Sensing of Environment, 81(2-3):416-426.)
Triangular Vegetation Index (TVI)	$\frac{[120 (N I R - G) - 200 (R - G)]}{2}$	(Broge, & Leblanc, 2001Broge, N. H., Leblanc, E. (2001). Comparing prediction power and stability of broadband and hyperspectral vegetation indices for estimation of green leaf area index and canopy chlorophyll density. Remote Sensing of Environment, 76(2):156-172.)
Wide Dynamic Range Vegetation Index (WDRVI)	$\frac{(0.2 N I R - R)}{(0.2 N I R + R)}$	(Gitelson, 2013Gitelson, A. A. (2013). Remote estimation of crop fractional vegetation cover: The use of noise equivalent as an indicator of performance of vegetation indices. International Journal of Remote Sensing, 34(17):6054-6066.)

Data Type	Sample size	Minimum value	Maximum value	Average value	Standard deviation	Coefficient of variation(%)
Training set	80	27.14	49.9	37.08	6.05	16.30
Test set	20	27.46	48.6	37.39	6.86	18.35
All	100	27.14	49.9	37.03	6.14	16.6

Model selection		Parameter settings	R²	MSE	RMSE	MAE
RF		Bootstrap = True Maximum depth = 9 Number of trees =100	0.53	21.8	4.67	3.76
Enter all remote sensing variables	BPNN	Hidden layer dimension =14 Optimizers=Adam Activation functions = Sigmoid Early Stop Method=True Epochs=5000	0.54	21.41	4.63	3.46
	RBFN	Hidden layer dimension =10 Optimizers=Adam Early Stop Method=True Epochs=5000	0.51	22.48	4.74	3.68
	GRNN	Smoothing factor=0.7	0.59	18.92	4.34	3.72
	SSA-GRNN	Number of individuals = 30 Number of iterations = 30 Dimensionality = 1 Lower bound = 0.01 Upper bound = 2	0.61	18.62	4.31	2.57
Enter important remote sensing variables	BPNN	Hidden layer dimension =12 Optimizers=Adam Activation functions = Sigmoid Early Stop Method=True Epochs=5000	0.58	19.26	4.39	3.73
	RBFN	Hidden layer dimension =6 Optimizers=Adam Early Stop Method=True Epochs=5000	0.66	15.45	3.93	3.27
	GRNN	Smoothing factor=0.55	0.77	10.11	3.18	2.42
	SSA-GRNN	Number of individuals = 30 Number of iterations = 30 Dimensionality = 1 Lower bound = 0.01 Upper bound = 2	0.84	7.04	2.65	1.92

Brasil

Brasil

Predição do conteúdo relativo de clorofila na copa da planta de chá usando algoritmo GRNN otimizado e imagens multiespectrais RPA

RESUMO