Proximal hyperspectral analysis in grape leaves for region and variety identification

ABSTRACT: Reflectance measurements of plants the same species can produce sets data with differences between spectra, due to factors that be external plant, like environment where plant grows, and internal factors, for different varieties. This paper reports results analysis radiometric performed on leaves vines several grape varieties sites. The objective research was, after application techniques dimensionality reduction definition most relevant wavelengths, evaluate four machine learning models applied observational sample aiming discriminate classes region variety in vineyards. tested classification were Canonical Discrimination Analysis (CDA), Light Gradient Boosting Machine (LGBM), Random Forest (RF), Support Vector (SVM). From results, we reported LGBM model obtained better accuracy spectral discrimination by region, a value 0.93, followed RF model. Regarding varieties, these two also achieved accuracies 0.88 0.89. wavelengths more at ultraviolet, those blue green regions. pointed toward importance defining characterization reflectance spectra revealed effective capability discriminating vineyards their or variety, using models.