Classificação individual de sementes de mamona usando espectroscopia de reflectância no visível, imagens digitais e análises multivariadas

Abstract

This work presents two methods based on digital image analysis, diffuse reflectance spectroscopy in the visible espectral range and multivariate analysis to classify seeds castor with respect to the type of cultivar and genotype. For this purpose, two groups of seeds commonly used in Brazilian plantations were evaluated: BRS Nordestina e BRS Paraguaçu (group I), BRS Energia cultivar and CNPA 2009-7 (group II) The diffuse reflectance spectra were recorded in the region 400-800 nm obtained by a spectrophotometer VIS-NIR. The Images of these two groups were registered from a webcam and frequency distribution of color indixes in the red-green-blue channels, hue, saturation, intensity and grayscale was obtained. The discriminant analysis by partial least squares (PLS-DA) and linear discriminant analysis were applied separately for each group of seed. The best results were obtained using the PLS-DA model correctly classified samples to visible 96.2% and 92.5% of test samples for the group I and II, respectively. For digital images PLS-DA also achieved the best result hitting 98.7% and 100% for group I and II, respectively. The methods here proposed based on digital image analysis and diffuse reflectance spectroscopy in the visible range has advantages for not using reagents, are fast, inexpensive and are promising in the classification of castor alternatives, according to the type of cultivar and genotypes