Prediction of banana color and firmness using a novel wavelengths selection method of hyperspectral imaging

This study investigated the feasibility of using hyperspectral imaging for determining banana color (L∗, a∗ and b∗) and firmness as well as classifying ripe and unripe samples. The hyperspectral images at wavelengths 380-1023 nm were acquired. Partial least squares (PLS) models were built to predict color and firmness. Two-wavelength combination method λi-λjλi+λj,λi2-λj2λi2+λj2,λiλjandλi-λj was used to identify the effective wavelengths. Based on the selected wavelengths, PLS models obtained good results with the coefficient of determination in prediction (Rp2) of 0.795 for L∗, 0.972 for a∗, 0.773 for b∗ and 0.760 for firmness. The corresponding residual predictive deviation (RPD) values were 2.234, 6.098, 2.119 and 2.062, respectively. The classification results of ripe and unripe samples were excellent in two different principal components spaces (PC1 + PC2 and PC1 + PC3). It indicated hyperspectral imaging can be used to non-destructively determine banana color and firmness as well as classify ripe and unripe samples.