Comparative assessment of feature-wavelength eligibility for measurement of water binding capacity and specific gravity of tuber using diverse spectral indices stemmed from hyperspectral images

A number of investigations have been conducted to explore non-contacting measuring techniques for predicting chemical components of agricultural product, but there is no research on non-destructive inspection and quantification of water binding capacity (WBC) and specific gravity (SG) in tuber. The purpose of this study was to exploit a rapid analytical technique using reflectance spectra (RS), generalised logarithm spectra (GLS), absorbance spectra (AS), and power spectra (PS) derived from spectral image data to develop partial least squares regression (PLSR) and locally weighted principal component regression (LWPCR) models that predicted tuber WBC and SG. Based upon the RS, GLS, AS, and PS, corresponding feature wavelengths were then respectively selected by using genetic algorithm (GA), first-derivative and mean centering iteration algorithm (FMCIA), and reverse variable algorithm (RVA). Compared to FMCIA and GA, the method of RVA achieved the highest accuracy based on the RVA-PS-LWPCR model for predicting WBC and SG. Then, all combinations of feature wavelengths were refined with the method of regression coefficient (RC). The simplified GA-RC-PS-LWPCR model obtained highest accuracy to measure WBC, resulting in a coefficient of determination in prediction (R2P) of 0.966 and root mean square error of prediction (RMSEP) of 0.199. Besides, the FMCIA-RC-GLS-LWPCR model showed the best performance to determine SG, with R2P of 0.978 and RMSEP of 0.009. The optimal models were then applied to each pixel of the spectral image to generate distribution maps of WBC and SG of tested samples. Furthermore, the overall performances of wavelength selection methods in terms of FMCIA-RC and RVA-RC were equivalent and slightly better than GA-RC. The results demonstrated that effective wavelength selection method can improve the performance of multispectral imaging system for detection of WBC and SG.