Near-infrared spectroscopy combined with effective variable selection algorithm for rapid detection of rice taste quality

Abstract

Rice taste quality determines the satisfaction of consumers and whether it can be profitable for growers. The accurate and fast determination of the taste quality of rice helps in evaluating the commodity price of rice. In this study, partial least squares regression (PLSR) and near-infrared spectroscopy (NIR) were combined to predict changes in rice taste quality. The coefficient determination of cross-validation (R2CV), the coefficient determination of prediction (R2P), and the residual predictive deviation (RPD) of the PLSR model of the Savitzky-Golay first derivative spectrum were 0.84, 0.86, and 2.76, respectively. Three methods were applied to screen the valid wavelengths, including competitive adaptive reweighted sampling (CARS), successive projections algorithm (SPA), and random frog (RF). The RF algorithm screened 10 effective wavelengths and had the best prediction ability. After mining the effective wavelength by RF, the R2CV, R2P, and RPD were 0.89, 0.93, and 3.8, respectively. This study provided a rapid and accurate method for determining the taste quality of large-scale rice samples.