Combination of near-infrared spectroscopy and key wavelength-based screening algorithm for rapid determination of rice protein content

Abstract

Rice taste quality is one of the most important factors influencing rice marketing and distribution, and rice with a high taste quality is more popular with consumers and has a higher price. Accurate and rapid determination of rice protein content helps to assess the rice taste quality and aids in marketing. In this study, NIR spectra of 84 rice samples combined with partial least squares regression (PLSR) were used to model protein content, and different selection algorithms based on key wavelengths (competitive adaptive reweighted sampling, CARS; Monte-Carlo uninformative wavelength elimination, MC-UVE; random frog, RF) were used to understand the accuracy of NIR spectra in predicting the rice protein content. Our results showed that the R2P and RPD of the original full-spectrum PLSR model were 0.83 and 1.95, respectively. After the second-order derivative preprocessing, the R2P and RPD of the full spectrum were improved to 0.95 and 4.14. Both CARS and MC-UVE increased the R2P of the PLSR model to 0.97 and the RPD to 5.57 and 5.65, respectively. R2C and R2CV in the PLSR model based on CARS algorithm were 0.93 and 0.91, respectively. The CARS algorithm had excellent results in predicting the rice protein content.