Efficient protein quantification method by near-infrared spectroscopy for multi-particle size rice samples

Protein, as a core parameter for assessing the rice nutritional quality, shows a close correlation between its content and cooking flavor characteristics. Aiming at complicated operation process, high time-cost and high reagent consumption of traditional chemical methods for detecting protein content, this study investigated the rapid detection method for protein content in milled rice with different particle sizes of pulverization (0.5 mm, 1.0 mm, 1.5 mm, and whole particle) using near-infrared spectroscopy. By combining wavelength selections with multivariate quantitative corrections, the optimal construction method of the spectral regression model for rice protein content with different particle sizes was explored. Following comparative analyses, it was determined that the 1.0 mm particle size model constructed by the uninformative variable elimination combined with the deep extreme learning mechanism exhibited optimal performance. Coefficients of determination for the validation and external test sets were found to be 0.9697 and 0.9641, with relative root mean square errors of 2.50% and 3.01%, and residual prediction deviations of 5.8383 and 5.4370, respectively. This method not only provides a reliable nonlinear modeling method for the analysis of grain quality, but also provides an innovative solution for the adaptation of multi-scale data and efficient algorithms in the spectral quantitative analysis of agricultural products. • Spectral model of intact rice particle fulfills rapid protein testing requirements. • Optimal rice particle size of 1.0 mm balances detection accuracy and milling costs. • Feature wavelengths coupled with multiple models achieve precise protein detection • UVE-DELM combined with 1.0 mm particle size achieves optimal detection accuracy.