ABSTRACT
Breeding of high-quality rice requires quick methods to evaluate the quality characteristics such as milling, grain appearance, nutritional, eating, and cooking qualities. Because routine measurements of these quality traits are time consuming and expensive, a rapid predictive method based on near-infrared spectroscopy (NIRS) can be applied to measure these quality parameters. In this study, calibration models for measurement of grain quality were developed using a total of 570 brown and milled rice samples. The results indicated that the models developed from the spectra of brown rice for all the quality traits had the coefficient of determination for external validation (R2) larger than 0.64 except for gel consistency. The best model was developed for the protein content, with R2 of 0.94 for external validation. The model for the total score of physicochemical characteristics (TSPC), a comprehensive index reflecting all other traits, had R2 of 0.70 and SD/SEP of 1.70, which indicates that high or low TSPC for a given rice could be discriminated by NIRS. The models developed from brown rice were as accurate as those from milled rice. Results suggest that NIRS-based predictions for rice quality traits may be used as indicator traits to improve rice quality in breeding programs.