Single-kernel NIR spectroscopy for non-destructive rice bran color discrimination across diverse hull types and production environments

Uniformity of rice bran color is important in the whole grain rice market as well as in seed rice production. Normally, determining bran color requires the removal of the outer hull, which is a destructive process, time-consuming, and an obstacle to the production of nutritious pigmented bran varieties. In this study, single-kernel near-infrared (SKNIR) spectroscopy (905-1688 nm) and multivariate techniques were explored to discriminate rough rice based on bran color (brown, purple, red bran) in rice varieties with similar hull colors, different hull colors, and different growing environments in the US. Classification models were developed using partial least squares discriminant analysis (PLS-DA) and quadratic discriminant analysis (QDA) in combination with variable selection techniques. The results revealed that brown bran is the easiest to identify among the three bran colors. In rough rice varieties with similar straw hulls and different growing environments, the lowest precision was found in red bran. The models from QDA, which performed better than PLS-DA, achieved a recall (true positive rate) of 90-100% and a maximum false positive rate of 7%. These results demonstrate the potential of SKNIR for rapid and non-destructive separation of rough rice according to bran color, facilitating sorting and quality control applications for breeding and processing of pigmented rice varieties.