Use of morphological and chemical features from near infrared-hyperspectral imaging (NIR-HSI) for advanced classification of rice varieties: A chemometric approach

Rice is one of the world's most important staple crops, but its large varietal diversity makes authenticity control and fraud prevention difficult. This study investigated whether near-infrared hyperspectral imaging (NIR-HSI) combined with chemometrics could support rice variety discrimination through morphological and chemical information. A total of 47 rice varieties, 36 of which Italian and 11 from different foreign countries, were analysed, with 115 grains per variety, and the resulting image data were explored by principal component analysis and classified using partial least squares discriminant analysis (PLS-DA) and hierarchical modelling. Morphological features gave the best performance: the five grain-shape classes were classified with 94.1 % NER and up to 96.0 % class accuracy, while hierarchical morphology-based models further improved discrimination, separating 11–12 groups with accuracies above 93 % and reaching 99.6 % in the best case. NIR spectra were less informative for variety discrimination: amylose-based PLS-DA achieved 86.7 % NER and 91.0 % accuracy, whereas hierarchical spectral models showed high specificity but low sensitivities for several classes. Data fusion of morphological and spectral variables did not improve the predictive performance beyond morphology alone, suggesting limited complementarity between the two information sources. Overall, the results demonstrate that NIR-hyperspectral imaging is a rapid, non-destructive, and promising approach for rice authentication, with morphological descriptors providing the most robust basis for classification. These findings support the use of chemometric imaging strategies to strengthen quality control and help prevent food fraud in the rice supply chain.