A comprehensive analysis of NIR and MIR data fusion for rice authentication

This study investigated the potential of mid-infrared (MIR) and near-infrared (NIR) spectroscopy, combined with chemometric modelling, for the authentication of jasmine rice and the detection of adulteration at different levels. Spectral data were analysed both individually and through a multiblock data fusion strategy using Sequential Preprocessing through Orthogonalization (SPORT). Partial Least Squares–Discriminant Analysis (PLS-DA) models were first developed separately on MIR and NIR datasets using several preprocessing techniques (SNV, first derivative, and second derivative). When analysed individually, MIR showed better predictive performance than NIR in terms of classification accuracy. The SPORT-based multiblock approach, integrating complementary MIR and NIR information, improved the exploitation of spectral variability and provided a more robust framework for modelling complex authenticity patterns. When applied to an independent test set, the model provided satisfactory performance, achieving classification rates of ~71% for authentic rice and 66.7% and 81.8% for the 5% and 10% adulteration levels. Classification errors mainly involved samples at one adulteration level being assigned to the other adulterated class, reflecting the high chemical and spectral similarity between samples containing 5% and 10% of the same adulterant, while discrimination between authentic and adulterated rice remained robust.