Multivariate curve resolution followed by partial least squares-discriminant analysis combined with Vis-NIR hyperspectral imaging for rice authentication

Rice serves as a staple food for nearly half the global population, especially in Asia, where it is a major agricultural commodity. Nonetheless, deceptive practices, like blending premium and inferior rice varieties and selling them at inflated prices, present a considerable challenge to the industry. This study aims to authenticate rice samples using visible-short wavelength hyperspectral imaging (Vis-SWNIR HSI). To achieve this goal, 163 intact rice samples were sourced from three northern provinces of Iran (Gilan, Mazandaran and Golestan), including four different varieties (Hashemi, Shiroodi, Fajr and Neda). Samples were scanned by HSI device to record their cubic data. The HSI data of different samples were used in a column-wise augmented data matrix with pixels of different samples as rows and wavelengths as columns. The pure spatial (distribution maps) and spectral profiles of desired components were extracted using multivariate curve resolution-alternating least squares (MCR-ALS). As low-quality rice samples are considered adulterants for high-quality samples, the obtained data were used to differentiate samples by their origin as well as in adultertation detection. To the best of our knowledge, the MCR-ALS algorithm has not been used for rice authentication before using their intact forms. Further chemometric analyses, including principal component analysis (PCA) and partial least squares-discriminant analysis (PLS-DA), were applied to the spectral-resolved profiles of samples. This enables us to discriminate three different origins and four varieties of rice samples, as well as for adulteration detection, providing classification accuracies of 94.4 %, 82.75 % and 100 % in the prediction sets, respectively. Such an outcome indicates the practical feasibility of imaging methods for rapid, cost-effective and non-invasive food authentication purposes.