Applying multivariate curve resolution modelling combined with discriminant tools on near-infrared spectra for distinguishing between cheese varieties and stages of ripening

Abstract

In this study, near-infrared (NIR) spectra were employed to monitor the ripening process of two kinds of soft cheese produced in the Extremadura region of Spain, manufactured by two different producers, “Torta del Casar” and “Queso de la Serena”. Spectra were collected from the interior of the cheeses and the rind and analysed using appropriate chemometric techniques to distinguish between the two varieties and among different weeks of the maturation process. Different chemometric tools, including multivariate curve resolution with alternating least squares (MCRALS), linear discriminant analysis (LDA), quadratic discriminant analysis (QDA), and feed-forward artificial neural networks (FF-ANN), were utilised, resulting in outstanding discrimination outcomes with sensitivity, precision, specificity, and accuracy reaching values c.a. 1.00 in optimal scenarios. More comprehensive information was acquired from the rind spectra analysis, indicating that the sampling process can be performed without disturbing the cheese in a non-destructive way. Remarkably, the capability to distinguish between various weeks of ripening for both cheeses could enable manufacturers to produce market-ready products earlier than the typically established timeline.