Rapid and non-destructive spatially offset Raman spectroscopic analysis of packaged margarines and fat-spread products

Open Access

Abstract

This work was partially supported by University of Granada (Spain) within the framemork of the funding corresponding to program 'precompetitive research projects for young researchers'. Funding for open access charge: University of Granada/CBUA. AMJC wish to acknowledge the Department of Economic Transformation, Industry, Knowledge and Universities belong to Regional Andalusia Government (Spain) for the Postdoctoral fellowship (DOC_00121). In addition, AAC wants to express their sincere gratitude to the Spanish Ministry of Universities for a pre-doctoral fellowship FPU (FPU20/04711, Formaci ' on del Profesorado Universitario). ; Spatially offset Raman spectroscopy (SORS) is a novel technique capable of measuring samples through the original packaging and recovering the spectra without the contribution of surface layers. Here, a portable SORS equipment was used to measure 62 samples of margarines and fat spreads through the original plastic container. Chemometric tools were used to analyse the data obtained. A total of 25 classification models were developed based on: (i) geographical origin, (ii) vegetable oils and (iii) some significant minor constituents present in the samples. Partial least squares-discriminant analysis (PLS-DA), support vector machine (SVM) and soft independent modelling of class analogy (SIMCA) were used for model classification. Quantitative analysis using the partial least squares regression (PLSR) method was also performed to determine the total fat content. In parallel, a benchtop conventional Raman spectrometer was used to analyse the same samples, develop the models with the same training and validation sets in order to compare the results. The calculated classification performance metrics showed better classification models from SORS data than conventional Raman spectroscopy (CRS), highlighting the one-class SIMCA models for margarines containing phytosterols, olive oil or linseed oil. These models exhibited very high predictability (performance parameters with values equal to or ...