Cold-Pressed Walnut-Oil Adulteration with Edible Oils Detection Using Vis-NIR Spectroscopy

The aim of this study is to evaluate the usefulness of UV-Vis-NIR spectroscopy as a tool for detecting the adulteration of cold-pressed walnut oil and other edible oils (rapeseed, sunflower, and soybean oils) at varying percentages. The spectra were recorded between 200 and 1800 nm, but the analyses focused on 350-1650 nm due to high UV and NIR absorption. Color was determined in CIEL*a*b* coordinates to achieve the differences among the samples. The spectra were submitted to several pre-treatment (none, normalization, SNV, MSC, baseline/detrend, first/second derivative, and 1st-order smoothing) to improve the statistical model's parameters. The differentiation of the samples was carried out using an unsupervised method (principal component analysis-PCA) and two supervised methods (linear discriminant analysis-LDA and partial least squares linear discriminant analysis-PLS-DA). Partial least squares regression (PLS-R) was used for predicting the degree of adulteration. Separation between the authentic and adulterated samples was visible in the PCA scores plot, primarily along the spectral regions of 420-500 nm (pigment-related absorption band) and 1150-1450 nm (lipid-associated band). PLS-DA was superior to DA for the discrimination of authentic/adulterated samples, with baseline spectra of 350-1650 nm yielding a 100% overall accuracy and near-perfect accuracy with MSC (98.48%). PLS-R was able to predict the adulteration level, depending on the pre-treatment applied.