Development of a computer vision system for the classification of olive oil samples with different harvesting years and estimation of chlorophyll and carotenoid contents: A comparison of the proposed method's efficiency with UV-Vis spectroscopy

This study focused on the potential of UV-Vis spectroscopy and digital images and processing combined with chemometrics for classifying olive oil samples in terms of variety and harvest year and estimating pigment content. Different classification models were developed utilizing data from the RGB color space to classify olive oils. The classification of olive oil types was performed using Partial Least Squares-Discriminant Analysis (PLSDA), Principal Component Analysis-Support Vector Machine (PCA-SVM), and k-nearest neighbor (k-NN). PLS-DA was further employed for harvest-year classification. The PLS-DA model achieved 100% accuracy in most classes except one, while k-NN and PCA-SVM exhibited flawless 100% accuracy across all prediction dataset classes. PLS-DA also demonstrated perfect classification based on harvest year. Pigment estimation utilized PLSR, revealing RGB images as the most effective for carotenoid prediction (R2Pre = 0.995 and RMSEP = 0.304), and UV-Vis spectroscopy excelled in chlorophyll prediction (R2Pre = 0.998 and RMSEP = 0.047). These results suggest that imaging systems may accurately anticipate pigment concentrations and, in some situations, provide comparable precision to spectroscopic methods. The aformentioned findings highlight the efficacy of this approach, providing a potential alternative for swiftly identifying olive oil harvest years, freshness and estimating pigment concentrations. Moreover, the method is cost-effective, reagent-free, and accelerates sample analysis.