Fusion of dielectric spectroscopy and computer vision for quality characterization of olive oil during storage

Oxidation level and quality characteristics of olive oil require monitoring during storage to ensure that their amounts are maintained in the lawful thresholds. It is especially important for licensing their commercialization as high-value virgin olive oils. The present research proposes a novel approach based on the fusion of dielectric spectroscopy and computer vision for the characterization of olive oil quality indices during storage in order to reduce the time of analysis, reagent consumption, manpower and high-cost equipment. Colour features in RGB, HSV and L∗a∗b∗ spaces were extracted as well as dielectric features in the frequency range of 40 kHz to 20 MHz for each olive oil sample. After data pre-processing, classification and prediction models were developed and compared. Several machine learning techniques were investigated for storage time classification and quality indices prediction including artificial neural network (ANN), support vector machine (SVM), Bayesian network (BN) and multiple linear regression (MLR). The best result in the classification of olive oils during the storage period was obtained by BN technique with 100% accuracy. Among predictive models, the SVM with RBF kernel had the best results (R = 0.969, 0.988 and 0.976) for prediction of peroxide value (PV), UV absorbance at 232 nm (K232) and chlorophyll, respectively. Also, the SVM with normalized polynomial kernel had the best results (R = 0.989, 0.976, 0.969 and 0.969) for prediction of p-Anisidine value (AV), total oxidation value (TOTOX), UV absorbance at 268 nm (K268) and carotenoid, respectively. The ANN with 40-2-1 topology gave the best result (R = 0.977) for modelling free acidity (FA). Results of this research can be utilized for developing an efficient and reliable system for olive oil quality evaluation and monitoring by industry.