Identification of Edible Oil Based on Multi-source Spectra Data Fusion

An approach based on multi-source spectra data fusion for identification of edible oil is proposed. A qualitative model based on fusion of Raman spectra and near-infrared spectroscopy (Raman-NIR) was established and compared with conventional single-spectra model. The spectra data was pre-processed using the moving average method (MA11), the Savitzky-Golay method (SG9), the adaptive iteratively reweighted penalized least squares method (airPLS), the normalization method (Nor), the multiplicative scatter correction method (MSC), and the standard normal variant and standard normal variant transformation de-trending method (SNV-DT). Then, optimized characteristic variables were selected using the competitive adaptiive reweighted sampling method (CARS-SPA) and the backward interval partial least squares method (BiPLS). Based on that, a model for identification of edible oil was established using the support vector classification method (SVC). The results revealed that the SVC model established can accurately identify and classify eight different edible oil (soybean oil, peanut oil, rapeseed oil, tea seed oil, rice oil, corn oil, sunflower oil, and palm oil). The prediction accuracy for samples in calibration set and prediction set by the proposed model can be 100%, which is superior to that of conventional single-spectra model. The proposed model exhibits excellent generalization capability. Additionally, the study suggests that the Raman-NIR fusion shows improved efficiency in identification of edible oil and great potential for practical application.