Rapid identification of edible oil and swill-cooked dirty oil by using near-infrared spectroscopy and sparse representation classification

Abstract

Rapid identification of edible oil and swill-cooked dirty oil is a challenging and important task in the field of food safety. The main object of this investigation was to distinguish edible oil (QO) and swill-cooked dirty oil (SO) by employing near-infrared (NIR) spectroscopy and the sparse representation classification (SRC) method. Because of the diversity and uncertainty of the species in swill-cooked dirty oil, building a classification model based on NIR spectroscopy faces the problems of complex systems and small numbers of samples. To improve the stability and accuracy of the identification, in the SRC method, the redundant dictionaries for QO and SO were trained, and the sparse representation coefficients for spectra in a validation set under both dictionaries were calculated. Then the spectra in the validation set were reconstructed with the sparse representation coefficients and the corresponding dictionary. Finally, the reconstruction errors under the QO and SO dictionaries were used as indicators for classification. Moreover, a simplified SRC algorithm (SRC-S) that directly uses the calibration set spectra as redundant dictionaries was proposed, and this removed the dictionary training process and avoided information loss during training. Compared with linear discriminant analysis (LDA) and partial least squares discriminant analysis (PLS-DA), the experimental results showed that the SRC-S outperformed SRC, and it reached a maximum classification accuracy of 95.37%, which proved that SRC-S and NIR spectroscopy can distinguish QO and SO.