Study on Detection Methods for Frying Times of Soybean Oil Based on NIRS

Abstract

In the process of deep frying, oil can produce deleterious compounds which are harmful to human health. On the basis of analyzing the changing mechanisms of chemistries in repeatedly used oil, the study proposed a method for rapid detecting the frying times of oil based on near infrared spectroscopy (NIRS) technology. First derivative (D1), second derivative (D2) and standard normal variable transformation (SNV) were served as pretreatment methods, and characteristic wavelengths which sensitive to frying times were extracted by correlation coefficient (CC) method. Support vector machine (SVM), partial least squares regression (PLSR) and radial basis function neural network (RBFNN) were utilized to establish qualitative and quantitative analysis models. It turned out that the qualitative and quantitative analysis models had the best performance when D2 was used to pretreatment spectra and six characteristic wavelengths were extracted. More precisely, classification accuracy of the best SVM model reached 94%. Also, the performance of the best PLSR model was superior to the best RBFNN model, in which the values of correlation coefficient (R2), root mean square error of prediction (RMSEP), residual predictive deviation (RPD) were 0.9937, 0.3477 and 12.5803 respectively. The overall results indicated that the proposed method had a great potential to accurate detect frying times of oil.