Research On The Detection Method Of Oil Frying Frequency Based On NIRS

In the process of deep-frying,oil can produce deleterious compounds that are harmful to human health.This article proposed rapid detection methods for oil frying times based on near infrared spectroscopy(NIRS)technology.10 experiments were conducted and in each experiment,frozen French Fries were fried in the same batch of soybean oil for 15 times.First derivative(1D),second derivative(2D)and standard normal variable transformation(SNV)were used as the preprocessing methods.Characteristic wave points sensitive to frying times were extracted by correlation coefficient method and successive projections algorithm(SPA).Support vector machine(SVM),partial least squares regression(PLSR)and radial basis function(RBF)neural network were used to establish detection models in the prediction of oil frying times,and the best data-processing methods were determined by comparing the performance of different models.Through analyzing,it could be found that the SVM detection model had the best performance when 2D was served as the preprocessing method,and correlation coefficient method was used to extract the characteristic wave points.Six characteristic wave points were extracted and discriminant accuracy of the best SVM model reached94%.Besides,performance of the best PLSR model was superior to that of the best RBF neural network model,in which 1D was used as the preprocessing method,and seven characteristic wave points were extracted by SPA.Values of the evaluation parameters R~2,RMSEP and RPD were 0.9972,0.2194 and 18.9594,respectively.Results showed that the analysis methods proposed in this study could make accurate predictions of oil frying times,and a new way of detection of food safety was provided.