Research On Quality Detection And Analysis Of Edible Oil Based On NIRS Technology

Frying has become one of the most common cooking methods,which can make food have a good flavor and crisp taste.But,with the increase of frying times,the harmful substances in edible oils will increase,affecting the quality of the edible oil.Research on the quality of edible oil at domestic and foreign mostly adopts chemical analysis methods,which evaluate the quality of edible oil by detecting the content of total polar compounds(TPC),acid value,peroxide and other substances in edible oil.However,the chemical analysis methods were more cumbersome and will destroy the experimental samples,so this paper proposes to use near-infrared spectroscopy to detect frying oil samples,establish a relationship between TPC and frying times,and use frying times as a measure to evaluate the quality of edible oil.In this study,soybean oil,peanut oil,and rapeseed oil were fried for multiple times according to the standard experimental procedure,and 150 fried samples were obtained for each oil.Firstly,a near-infrared analyzer was used to collect the near-infrared spectra of samples.Then,by using a liquid chromatograph to measure the content of TPC in the edible oil sample,the analysis verified the linear correlation between the frying times and the content of TPC,and based on this,the quality of the edible oil was evaluated by frying times.First,for the near-infrared spectra data,the first-order derivation(D1),second-order derivation(D2),multivariate scattering correction(MSC)and standard normalized variable(SNV)were used to pre-process the near-infrared spectra to reduce the spectral noise.Then,the feature selection methods were used to extract the spectral feature wavelengths,and the classification and regression models were respectively established according to the feature wavelengths to predict frying times.Among them,the classification model established differential prediction model based on the characteristics of the spectra processed by D1 and the correlation coefficient method,and finally obtains a prediction result of 100%within the error range of±1.The regression prediction method used sample set partitioning based on joint x-y distance(SPXY)algorithm to divide the sample set,used the single variable feature selection method to extract the feature wavelength,and established partial least square regression(PLSR)model,support vector regression(SVR)model and Bayesian ridge regression(BRR)model respectively.By comparing the coefficient of determination(R~2),prediction root-mean-square error(RMSEP)and model training time of each model,the BRR model was determined as the best model to predict the frying times of frying oil,and the final result was that R~2 was0.9852 and the RMSEP was 0.5463.The research showed that the classification and regression model used in this paper can effectively predict the frying times and then determine the quality of edible oil,and also provide an efficient and feasible identification method for the rapid detection of the quality of edible oil.