Study On Detection Method Of Trans Fatty Acid Content In Edible Oil By Laser Raman Technique

In order to achieve the rapid detection of trans fatty acids(TFAs)content for edible oil after heating,the rapeseed oil,soybean oil and corn oil were heated at190°C(usually frying temperature)at different times(0,30,60,90,120,150 min),and then the samples were detected by confocal laser Raman spectrometer,36 Raman spectra were obtained from each sample,and a total of 648 Raman spectra were collected.Through the analysis of Raman spectroscopy data,the qualitative and quantitative analysis model of TFAs content in edible oil was established to achieve the rapid and accurate detection and evaluation of TFAs content.The main conclusions of this paper are as follows:1.Polynomial smoothing and standard normal variable transformation was used to preprocess the original spectral data to remove the interference of background and noise.2.The no information variable elimination method(UVE)was used to screen the characteristic variables of the full-band Raman spectroscopy data,75,153 and 23wavelength variables were selected from all wavelengths of the rapeseed oil,soybean oil and corn oil,respectively.Using the automatic peak finding algorithm,the peak value,peak valley and peak area of the Raman spectra of rapeseed oil,soybean oil and corn oil were used as feature representation information.The spectral peak decomposition characteristics of the three edible oils were further screened using the no-information variable elimination method.3.First,qualitative and quantitative models were established based on the full spectral data and UVE screening feature spectral variables,respectively.The content of TFAs of the three edible oils with different heating times were determined qualitatively and quantitatively,and the experimental results were compared and analyzed.The results show that the qualitative discriminant model was established by Fisher discriminant analysis(FDA)based on the selected variables.The discriminant accuracy rate was increased from 40%~50%to over 90%,which indicates that the selected variables can better characterize the feature information of the samples.At the same time,the mathematical prediction model of TFAs content was established by using partial least squares regression(PLSR),BP neural network(BPNN)and support vector machine regression(Support vector machine)The regression,SVR)method based on the selected variables and the full-spectrum data.Through the comparative analysis of the prediction results,it was shown that the non-information variable elimination combined with the support vector regression machine method(UVE-SVR)had a good detection effect.R~2 of rapeseed oil,soybean oil and corn oil were upgraded from 0.850 4,0.935 4 and 0.753 4 to 0.959 6,0.995 0 and 0.952 8,respectively.Second,the analysis was performed based on the Raman peak decomposition method.The FDA results showed that in the different kinds of edible oils,with the change of the number of spectral peaks,the identification correct rate also changed,whether it was a single characterization pattern or a multi-feature characterization pattern.The UVE algorithm was used to further screen the 3 characteristic combination variables,and the FDA identification accuracy rate of the three edible oils was increased to 100%.Based on the characteristic values of Raman peaks,the contents of trans fatty acids in three kinds of edible oils were predicted by SVR,PLSR and BPNN respectively,and the results were compared and analyzed.The results of the analysis showed that the content of trans fatty acids in rapeseed oil,soybean oil and corn oil was predicted by SVR,the R~2 of the training set was above 0.99,and the R~2 of the prediction set were0.983,0.987 and 0.982,respectively.When BPNN was used,the R~2 of the prediction set is 0.981,0.991,and 0.989,respectively.The results show that it is effective to predict the contents of the trans fatty acids in edible oil by using the characteristic variables extracted by spectral peak decomposition method and UVE as the input vector of SVR model and BP neural network model.Meanwhile,it can also improve the accuracy and robustness of the model,and greatly reduce the computational complexity of the model.