Identification Of Vegetable Oils By Near Infrared-Raman Spectroscopy And Multispectral Fusion

The consumption of plant edible oil in China has been increasing with the improvement of people's living standards.Some unscrupulous traders have sought to obtain benefits,shoddy,and use inferior products to produce a lot of counterfeit edible oils,even for the recovery of catering and secondary oils.Reprocessing is sold to consumers and endangers people's health.At present,traditional chemical detection methods require higher professional knowledge and longer detection time.The later developed spectral-based rapid detection methods generally rely only on single-spectrum data,are susceptible to noise and do not fully analyze sample information.Therefore,it is of great significance to explore a technology that can integrate multi-source spectral information and analyze and detect vegetable oil quickly and sensitively.Based on Raman spectroscopy and near-infrared spectroscopy,this paper studies the effective fusion methods of these two kinds of spectra,and studies the method of establishing a detection model with better performance,so as to achieve rapid identification of plant edible oil varieties and rapid detection of recovered oil.The main contents are as follows:(1)In order to rapidly identify the varieties of edible vegetable oils,a fusion method of Raman spectroscopy and near-infrared spectroscopy was investigated.Combined with support vector machine classification(SVC)method,the performance of the SVC model established by serial fusion,wavelet fusion,and typical correlation analysis fusion data was studied.Studies have shown that these three fusion methods can rapidly identify plant edible oil varieties.After exploring a variety of spectral preprocessing methods and model parameter optimization methods,the prediction accuracy rate of the model based on serial fusion and wavelet fusion reached 100%,and the prediction accuracy rate of models based on typical correlation analysis fusion reached 89.74%.It shows that the serial fusion method and wavelet fusion method can establish the authenticity identification model of vegetable edible oil with better prediction effect.(2)In order to achieve rapid identification of the authenticity of plant edible oils,a method for the establishment of classification models was explored.Explore the performance of four classification models: Support Vector Machine Classification Model(SVC),Partial Least Squares Support Vector Machine(LS-SVM),Multicore Learning Support Vector Machine(MKL-SVM),and Partial Least Squares Linear Discriminant Analysis Model(PLS-LDA).Studies have shown that these four model establishment methods can achieve rapid identification of authenticity of plant edible oils,among which SVC,LS-SVM and PLS-LDA models have the highest prediction accuracy of 100%,and MKL-SVM has the highest prediction accuracy of 99.15.%.In terms of model generalization capability and stability,the SVC model has the best performance.-+0(3)A rapid oil recovery model is studied and established based on Raman and near-infrared spectroscopy.Considering the research on the authenticity identification model of vegetable oil,serial fusion and wavelet fusion technology combined with SVC,LS-SVM and PLS-LDA modeling methods were established to realize rapid detection of recovered oil.Among them,the PLS-LDA model based on the fusion of Raman and near-infrared spectroscopy serials has the best performance.The accuracy of prediction reaches 100%,the specificity and sensitivity reach 100%,and the error rate of cross-check is only 0.0058,which indicates that the model has strong Generalization ability and stability.