Research On Rapid Detection Method Of Edible Oil Quality And Variety Based On Data Fusion Of Near-infrared And Raman Spectroscopy

China is the largest country of edible oil consumption in the world.And edible oil plays a decisive role in our daily lives.Edible oil can provide essential nutrients for our body,such as phytosterol,unsaturated fatty acid,oryzanol,etc.,and has more and more effect on health.However,some unscrupulous producers and traders often mixed the high valued edible vegetable oils with cheap edible oil.Adulteration of edible oils not only lowers the quality of the oils' hygiene and nutrition,which do great harm to the consumers' health,rights,and interests.Currently the traditional chemical methods cannot serve as an on-site detection approach,has many disadvantages,such as requiring specialized personnel operating instruments and a variety of chemical instruments and reagents,and time-consuming detection.Hence for the safety control for oil companies and to protect edible oil market and interests of consumers,it is essentially necessary to research a highly sensitive analysis system which can detect edible oil rapidly and dynamically.In recent years,the use of Raman and near-infrared spectrum technology for edible oil quality determination have been widely studied,because spectrum technology goes fast,convenient,non-destructive,without pretreatment to samples,on-line analysis,and pollution-free.It has been used to detect rapidly in food,chemical industry and other fields.This study aimed to detect the edible oil varieties,adulteration and authenticity of peanut oil,fatty acids by combining data fusion of Raman and NIR with chemometries algoriths.Qualitative discriminant models and quantity prediction models were established by using data fusion technology to detect authenticity of edible oil.The main research results are summarized below:(1)Qualitative discriminant models which were established based on data level and characteristic level fusion of multi-source spectroscopy could implement the prediction of the varieties of edible oil rapidly.The accuracy of prediction set for Raman-NIR-SVC model which was established by combining characteristic level fusion of Raman and NIR spectroscopy with support vector machine classification(SVC)reached 100%;Compared with single spectral analysis and data level fusion of multi-source spectroscopy,prediction efficiency of SVC model which was established based on characteristic level fusion of Raman and NIR with SVC is better;According to the analysis,combining characteristic extraction methods and characteristic level fusion of multi-source spectroscopy technology is significance to study the comprehensive identification of edible oils.(2)Qualitative discriminant models and quantity prediction models which was established based on data level and characteristic level fusion of multi-source spectroscopy data could implement the detection of authenticity and adulteration of peanut oil rapidly.Compared with single spectral analysis and data level fusion of multi-source spectroscopy,prediction results of adulteration content based SVR model which was established by combining characteristic level fusion of Raman and NIR with SVR is better,and the coefficient of determination R2 of prediction set for SVR model reached 0.9819.Compared with single spectral analysis and data level fusion of multi-source spectroscopy,prediction results of authenticity for peanut oil based PLS-LDA model which was established by combining characteristic level fusion of Raman and NIR with PLS-LDA is better;the accuracy of prediction set for PLS-LDA model reached 100%.And it reflects the very good complementarity of Raman and near infrared spectroscopy.The detection results of authenticity and adulteration of peanut oil are better by using model which was established based on characteristic extraction methods.(3)Quantity prediction models which was established based on data level and characteristic level fusion of multi-source spectroscopy could implement the detection of content of MUFA and PUFA rapidly.Compared with single spectral analysis and data level fusion of multi-source spectroscopy,prediction results of MUFA content based SVR model which was established by combining characteristic level fusion of Raman and NIR with SVR is better,and the R2 of prediction set reached 0.9773.Compared with single spectral analysis and data level fusion of multi-source spectroscopy,prediction results of PUFA content based SVR model which was established by combining characteristic level fusion of Raman and NIR with SVR is better,not only is the R2 of sample set above 0.9839 but also the difference of MSE of sample set was smaller.The SVR model has advantageous properties such as better prediction ability.So studying characteristic extraction methods is significative.