Study On Identification Of Edible Oil And Adulteration Of Camellia Oil Based On Molecular Vibration Spectroscopy

Camellia oil is highly sought after by consumers because of its rich nutritional value and multifunctional use,yet it is frequently adulterated.In this study,molecular vibration spectroscopy combined with chemometrics was used to establish a rapid and efficient method for the identification of edible oil types and camellia oil adulteration.The main research results are as follows:(1)A discrimination model for six types of edible oil species was constructed based on molecular vibration spectroscopy.After the spectra were multivariate scattering corrected(MSC),partial least squares(PLS)and convolutional neural networks(CNN)were used to construct the six edible oil discriminative models,respectively.The discrimination accuracy of the PLS and CNN models constructed based on Raman spectra were 100%and 94.74%,respectively;the mid-infraredspectral(MIR)models were 100%and 94.74%,respectively;and the near-infrared spectral(NIR)models were 94.40%and89.47%,respectively.All models identified 100%of the camellia oil samples,with misclassifications only occurring between the other five vegetable oils.(2)The models of camellia oil adulteration identification and quantitative detection of adulteration were constructed based on molecular vibration spectra.After the spectra were preprocessed by MSC,the accuracy of the established PLS and CNN models for discrimination of six types of the adulterated camellia oil was 83.87%and 80.65%(Raman),87.10%and 83.87%(MIR)and 80.64%and 83.87%(NIR),respectively.The identification rates of the proposed models were all 100%for the unadulterated camellia oil samples.The relative prediction deviations(RPD)of the PLS models were 2.77(Raman),2.52(MIR)and 1.48(NIR)for the adulteration detection models,respectively.In addition,the selected seven Raman peak intensity ratios and infrared spectral peak intensities of unsaturated hydrocarbon(=CH)bonds could be achieved for the determination of single species vegetable oil adulteration,the correlation coefficients R~2were 0.900～0.998 and 0.864～0.996,respectively.(3)The accuracy of the PLS and CNN models for the six types of edible oil species was 100%and 94.74%(Raman-MIR fusion),100%and 94.74%(Raman-NIR fusion),and100%and 100%(MIR-NIR fusion),respectively.The accuracy of the PLS and CNN models for camellia oil adulteration discrimination was 100%and 90.32%(Raman-MIR fusion),93.55%and 96.77%(Raman-NIR fusion),and 100%and 90.32%(MIR-NIR fusion),respectively.The PLS quantitative model prediction based on fused data outperforms the PLS quantitative model effect based on single spectra,with the RPD values improving to 2.99(Raman-MIR fusion and Raman-NIR fusion)and 1.93(MIR-NIR fusion),respectively.Therefore,performing data fusion can improve the performance of the model to some extent.