Rapid Authentication Of Camellia Oil Based On UV/VIS/NIR Spectroscopy

In order to explore a rapid,accurate and stable method for the authentication of adulterated camellia oil,this paper conducted a rapid identification study of adulterated camellia oil by ultraviolet/visible/near infrared spectroscopy(UV/VIS/NIR)(200-1100 nm).Acquisition of transmission spectra of adulterated camellia oil sample matrices with an in-house built experimental platform;A qualitative and rapid authentication model of adulterated camellia oil with high accuracy,sensitivity and specificity was established by comparing various spectrum pre-processing methods,feature wavelength selection methods and machine learning modeling methods;The transmission spectra were pre-processed and the effective wavelengths were screened using various feature extraction algorithms,a fast prediction model of adulterated camellia oil content was established on this basis.The robustness of the proposed model was verified by analyzing the wavelength characteristics in terms of the degree of correlation between the adulteration rate and the absorbance of the characteristic wavelength;The qualitative and quantitative model authentication performance was evaluated by a re-prepared matrix of adulterated samples.This study provides a model basis for the development of a low-cost spectrum nondestructive detection device for adulterated camellia oil.The details of the study are as follows:1)Preparation of adulterated camellia oil sample matrix and construction of experimental platform for spectrum acquisition.The sample matrices of camellia oil adulterated with soybean oil,rapeseed oil,corn oil,peanut oil were firstly prepared according to different ratios,then the sample spectra in the range of 200-1100 nm were collected by the transmission experimental platform built independently,finally the sample transmission spectrum curves were obtained by the calculation of absorbance.2)Establish a qualitative and rapid authentication model for adulterated camellia oil.Firstly,the full spectrum was preprocessed with MSC,SNV,MA,SG,1stDeriv,2ndDeriv,SG-1stDeriv,SG-2ndDeriv,SG-CWT(Continuous Wavelet Transform),then the Competitive Adaptive Reweighted Sampling(CARS),Successive Projections Algorithm(SPA),Bootstrapping Soft Shrinkage algorithm(BOSS),Iterative Variable Subset Optimization algorithm(IVSO)were used for feature wavelength screening,finally an adulterated camellia oil authentication model based on Extreme Gradient Boosting algorithm(XGBoost)was developed.The results showed that the original spectrum were pre-processed by SG-CWT(decomposition scale 25,L5)and screened by IVSO,the established XGBoost model had the best discrimination performance,with the accuracy,sensitivity and specificity of the test set reaching 98.92%,100%,98.84%,respectively.And its authentication accuracy can reach 100%when the proportion of adulteration is more than 1%,while the authentication accuracy of camellia oil for adulteration proportion of 1%also reached 85.71%.In addition,the model was experimentally evaluated by re-collecting the spectrum of samples with multiple adulteration ratios,the results showed that the SG-CWT(L5)-IVSO-XGBoost discrimination model could achieve 98.57%accuracy,100%sensitivity,98.33%specificity for 70 adulterated samples,respectively.It was demonstrated that the proposed qualitative model showed good accuracy in the authentication of independent samples.3)A fast prediction model(quantitative)of adulterated camellia oil content was established,the robustness of the model was verified by the proposed characteristic wavelength characteristics.After SG-CWT pre-processing of the transmission spectrum of the collected samples,CARS,SPA,BOSS and IVSO were used for feature wavelength selection,and fast prediction models based on support vector machine(SVM)and random forest(RF)were established for the adulteration content of camellia oil,respectively.The results showed that the prediction set coefficients of the models for predicting the content of camellia oil blended with soybean oil(BOOS-SVM),corn oil(IVSO-RF),peanut oil(IVSO-RF),rapeseed oil(IVSO-RF)were 0.9928,0.9845,0.9674,0.9937,the root mean square errors were 0.0317,0.0398,0.0585,0.0267,the mean absolute errors were 0.0170,0.0282,0.0392,0.0155,respectively.Moreover,the wavelength characteristics analysis showed that the adulteration rates of the proposed wavelengths in the three wavelength bands were strongly correlated with the absorbance,which proved the robustness of the model.In addition,the adulteration effectiveness of the proposed adulteration content prediction model was verified with 60 sets of new samples.The experimental results showed that terms of quantitative models,the RP2 of the prediction models for soybean oil(SG-CWT(L5)-BOOS-SVM),corn oil(SG-CWT(L5)-IVSO-RF),peanut oil(SG-CWT(L5)-IVSO-RF),rapeseed oil(SG-CWT(L5)-IVSO-RF)adulterated with camellia oil content were 0.9906,0.9729,0.9626,0.9892,RMSEP were 0.0322,0.0548,0.0647,0.0352,MAEP were 0.0166,0.0309,0.0316,0.0231,respectively.In summary,the proposed quantitative models are proved to show strong robustness in the authentication of independent samples.