Research On Detection Method Of Coumarin-Based Food Additives Based On Terahertz Spectroscopy

With the rapid development of the food industry,the safe use of food additives is directly related to people’s health as essential auxiliary materials in food production,the importance of which is self-evident.For the shortcomings of traditional detection methods of coumarin food additives and the limitations of current spectroscopy,terahertz Time-Domain Spectroscopy(THz-TDS)technology is gradually applied to food safety detection as a new spectral detection method because of its unique characteristics such as penetrability,security,and fingerprint.This paper uses coumarin-based food additives as the detection object.The combination of THz-TDS technology and machine learning algorithm is applied to the detection of food additives(qualitative identification and quantitative analysis).The main research contents are as follows:(1)The research on THz spectral measurement and analysis of coumarin-based food additives.First,standard samples were prepared according to the characteristics of coumarin-based food additives.Then,the THz spectra of the samples were measured by the THz-TDS system.Next,the data was preprocessed by operations such as De-averaging,Gaussian window,and Data smoothing.By Fast Fourier Transform and Optical Parameters,the absorbance spectra of the studied substances were finally obtained and analyzed,and it would provide a data basis for subsequent models.(2)The research on qualitative identification of coumarin-based food additives.The216 sets of THz spectra of six coumarin-based food additives were measured by the THz-TDS system.To deal with the problem of the long-running time of the classification model due to the relatively high dimension of sample absorbance data,a combination method(P-t-SNE)of manifold learning t-distributed stochastic neighborhood embedding(t-SNE)and principal component analysis(PCA)was used to extract features of spectral data for simplify modeling time and retaining enough useful information.To deal with the problem that the traditional grid search method(GS)and cross-validation(CV)method were not sensitive to parameter optimization of the support vector machine(SVM),the differential evolution(DE)algorithm was introduced to improve the performance of the gray wolf optimization(GWO)algorithm,and then the combined new algorithm(DEGWO)could find the best parameter combination of SVM.The original GWO-SVM model and the improved DEGWO-SVM model were established to compare the modeling effects.The research result showed that the P-t-SNE-DEGWO-SVM method proposed in this paper had the best recognition effect,with the calibration set accuracy of 100% and the prediction set accuracy of 98.61%,which provided a good reference value for establishing an efficient and accurate food additive identification model.(3)The research on quantitative analysis of the two-component mixture of coumarin and its derivatives.The powder of coumarin and 6-methyl coumarin was proportioned according to the isocratic method,and a total of 396 sets of full spectral data with 11 concentrations were constructed.The linear model partial least squares(PLSR)and nonlinear model support vector regression(SVR)were used to establish prediction models of coumarin content in this paper.It was found that the modeling effect of SVR was much stronger than that of PLSR.To further improve the prediction accuracy of the quantitative model,the competitive adaptive reweighting(CARS)algorithm was used to select variables for THz full spectra,and then PLSR and SVR models were established again to find the best quantitative analysis model.The research result showed that the CARS-SVR model established in this paper had the best quantitative prediction effect,with the calibration correlation coefficient of 0.9962 and the calibration root mean square error of2.8513%,which provided an effective quantitative regression analysis method for the content prediction of coumarin and its derivatives.