Research On Detection Method Of Citri Reticulatae Pericarpium Quality Based On Terahertz Time Domain Spectroscopy Technology

Citri Reticulatae Pericarpium(CRP)is the dried,ripe peel of Citrus reticulata Blanco(Rutaceae)and its cultivated varieties,which is used for both medicinal and dietary purposes.Due to the influence of its origin,climate and ecological environment,there is a great difference between the CRP of different origins,which directly leads to differences in its medicinal effects and value.As an authentic medicinal material,Xinhui CRP produced in Xinhui County,Guangdong Province is the best,and its quality is closely related to the aging time.However,CRP of different storage ages have similar morphological characteristics,which makes it quite difficult to distinguish.Therefore,the detection of the quality of different-age CRP is of great guidance.In this study,terahertz time-domain spectroscopy(THz-TDS)combined with chemometric methods and deep learning algorithms were used for the identification of different-age CRP.Additionally,the THz fingerprint peaks of flavonoids were also investigated,and the possibility of THz-TDS for the rapid detection of flavonoids in different-age CRP was also explored.The main conclusions of this topic are as follows:(1)The THz-TDS technique based on tablet pressing method realized the rapid identification of the age of different-age CRP.The THz spectra of the Xinhui CRP stored for 1,3,5,10,15 and 20 years were obtained by transmission THz time-domain spectrometer,and the absorption coefficient spectra and the refractive index spectra in the range of 0.2-1.5 THz were selected for analysis,and the classification models of partial least squares discriminant analysis(PLS-DA),random forest(RF),support vector machines(SVM),least squares support vector machines(LS-SVM)and convolutional neural network(CNN)were established.The results showed that the CNN model outperformed PLS-DA,RF,SVM and LS-SVM models in terms of correct classification in both absorption coefficient spectra and refractive index spectra,and reached the highest accuracy of 95.63%in the absorption coefficient spectra and 90.62%in the refractive index spectra.In addition,the t-SNE algorithm was used to visualize the pre-processed spectra and CNN intermediate layer data,and the results showed that the boundaries of the sample distribution were not obvious after pre-processing,and both had cross overlap,while the output data points of CNN model intermediate layer gradually changed from overlapping to clearly separable,demonstrating the feasibility of the THz-TDS technique for the identification of different-age CRP.(2)The THz-TDS technique based on in situ method realized the rapid nondestructive identification of different-age CRP.The absorption coefficient spectra and the refractive index spectra of Xinhui CRP stored for 1,3,5,10,15 and 20 years were analyzed in the range of 0.2-1.5 THz,and PLS-DA,RF,SVM,LS-SVM and CNN models were established.The results showed that the refractive index spectra provided better discrimination models than were obtained with the absorption coefficient spectra,and the CNN model outperformed PLS-DA,RF,SVM and LS-SVM models,reaching the highest accuracy of 78.40%in the absorption coefficient spectra and 79.63%in the refractive index spectra.The output data of CNN middle layer were visualized by t-SNE algorithm,and it was found that the data points also gradually changed from overlapping to obviously separable,which indicated that THz-TDS technology combined with CNN model provided theoretical feasibility for nondestructive detection of different-age CRP.In addition,the colour analysis of the different-age CRP revealed a decreasing trend in the L~*,a~*and b~*values of the samples aged within 10 years,which confirms that the aging of the CRP is an orange to brown process,while the L~*,a~*and b~*values of the samples aged 15 and 20 years did not show a significant change.(3)The THz-TDS technique based on tablet pressing method was used for the detection of hesperidin,nobiletin and tangeretin.The THz characteristic absorption spectra of three flavonoids in the range of 0.2-2 THz were selected for analysis,and the theoretical characteristic absorption peaks were calculated using the density functional theory(DFT)method.The results showed that hesperidin showed two distinctive characteristic absorption peaks at 1.63 and 1.89 THz respectively.Nobiletin had four characteristic absorption peaks at 0.42,0.66,0.97 and 1.51 THz respectively.In contrast,tangeretin did not show a distinct characteristic absorption peak,but a broad absorption peak could be observed at 1.03 THz.Comparing the vibration frequencies and modes calculated by experiment and DFT,it was found that the absorption peaks of the flavonoids were mainly caused by torsional vibration of dihedral angle.In addition,the analysis of flavonoids in different-age CRP using high performance liquid chromatography(HPLC)revealed that the flavonoid content showed fluctuations with increasing storage age.Modeling analysis was then carried out using the absorption coefficient spectra in the range of 0.2-1.8 THz with the corresponding flavonoid content values,and it was found that both the support vector machine(SVM)and least square support vector machine(LS-SVM)models did not predict well,and the correlation coefficient R_p of prediction sets was about 0.5.