| As a valuable traditional Chinese medicine(TCM),Panax notoginseng(PN)has an increasing market demand because of its great medicinal value.However,due to the harsh growth conditions and the lack of resources,the price of PN is keeping rising.PN are often adulterated with some low-cost counterfeits with similar appearances,such as Rhizoma curcumae(RC),Curcuma longa(CL)and Rhizoma alpiniae offcinarum(RAO)on the market.It is difficult to identify its similar TCMs and quantify adulteration of PN due to the complex composition of TCMs.Near-infrared(NIR)spectroscopy and Ultraviolet-visible(UV-vis)diffuse reflectance spectroscopy are widely used for qualitative and quantitative analysis of complex samples,which are non-destructive,fast,and simple operation.In this thesis,the feasibility of identification of PN and its similar TCMs and quantification adulteration of PN using NIR spectroscopy and UVvis diffuse reflectance spectroscopy combined with chemometrics are investigated.The specific research contents are as follows.1.NIR spectroscopy combined with chemical pattern recognition are used to distinguish PN and its similar TCMs.25 PN,28 RC,28 CL and 28 RAO were purchased from the Tianjin pharmacy,which were a total of 109 samples.The NIR spectra were measured directly after each sample was ground and passed through a 120 mesh sieve.The Kennard and Stone(KS)algorithm was used to divide 109 samples into a training set with 70 samples and a prediction set with 39 samples.Then hierarchical cluster analysis(HCA),partial least squares-discriminant analysis(PLS-DA),artificial neural networks(ANN),support vector machine(SVM)and extreme learning machine(ELM)were used to build model based on the optimal parameters using training set and compare their discrimination capabilities for prediction set samples.The results showed that the PLS-DA and SVM can achieve 100% classification accuracy.2.NIR spectroscopy combined with multivariate calibration is used to quantify the component of PN in multiple adulterated PN.Three datasets of binary adulterations,three datasets of ternary adulteration and one dataset of quarternary adulteration of PN adulterated with RC,CL and RAO were designed respectively,which were a total of seven datasets.Every dataset was divided into a training set of 2/3 samples and a prediction set of 1/3 samples by overall order and local random.Five models of principal component regression(PCR),support vector regression(SVR),partial least squares regression(PLSR),ANN and ELM were established using training set samples.The prediction accuracy and efficiency were evaluated and compared for each dataset separately.PLSR was an optimal modeling method of every corresponding prediction set by comprehensive consideration of prediction accuracy,overfitting and efficiency.In addition,considering the effect of preprocessing on the prediction results,it was found that appropriate preprocessing method can further improve the prediction accuracy of PLSR model.3.UV-vis diffuse reflectance spectroscopy combined with chemical pattern recognition are used to distinguish PN and its similar TCMs including RC,CL and RAO.First,UV-vis diffuse reflectance spectra of 109 pure TCMs were collected.109 samples were divided into a training set with 70 samples and a prediction set with 39 samples according to the KS algorithm.Then the parameters were optimized and the four classification models of HCA,PLS-DA,ANN and SVM were established based on the optimal parameters using the training set samples.It demonstrated that PLS-DA was an accurate identification method for prediction set and the prediction accuracy rate can reach 100%.4.UV-vis diffuse reflectance spectroscopy combined with multivariate calibration are used to quantify the quaternary adulteration of PN.Firstly,75 samples of quaternary adulteration using PN,RC,CL and RAO according to different mass percentage were designed and their UV-vis diffuse reflectance spectra were collected.The KS algorithm was used to divide 75 adulteration into a training set with 50 samples and a prediction set with 25 samples.Secondly,the methods of Monte Carlo Cross-Validation(MCCV)combined with F-test were used to determine the optimal parameter of PLSR.And then the models of PLSR for the four components were established using training set based on the optimal factors.At the same time,the effects of six preprocessing methods and three variable selection methods on the modeling were considered.It was found that combining the appropriate pre-processing and variable selection methods can improve the prediction performance of the model.Meanwhile,each component of the prediction set was predicted in adulterate PN and the correlation coefficient Rp were all above 0.98. |
PDF Full Text Request