Study Of Spectrum Fingerprint Of Dendrobium Officinale And Similar Species In Yunnan Province

The quality of herbal medicine is affected by several factors,such as the botanical origins,harvesting period,geographical origins,botanical parts,et al.These factors lead to the fluctuated quality.Spectroscopy is a green and environmental protection method,which can provide a rapid,effective,and convenient technology reference for quality control of herbal medicine combined with content data of index components of chromatography and statistical chemometrics methods.In the thesis,Dendrobium plants were used as research object for investigating their botanical origins,harvesting period of,geographical origins,and botanical parts D.officinale,using ultraviolet-visible spectroscopy(UV-Vis),attenuate total reflection-Fourier transform infrared spectroscopy(ATR-FTIR),near infrared spectroscopy(NIR),and high performance liquid chromatography(HPLC).In addition,the analytical methods include three unsupervised(principal component analysis,PCA;hierarchical cluster analysis,HCA;t-distributed stochastic neighbor embedding,T-SNE),four supervised methods(k-nearest neighbor,KNN;partial least squares discriminant analysis,PLS-DA;grid search-support vector machine,GS-SVM;random forest,RF),and support vector machine regression(SVMR)for deep analysis of spectral characteristics of Dendrobium.The discrimiantion results of D.officinale from similar species indicated that low-level fusion strategy was the best fusion method for discrimination of different Dendrobium species.GS-SVM and PLS-DA were the optimal discrimination models with 100% accuracy rates in calibration and validation sets.Herein,the important variables combined with PLS-DA could improve the robustness of the model.The discrimination results of harvesting time in D.officinale showed that combined important variables from stems and leaves spectra could improve the discrimination accuracy with 94.44% and 97.92% in calibration and validation sets,respectively.The advantage data fusion from different botanical parts may be used for the species discrimiantion using spectra dataset.Furthermore,the SVMR model was established using contents of quercetin and erianin and spectra.The results indicated that the optimal pretreatment methods combined with grid search could improved the predictive results of regression model.The systematic method for quality assessment of D.officinale could provide theoretical basis for rational resource development of the specie.