论文部分内容阅读
利用近红外光谱鉴别云南不同产区药用真菌茯苓。采用光谱标准偏差法(SSD)和蒙特卡罗无信息变量消除法(MC-UVE)逐步筛选光谱信息,利用K-S算法将白茯苓和茯苓皮划分训练集和验证集,并结合偏最小二乘判别分析法(PLS-DA)分别构建不同产区白茯苓和茯苓皮的分类模型,进一步建立不同产区茯苓的Fisher判别方程。结果表明:(1)通过主成分-马氏距离(PCA-MD)分析,白茯苓和茯苓皮的近红外光谱在主成分得分空间内呈现出较大差异,构建不同产区茯苓鉴别模型应将白茯苓及茯苓皮分开。(2)最优主成分数为5时,采用SSD筛选的886个变量(7 501.74–4 088.35cm~(-1))构建的白茯苓和茯苓皮的分类模型,其R~2、RMSECV和RMSEP分别为0.986、0.988;0.320、0.283;0.425、0.395;采用MC-UVE法分别筛选出白茯苓和茯苓皮光谱信息(34个、22个变量)建立的分类模型,其R~2、RMSECV和RMSEP分别为0.993、0.991;0.224、0.255;0.298、0.355。采用MC-UVE结合PLS-DA法建立的分类模型,有效降低了冗余信息,白茯苓和茯苓皮的R~2均有所提升,RMSECV和RMSEP均有所降低,预测正确率分别由85.71%和83.33%,提高至100%。(3)进一步采用逐步判别分析法筛选出白茯苓(6个)和茯苓皮(4个)光谱变量,建立Fisher判别方程,回代验证正确率均大于85.7%,交叉验证正确率均大于66.7%,表明所建立的Fisher判别方程能较好地鉴别不同产区的茯苓。
Identification of Medicinal Fungus Poria in Different Areas of Yunnan by Near Infrared Spectroscopy. Spectral information was gradually screened by Spectral Standard Deviation Method (SSD) and Monte-Carlo Elimination of Information Variables (MC-UVE). The KS algorithm was used to classify the Poria and Poria cocos skin in the training set and the verification set. Combining Partial Least Squares (PLS-DA) were used to construct taxonomic models of Poria cocos and Poria cocos from different areas to further establish Fisher discriminant equations of Poria cocos in different areas. The results showed that: (1) Near infrared spectra of Poria and Poria cocos skin showed significant differences in the principal component score space by PCA-MD analysis. Poria Cocos and tuckahoe separated. (2) When the optimal principal component was 5, the classification model of Poria cocos and Poria cocos with 886 variables selected by SSD (7 501.74-4 088.35 cm -1) was established. The R 2, RMSECV and RMSEP were 0.986,0.988; 0.320,0.283; 0.425,0.395 respectively. The classification models established by MC-UVE method were used to separate the spectral information of Poria and Poria cocos (34 and 22 variables) RMSEP were 0.993,0.991; 0.224,0.255; 0.298,0.355, respectively. The classification model established by MC-UVE combined with PLS-DA method reduced the redundant information effectively. R ~ 2 of both Poria cocos and Poria cocos skin increased, and both RMSECV and RMSEP decreased. The accuracy of prediction was 85.71% And 83.33%, increased to 100%. (3) Stepwise discriminant analysis was used to screen the spectral variables of Poria cocos (6) and Poria cocos (4), Fisher discriminant equations were established, and the correct rates were all more than 85.7% by back-generation verification. The correct rates of cross validation were more than 66.7% , Indicating that the established Fisher discriminant equation can better identify Poria in different producing areas.