中药皂角刺中山皂角刺和野皂角刺掺伪模式识别研究

Discrimination and Quantification of Gleditisa sinensis Powder with Adulterants Using NIR Combined with Pattern Recognition Analysis

目的建立中药皂角刺中山皂角刺和野皂角刺掺伪的快速分析鉴别方法。方法利用近红外光谱法结合线性判别分析(LDA)、支持向量级(SVM)和人工神经网络法(BP-NN),建立中药皂角刺中山皂角刺和野皂角刺掺伪的快速无损鉴别模式识别模型;采用偏最小二乘(PLS)回归分析建立掺伪品中山皂角刺和野皂角刺伪品的掺伪量预测模型。结果对于皂角刺正品和其山皂角刺掺伪及野皂角刺掺伪品SVM法分类效果优于LDA和BP-NN法。谱段选择5000~4200 cm^-1、采用平滑-归一化法预处理数据方法时,SVM建模训练集、验证集和测试集分类准确率分别为100%、100%和96.4%。PLS回归模型结果显示,山皂角刺掺伪预测集rp、预测集均方根误(RMSEP)和偏差(bias)值分别为0.993、2.91%和-0.3303;野皂角刺掺伪预测集rp、RMSE和bias值分别为0.995、2.57%和0.3649。结论本实验建立的模式识别模型及回归方法能够准确快速判别皂角刺及山皂角刺和野皂角刺掺伪品并能较准确预测正品中山皂角刺和野皂角刺掺伪量。

OBJECTIVE To discriminate and quantify of Gleditsia japonica Miq.thorn(SZJ)and Gleditsia microphylla Gordon ex Y.T.Lee thorn(YZJ)in the Gleditsia sinensis Lam thorn(GST).METHODS Fourier transform near-infrared spectroscopy(FT-NIR)combined with linear discriminate analysis(LDA),support vector machine(SVM),as while as back propagation neural network(BPNN)algorithms were applied to construct the identification models.The SZJ and YZJ content in adulterated GST were determined by partial least squares regression(PLSR).RESULTS The SVM models performance best compared with LDA and BP-NN models for it could reach 100%accuracy in training and validation set for identifying authentic GST and GST adulterated with SZJ and YZJ based on the spectral region of 5000-4200 cm^-1 combined with SG+VN processing.The rp,RMSEP(the root mean standard error of prediction)and bias for the prediction by PLS regression model were 0.993,2.919%and-0.3303 for SZJ,0.995,2.57%and 0.3649 for YZJ,respectively.CONCLUSION Our results suggest that the combination of NIR spectroscopy and chemometric methods offers a simple,fast and reliable method for classifification and quantifification of SZJ and YZJ adulterants in the GST.