基于NIRS技术和PCA-SVM算法6种树脂及其他类中药的快速鉴别

目的:利用近红外漫反射光谱(NIRS)法,结合主成分分析(PCA)和支持向量机(SVM)联用算法,建立6种树脂及其他类中药安息香(Benzoinum),琥珀(Succinum),没药(Myrrha),乳香(Olibanum),松香(Colophonium),天竺黄(Bambusaen Concretio Silicea)的NIR模式识别模型,用于该6味中药的快速鉴别。方法:收集上述6种中药样品,经性状鉴别和理化分析确定正品药材55批,粉碎成均匀粉末,在4 000~12 000 cm~(-1)光谱区,采集各样品粉末的NIR光谱,选取特征谱段9 000~5 400,5 000~4 000 cm~(-1)为建模谱段,分别采用矢量归一化法(vector normalization,VN),一阶导数法(first derivative,FD),二阶导数法(second derivative,SD)3种不同光谱预处理方法进行预处理并分别进行PCA降维。根据主成分空间散点图,优选最佳预处理方法。利用最佳预处理方法处理后光谱的PCA降维数据,建立SVM模式识别模型,SVM模型参数c和g采用网格搜索法结合五折交叉验证进行寻优。对比不同主成分数所建PCA-SVM模型的预测准确率,确定最佳的主成分数,最终建立6种中药NIR快速鉴别模型。结果:在9 000~5 400,5 000~4 000 cm~(-1)建模谱段,确定最佳光谱预处理方法为SD,SD预处理光谱PCA降维后,确定最佳主成分数为3个,累计贡献率达93.57%。经网格搜索法确定最佳SVM建模参数组为c=65 536,g=512。所建PCA-SVM模型对训练集和验证集样品预测正确率均达100%,模型五折交叉验证准确率亦达100%。结论:所建的6种中药NIR光谱PCA-SVM鉴别模型,预测准确率高,模型预测能力强,结合NIRS技术无损、快速的优点,该模型可用于上述6种中药的无损、快速鉴别。

Rapid Identification of 6 Kinds of Traditional Chinese Medicines Containing Resins and Other Components Based on Near Infrared Refectance Spectroscopy and PCA-SVM Algorithm

Objective: To establish the pattern discernment model with the use of near-infrared reflectance spectroscopy (NIRS), principal component analysis (PCA) and support vector machine (SVM) algorithms for rapid identification of Benzoinum, Succinum, Myrrha, Olibanum, Colophonium and Bambusae Concretio Silicea, all of which are the traditional Chinese medicines (TCMs) containing resins and other components. Method: According to morphological identification and conventional physical and chemical analysis on the above 6 kinds of samples collected form major national medicinal materials markets, a total of 55 batches of samples were verified as genuine medicinal herbs. These samples were smashed into uniform powders, and the NIR spectra of sample powder were scanned at 4 000-12 000 cm^-1. The characteristic bands at 9 000-5 400, 5 000-4 000 cm^-1 were used to establish models for pre-treatment by vector normalization (VN), first derivative (FD) and second derivative (SD) respectively, and in addition, PCA dimension reduction was also conducted. The best pre-treatment method was selected according to the three-dimensional scatter diagram of principal components, and the PCA dimension reduction data after best pre-treatment method were used to establish the pattern discernment models based on the SVM algorithm. The SVM model parameters c and g were optimized by using grid search method combined with 5-fold cross validation method. By comparing the forecast accuracy of PCA-SVM models established with different number of principal components, the best number was optimized and finally, NIR rapid identification model was established for 6 Chinese herbs. Result: At 9 000-5 400, 5 000-4 000 cm^-1, SD was determined as the best pre-treatment method, and after SD pre-treatment for PCA dimension reduction, the best number of principal components was determined as 3, with an accumulative contribution rate of 93.57%. According to the grid search method, c=65 536, g=512, which were the optimum parameters of the SVM model. In the established PCA-SVM models, the forecast accuracy was 100% for both training set and verification set, and the forecast accuracy was also 100% for 5-fold cross validation method. Conclusion: The NIRS identification model based on PCA-SVM had high forecast accuracy and strong prediction ability, which can be used for rapid, nondestructive and accurate identification of above 6 kinds of TCMs when it was used in combination with the advantages of NIRS.