基于主成分分析和支持向量机的太赫兹光谱冰片鉴别

利用主成分分析方法结合支持向量机建立了太赫兹时域光谱冰片种类鉴别模型。冰片是一些常用中成药的重要成分，由于其来源多、真假易混淆，在制药和交易环节，迫切需要快速、简便、准确的检测、鉴别方法。太赫兹时域光谱技术是利用太赫兹脉冲表征物质性质的一种新兴光谱技术。实验使用透射式太赫兹时域光谱系统分别获得了艾片、合成冰片和梅片三种冰片在0.2～2 THz之间的吸收谱线。通过主成分分析，做出了第一、第二主成分二维得分图以及第一、二、三主成分三维得分图，两者对三种不同种类冰片都具有很好的聚类效果。用前十个主成分的得分值矩阵代替原光谱数据，通过对三种冰片的60组样本训练，对未知的60组样本鉴别，建立了四种不同核函数的支持向量机模型。对比结果表明，径向基核函数构建的支持向量机对三种冰片的分类鉴别准确率均为100%，由此我们确定选择具有径向基核函数的支持向量机建立冰片种类的鉴别模型。此外，在含噪情况下，四种核函数SVM获得的总分类准确率都在85%以上，说明支持向量机具有很强的泛化能力。主成分分析结合支持向量机方法对冰片太赫兹光谱具有很好的分类和鉴别效果，为冰片等中成药剂的种类鉴别提供了一种新思路。

Discrimination of Varieties of Borneol Using Terahertz Spectra Based on Principal Component Analysis and Support Vector Machine

In the present paper, the terahertz time-domain spectroscopy (THz-TDS) identification model of borneol based on principal component analysis (PCA) and support vector machine (SVM) was established. As one Chinese common agent, borneol needs a rapid, simple and accurate detection and identification method for its different source and being easily confused in the pharmaceutical and trade links. In order to assure the quality of borneol product and guard the consumer’s right, quickly, efficiently and correctly identifying borneol has significant meaning to the production and transaction of borneol. Terahertz time-domain spectroscopy is a new spectroscopy approach to characterize material using terahertz pulse. The absorption terahertz spectra of blumea camphor，borneol camphor and synthetic borneol were measured in the range of 0.2 to 2 THz with the transmission THz-TDS. The PCA scores of 2D plots (PC1×PC2) and 3D plots (PC1×PC2×PC3) of three kinds of borneol samples were obtained through PCA analysis, and both of them have good clustering effect on the 3 different kinds of borneol. The value matrix of the first 10 principal components (PCs) was used to replace the original spectrum data, and the 60 samples of the three kinds of borneol were trained and then the unknown 60 samples were identified. Four kinds of support vector machine model of different kernel functions were set up in this way. Results show that the accuracy of identification and classification of SVM RBF kernel function for three kinds of borneol is 100%, and we selected the SVM with the radial basis kernel function to establish the borneol identification model. In addition, in the noisy case, the classification accuracy rates of four SVM kernel function are above 85%, and this indicates that SVM has strong generalization ability. This study shows that PCA with SVM method of borneol terahertz spectroscopy has good classification and identification effects, and provides a new method for species identification of borneol in Chinese medicine.