基于支持向量机的中药太赫兹光谱鉴别  

Chinese Traditional Medicine Recognition by Support Vector Machine(SVM) Terahertz Spectrum

作　　者：陈艳江[1] 刘艳艳[1] 赵国忠[1] 王卫宁[1] 李福利[1] 

机构地区：[1]首都师范大学物理系,北京100048

出　　处：《光谱学与光谱分析》2009年第9期2346-2350,共5页Spectroscopy and Spectral Analysis

基　　金：北京市教育委员会科技发展计划面上项目(KM200710028004)资助

摘　　要：文章将支持向量机用于中药材太赫兹光谱识别。利用太赫兹光谱系统测得三组相似中药炙甘草和生甘草、南柴胡和北柴胡、山豆根和北豆根的太赫兹光谱,傅里叶变换后得到它们的吸收系数作为分类鉴别的特征数据。用线内积函数、多项式内积函数和径向基内积函数分别构建三种不同的支持向量机,并建立误差反传神经网络(BP神经网络),分别用支持向量机和BP神经网络对中药的特征数据进行鉴别。识别结果比较表明,支持向量机在小样本情况下对中药两分类识别的效果明显超过BP神经网络。Identification is very important for the development of Chinese traditional medicines. In recent years, rapid progress in ultrafast laser technology provides a steady and available source for,terahertz pulses generation, which greatly promotes the development of THz spectroscopy and imaging＇technique. SVM is amethod for recognition of two kinds of samples. Appling SVM to the identification of Chinese traditional medicines through THz spectrum is a new way. The experiment on three groups of Chinese traditional medicines （zhigancao and shengancao, nanchaihu and beichaihu, shandougen and beidougen） was studied. The THz frequency spectrum and absorptivity were obtained and used to construct the feature space of Chinese traditional medicines. Three kinds of SVM were build, which used three kinds of kernel functions. By comparison, a model of BP artificial neu- ral network was constructed. The result of using three kinds of SVM and BP artificial neural network to identify the Chinese traditional medicines showed that both methods have good prediction ability, but obviously the effect of SVM is better than BP artificial neural network for small samples. Using SVM in terahertz spectrum is a efficacious way for classification of Chinese traditional medicines.

关 键 词：太赫兹光谱识别 光谱学 支持向量机 

分 类 号：O433.4[机械工程—光学工程]