中国激光, 2019, 46 (6): 0614039, 网络出版: 2019-06-14
基于差分-主成分分析-支持向量机的有机化合物太赫兹吸收光谱识别方法 下载: 1053次
Terahertz-Spectral Identification of Organic Compounds Based on Differential PCA-SVM Method
太赫兹技术 光谱学 差分数据 主成分分析 支持向量机 terahertz technology spectroscopy differential data principal component analysis support vector machine
摘要
针对有机化合物的太赫兹时域光谱数据,提出了一种基于差分-主成分分析(PCA)-支持向量机(SVM)的有机化合物识别方法。基于物质样本的太赫兹时域信号计算得到太赫兹吸收光谱,对0.2~2.5 THz频率区间内的数据进行特征提取。在特征提取中,提出了基于差分数据的样本容量扩充方法,并结合PCA进行了特征的提取。利用SVM建立了提取的特征与物质类别对应关系的数学模型,并根据建立的模型对未知样本进行了识别研究。利用所提方法对15种有机化合物的太赫兹光谱数据进行了识别,正确识别率为93.33%。将所提方法与线性判别分析法及吸收峰频率-幅值法进行了对比,结果表明基于差分-PCA-SVM的有机化合物识别方法的正确识别率最高。
Abstract
This paper proposes a method for identifying organic compounds by applying a differential principal-component-analysis (PCA)-support-vector-machine (SVM) to the terahertz time-domain spectral data. First, the terahertz absorption spectrum is calculated according to the terahertz time-domain signal of the material sample; then, the features of the data in the frequency range of 0.2-2.5 THz are extracted. During the feature extraction, an expansion-of-sample-size method based on differential data is proposed and combined with the PCA method to achieve the feature extraction. Finally, the SVM is used to establish a mathematical model for the corresponding relationship between the extracted features and the material category, and the unknown samples are identified according to this model. The terahertz-spectral data of 15 organic compounds are identified using the proposed method, and the correct recognition rate is 93.33%. The experimental results show that the correct recognition rate of organic compounds by the proposed method is the highest when compared with those by the linear-discriminant analysis method and the absorption peak frequency-amplitude method.
刘俊秀, 杜彬, 邓玉强, 张建文, 祝海江. 基于差分-主成分分析-支持向量机的有机化合物太赫兹吸收光谱识别方法[J]. 中国激光, 2019, 46(6): 0614039. Junxiu Liu, Bin Du, Yuqiang Deng, Jianwen Zhang, Haijiang Zhu. Terahertz-Spectral Identification of Organic Compounds Based on Differential PCA-SVM Method[J]. Chinese Journal of Lasers, 2019, 46(6): 0614039.