关键词: 太赫兹时域光谱, 流形学习, 支持向量机, 核方法

Abstract: Terahertz time-domain spectroscopy is a new spectrum detection technology, and applied to various research fields：national security and defense, biomedical and food quality inspection. The classification of THz spectrums is one crucial domain of THz measurement. Because of the influence of noise, the result of traditional pattern classification methods is not satisfactory. Manifold learning and SVM are both effective approaches for non-linear pattern analysis problems and are also new research focus in current machine learning community. A new algorithm based on manifold learning and SVM is proposed in this paper. Compared to traditional SVM, this algorithm can improve the classification performance of SVM. Experimental result demonstrates that this algorithm can be applied to medical identification. It provides an effective method for the detection and identification of medical by terahertz spectroscopy technology.

Key words: TeraHertz Time-Domain Spectroscopy（THz-TDS）, manifold learning, Support Vector Machine（SVM）, kernel method