LIU Jiang-yong1, WANG Da-ming2
(1 College of Information and Engineering, Xiangtan University, Xiangtan 411105, Hunan, China；2 Representative Office, Air Force in Luoyang Region, Luoyang 471009, Henan, China)
Abstract:
Based on the support vector machines (SVMs), a fast hyperspectral data classification method is proposed. First, a feature reduction method based on principal component analysis (PCA) and Bhattacharyya distance is used. Then, the binary tree of SVMs (BTS) is adopted to reduce the number of binary SVMs for one classification. Finally, the simplifying support vector solution is utilized to further reduce the number of support vectors (SVs). The experiment adopted real hyperspectral data set. Compared with the other four methods, the results demonstrate that the proposed method can achieve a fast classification effectively.
KeyWords:
hyperspectral data classification; support vector machine; feature reduction; computational complexity
