相关向量机的高光谱影像分类研究.pdfVIP

39 6 Vol. 39, No. 6 2010 12 Acta Geodaetica et Ca rtographica Sinica Dec. , 2010 :2010) 1, 2 1 3 1 杨国鹏 , 余旭初 , 周 欣 , 张鹏强 1. , 450052; 2; , 100850; 3. , 450052 Research on Relevance Vector Ma chine for H perspectral Imager Classification 1,2 1 3 1 YANG Guopeng , YU Xuchu , ZHOU Xin , ZHANG Pengqiang 1. Institute of Surve ing and Mapping, Information Engineering Universit , Zhengzhou 450052, China; 2. Intelligence Institute of A irforce s Equipment Academ , Beijing 100850, China; 3. Institute of Surve ing a nd Ma pping, Information Engineering Universit , Zhengzhou450052, China Abstract : Though the support vector machine has been successfull applied in h perspectral imager classification, it ha s also several limitations. Releva nce vector ma chine ( RV M) is a sparse model in the Ba esian framework, its ma thematics model doesn t have regularization coefficient and its kernel funct ions don t need to sat isf Mercer s condition. RVM presents the good genera lization performance, and its predictions a re proba bilistic. In this pa per, we firstl a na l sis the disa dva nta ges of the support vector machine for h perspect ral ima ger classifica tion, and then a h perspectral imager classification method based on the relevance machine is brought forwa rd. We intro duce t he spa rse Ba esia n classification model, rega rd the RV Mlearning as the maximization of margina l likelihood, and select t he fast sequentia l sparse Ba esian lea rning algorithm. Throught he experiments of PHI a nd OMIS ima ge ries, the a dva nta ges of the relevance machine used in h perspectral imager cla ssifica tion are given out. Ke words : h perspectral imager ; sparse Ba esian model; releva nce vector ma chine; support vector ma chine :从分析支持向量机用于高光谱影像分类时存在的不 出发, 提出一种基于相关向量机的高光谱影像分类方法 在介