基于SVM的多源遥感影像分类研究  

Research on multi-source remote sensing image classification based on SVM

作　　者：贾萍[1] 李海涛[2] 林卉[3] 顾海燕 韩颜顺[2] 

机构地区：[1]国土资源部信息中心,北京100812 [2]中国测绘科学研究院摄影测量与遥感研究所,北京100039 [3]徐州师范大学国土信息与测绘工程系,江苏徐州221116

出　　处：《测绘科学》2008年第4期21-22,7,共3页Science of Surveying and Mapping

基　　金：国家高技术研究发展计划(编号:2007AA122151)

摘　　要：本文通过分析单源遥感影像分类的现状和困难,以SAR和SPOT-5影像为实验数据,提出了基于支持向量机(Support Vector Machine,SVM)理论的多源遥感影像分类方法。研究结果表明,本文的方法能够有效地解决单源影像信息分类效果破碎的问题,正确识别地物,对高维输入向量具有高的推广能力,正确率达到94.97%,比多源影像的最大似然分类(Maximum Likelihood Classification,MLC)方法正确率更高。On the basis of analyzing the actuality and difficulty of single-source image classification, a method of applying Support Vector Machine to multi-source Remote Sensing image classification is presented in this paper where SAR and SPOT-5 data are used. The results show that this method can solve the image classification fragmentation which is based on the single-source data, recognize objects accurately, has the good generalization ability with the high dimension vector, and has total accuracy of about 94. 97%. It has more accuracy than the Maximum Likelihood Classification （MLC） method using multi-source data.

关 键 词：支持向量机 多源影像 最大似然分类 精度评价 

分 类 号：TP75[自动化与计算机技术—检测技术与自动化装置]