基于SOM和SVM的遥感图像目标识别  

Remote Sensing Target Recognition Based on SOM and SVM

作　　者：张艳宁[1] 郑江滨[1] 候云舒[1] 赵荣椿[1] 

机构地区：[1]西北工业大学,陕西西安710072

出　　处：《系统工程与电子技术》2002年第7期9-11,共3页Systems Engineering and Electronics

基　　金：航空基础研究基金 ( 0 0F5 30 5 0 ) ;雷达信号处理重点实验室基金 ( 2 0 0 0JS0 1 4 1 HK0 311) ;西北工业大学高层人才资助基金( 5 2 1110 2 -380 0 -G14 10 7)资助课题

摘　　要：提出了一种基于自组织特征映射 (SOM)神经网络和支撑矢量机 (SVM)相结合的遥感图像目标识别方法。该方法首先利用SOM对目标进行聚类 ,然后应用SVM方法对其进行分类识别。最后将该方法应用于二值遥感图像的目标识别 ,与仅用支撑矢量机方法的识别结果进行了比较 ,结果表明 ,这一方法对二值遥感图像目标具有很好的分类识别效果 ,且训练时间大幅度缩短。An efficient method of remote sensing target recognition based on self-organizing map (SOM) neural network incorporating with support vector machines is proposed in this paper. Firstly, SOM is applied to cluster for the remote sensing target, then SVMs are used to classify each cluster. Finally, this method is used to recognize some binary remote sensing targets. Compared with SVM, the experiment results show that the method proposed will have better classification and recognition results for the binary remote images, which only contains object and background, and the tranning time is shorter than that by using the support vector machines only.

关 键 词：SOM SVM 遥感图像 目标识别 图像识别 支撑矢量机 自组织特征映射神经网络 

分 类 号：TN919.8[电子电信—通信与信息系统]