基于多分类器融合的SAR图像自动目标识别方法  

Radar target recognition based SAR image and classifiers fusion

作　　者：胡利平[1] 刘宏伟[1] 吴顺君[1] 

机构地区：[1]西安电子科技大学雷达信号处理国家重点实验室,陕西西安710071

出　　处：《系统工程与电子技术》2008年第5期839-842,共4页Systems Engineering and Electronics

基　　金：教育部长江学者和创新团队支持计划(IRT0645);国家自然科学基金(60772140);国防预研项目资助课题

摘　　要：给出一种内存需求小、计算复杂度低且性能较好的合成孔径雷达自动目标识别(SAR ATR)方法。先对原始图像预处理获得目标及阴影,然后提取目标和阴影的形状描述子以及基于极化映射提取目标及阴影的形状特征、目标的强度分布特征,最后基于平均准则融合分类器对目标进行分类。基于美国运动和静止目标获取与识别(MSTAR)计划录取的数据的实验结果表明,所提融合的分类器可获得比单个分类器好的识别性能,并且利用阴影信息可大大提高识别性能。A SAR ATR method characterized with low memory requirement, low computational complexity and good recognition performance is proposed. Firstly, smoothed target and shadow images are segmented via SAR image preTProcessing. Secondly, shape information of target and its shadow, intensity distributed information of target are extracted based on polar mapping, and shape descriptors of target and its shadow are also extracted. Finally, SAR targets are classified by the combined classifier based on the average rule. Experimental results based on the MSTAR （moving and stationary target acquisition and recognition） data show that the combined classifier can obtain much better performance than the individual classifiers, moreover, the inclusion of the shadow features results is a significant improvement in recognition performance.

关 键 词：合成孔径雷达 自动目标识别 形状描述子 极化映射 多分类器融合 

分 类 号：TN958[电子电信—信号与信息处理]