基于SAR图像样本的本征维数检测人造目标  

Man-made targets detection based on intrinsic dimension of SAR image samples

作　　者：张之光[1] 雷宏[1] Zhang Zhiguang Lei Hong(Institute of Electronics, Chinese Academy of Sciences, Beijing 100190,China)

机构地区：[1]中国科学院电子学研究所,北京100190

出　　处：《电子测量技术》2016年第9期34-39,共6页Electronic Measurement Technology

摘　　要：人造目标检测是合成孔径雷达(SAR)自动目标识别(ATR)系统的重要功能,当前适用于复杂背景中人造目标检测的手段较为有限,为此提出了一种基于多方位角SAR图像样本本征维数的检测方法。该方法不依赖于图像强度,而是估计出各感兴趣区域(ROI)大量不同方位角图像样本在高维度空间中的本征维数,根据本征维数定量衡量ROI内目标更趋近人造或是自然,由此检测出人造目标。采用Gotcha项目实测数据将本文方法与当前以分形维数检测人造目标的方法进行了对比,实验结果显示本征维数方法具有不依赖于图像强度的特点,并且在目标形态刻画上比分形维数方法取得更好效果。Man‐made targets detection is a crucial function of synthetic aperture radar （SAR） automatic targets recognition （ATR） systems . At present , there exist sort of limited methods which are applicable to complex background .To treat this case ,a method based on the intrinsic dimension of multi‐aspect angle image samples is proposed .This method doesn’t count on intensity of the image ,whereas it estimates the intrinsic dimension of images samples of the region of interest （ROI） in high dimensional space ,which are obtained from different aspect angles .The measurement of artificiality and naturalness is performed with the intrinsic dimension of ROI .In this way man‐made targets are detected . The real data from Gotcha project are adopted to compare our method with that of fractal dimension to detect man‐made targets .The experiment result shows that intrinsic dimension has a merit that it doesn’ t count on image intensity and it performs better than the fractal dimension method in depicting the shape of targets .

关 键 词：合成孔径雷达 人造目标检测 本征维数 分形维数 

分 类 号：TN959.1[电子电信—信号与信息处理]