车载前视GPSAR序列图像浅埋目标检测新方法  

A Novel Detection Method of Shallow Buried Objects with Sequence Images of the Vehicle-Mounted Forward-Looking GPSAR

作　　者：杨延光[1] 宋千[1] 周智敏[1] 金添[1] 张汉华[1] 

机构地区：[1]国防科技大学电子科学与工程学院,长沙410073

出　　处：《电子与信息学报》2009年第6期1292-1297,共6页Journal of Electronics & Information Technology

摘　　要：由于浅埋小目标的雷达截面较小、埋设环境复杂,导致单帧检测结果中存在大量虚警,该文提出一种"预筛选-交替后向跟踪-多帧确认"目标检测方法。首先利用CFAR检测、形态滤波和聚类分析对图像进行预筛选;然后利用当前距离多视图像对应的聚类中心标记待跟踪的潜在目标,采用点跟踪搜索和块相关检测相结合的交替后向跟踪获得潜在目标轨迹;最后计算潜在目标轨迹的加权历史置信度,进行目标确认,剔除自然杂波。实测数据处理结果表明:该方法可快速、稳健地实现目标检测,明显减少虚警。Since the small Radar Cross Section （RCS） and complicated buried environments of shallow buried targets will lead to large numbers of false alarms in the detection results of single frame image, a novel detection method called as ＂prescreening-alternately backward tracking-multiframe confirmation＂ is proposed in this paper. Firstly, the prescreening of images is made by the Constant False Alarm Rate （CFAR） detector, then followed by a morphological filter and clustering analysis. Secondly, the clustering centers of the current frame range multi-look image are utilized to mark the potential targets for tracking, and a new alternately backward tracking method is used to obtain the potential object tracks. Finally, the tracks are employed to compute the weighted historical confidences, which are exploited to eliminate the natural clutters and confirm targets. Experimental results of the real data show that the proposed approach can acquire the regions of interest fast and robustly, reduce the false alarms obviously.

关 键 词：前视地表穿透雷达 地雷检测 交替后向跟踪 块相关检测 历史置信度 

分 类 号：TN957.52[电子电信—信号与信息处理]