基于模糊语义的高分辨率SAR图像汽车检测算法

发布时间：2018-01-19 04:11

本文关键词： 高分辨率SAR 图像处理目标检测图像语义模糊隶属度　出处：《电子与信息学报》2017年04期 　论文类型：期刊论文

【摘要】：针对高分辨率SAR图像难于找到精确的背景杂波分布概率模型的问题,该文提出一种不需要背景杂波分布概率模型的高分辨率SAR图像自动检测汽车的新方法。该算法首先搜索场景中包含的亮区域和暗区域,其次采用模糊隶属度函数提取语义特征,筛选可能是汽车强散射区域的亮区域和可能是汽车遮挡区域的暗区域。再根据空间语义关系,对候选汽车强散射区域与候选汽车遮挡区域进行匹配,若匹配成功则计算它们源于同一辆汽车的隶属度。最后阈值选择高隶属度的目标进行合并输出。通过对Mini SAR图像进行汽车检测实验,表明该方法在不需要背景杂波分布概率模型的条件下仍然具有较高的检测率。
[Abstract]:It is difficult to find accurate probability model of background clutter distribution in high-resolution SAR images. In this paper, a new method for automatic vehicle detection from high-resolution SAR images without background clutter distribution probability model is proposed. The algorithm first searches the bright and dark areas contained in the scene. Secondly, the fuzzy membership function is used to extract the semantic features, and then the bright region which may be the strong scattering region of the vehicle and the dark area which may be the vehicle occlusion region are screened. Then according to the spatial semantic relationship. The strong scattering region of the candidate vehicle is matched with the occlusion region of the candidate vehicle. If the matching is successful, the membership degree of the same vehicle is calculated. Finally, the threshold value is selected to combine the target with high membership degree and the vehicle detection experiment is carried out on the Mini SAR image. It is shown that this method still has a high detection rate without the need of the probability model of background clutter distribution.
【作者单位】：西安电子科技大学计算机学院;西安电子科技大学智能感知与图像理解教育部重点实验室;
【基金】：国家973计划项目(2013CB329402) 国家自然科学基金(61573267);国家自然科学基金重大研究计划(91438201,91438103)~~
【分类号】：TN957.52
【正文快照】： 近些年来,许多城市通过安装设置大量的光学摄像头对通行车辆的数量和频度进行实时监控。但是光学摄像头对天气条件有较大的依赖性,遇到雨、雪、雾等天气或在夜晚等光线差的条件下会直接影响其成像效果。而合成孔径雷达成像技术则不受光照和气候等自然条件的制约,可以在能见度

【相似文献】