基于模糊语义的高分辨率SAR图像汽车检测算法

Detecting Cars in VHR SAR Images with a Fuzzy Semantic Algorithm

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。

It is hard to select a probability distribution model for very high resolution SAR images. This paper presents a novel method for the automatic detecting of cars from VHR SAR image without the probability distribution model. The proposed method starts with searching bright regions and dark regions by the gray feature. Subsequently, the fuzzy membership is employed to extract the semantic features of car from bright regions and dark regions. The potential scattering surface and shadow are matched and calculated with the spatial semantic relationship. Finally, the cars are selected from the matching. The efficiency of the proposed method is demonstrated by experiment which shows it still has high detection rate without the probability distribution model.