基于稀疏表示模型的SAR图像目标检测算法

针对合成孔径雷达（synthetic aperture radar，SAR）图像可视性差、目标区域小以及特征不明显等特性对目标检测造成的困难，将稀疏表示模型应用于SAR图像目标检测，提出一种基于稀疏表示模型的SAR图像目标检测算法。首先，利用K—SVD算法训练样本提取对样本最具描述能力的SIFT特征形成字典；其次，通过将进化机制和稀疏表示结合，逐步提取整幅图像中含有目标的图像块；最后，输出稀疏表示误差小于阈值的图像块的位置作为目标检测的结果。实验结果表明，与传统目标检测算法相比，该算法在检测率和运行效率方面均有一定的提高，取得了较好的效果。

Objects detection in SAR images via sparse representation moael

Considering detection difficulties caused by poor visibility, small target area and indistinctive features of synthetic aperture radar （SAR） images, we apply the sparse representation model to targets detection in SAR images, and propose a new algo- rithm to work on this problem. In first instance, employing K-SVD algorithm, a dictionary that consists of SIFT features can be trained from samples. Secondly, combining the evolution idea with sparse representation, patches that may contain targets can be extracted gradually. Finally, the location of image patches whose reconstruction error is below the predefined threshold is exported as detection result. Experimental results demonstrate the improvement of our proposed algorithm on detection rate as well as processing efficiency.