基于鲁棒主成分分析的SAR舰船检测
Ship detection in SAR images by robust principle component analysis

针对单极化SAR舰船目标恒虚警检测无标准杂波模型可选,且多目标情况下易发生目标检测不完整和弱目标丢失的问题,该文提出一种基于鲁棒主成分分析(robust principle component analysis, RPCA)的舰船检测方法,通过利用SAR图像中内在的海面低秩属性和舰船目标的稀疏属性,借助推导的增量Lagrange乘子算法,将SAR图像分解为低秩图像、噪声图像(两者之和对应海面)和稀疏图像(对应舰船)的和,从而一次性实现目标检测和杂波抑制,不依赖任何杂波模型和检测统计量。仿真实验验证了增量Lagrange乘子算法的有效性。实测数据处理实验中与平均单元恒虚警检测法和均方误差恒虚警检测法进行了对比,结果表明该方法可以正确从海杂波中检测出舰船目标,具有良好的鲁棒性。
Abstract：Single polarization SAR ship detection using a constant false alarm rate (CFAR) detector does not have a standard clutter model, so the system often gives incomplete targets and misses weak targets in multitarget detection. A ship detection method was developed based on robust principle component analysis (RPCA) to improve the detection. This method leverages the intrinsic properties of SAR images that the sea area is approximately low rank and there are few ships. SAR images can be decomposed into the sum of a low rank component, a noise component and a sparse component via RPCA, with the sum of the first two corresponding to the sea surface and the third corresponding to ships. Thus, ship detection and clutter suppression are achieved in one step without a clutter model or statistics. The augmented Lagrange multiplier method for RPCA is verified by simulations. For comparison, cell averaging CFAR (CA-CFAR) and mean square error CFAR (MSE-CFAR) are also used. Tests with real data show that this method correctly detects ships from sea clutter with robust detection performance.