基于改进的S-NMF方法的全极化SAR船舶检测

全极化SAR影像应用于海上船舶检测,如何在不降低检测率的同时,利用极化信息降低虚警率,是一个值得研究的问题。根据SAR极化矩阵3个特征值的非负性和稀疏性,提出了改进的非负矩阵分解S\|NMF(Sparseness-Nonnegative Matrix Factorization)方法:将最大和次优特征值用于NMF分解,因为两者占有目标97%以上的能量,能够保证最大的检测率,分解后所得结果为检测结果I;而强海杂波、“目标鬼影”等虚警在第三特征值上表现较弱,利用第三特征值与结果I相乘,可进一步强化目标弱化虚警。然后采用OS-CFAR算法对相乘后的图像进行检测,输出最终结果。最后用带有AIS数据的Radarsat-2全极化数据对该方法进行验证,并与SPAN方法、HV通道、PWF方法的检测结果进行对比,结果表明:该方法不但能够正确地检测出船舶目标,而且能够有效降低虚警。

Ship Detection from Fully Polarimetic SAR Images based on Improved S-NMF Algorithm

It’s worth to investigate that the false alarm is reduced with polarimetic information while keeping the detection ratio when the fully polarimetic SAR images used for ship detection.According to the nonnegative and sparseness characteristics of eigenvalues,an improved S-NMF (Sparseness-Nonnegative Matrix Factorization) based on eigenvalues is proposed in this paper,which is used for ship detection for SAR images and reduce false alarms effectively.We use the first two eigenvalues to consist matrix for NMF decomposition since they take most of the energy of the targets and could maintain the shapes of the targets,while the sparseness is decided by the clutter distribution,the decomposition result is regarded as Result I which contains all the ships with some false alarms;while the third eigenvalue which could get rid of the false alarms,such as “ghosts” caused by the ship movement,noi,etc.is used to multiply the Result I to get the Result II which could achieve higher TCR(Target-Clutter Ratio).Then the OS-CFAR method is used for ship detection,and become the final detection results.In this paper,this detection algorithm is validated through Radarsat-2 fully polarimetic SLC images of South Sea of China and the corresponding AIS data,and compared with the SPAN method,HV channel image and PWF method.The detection and comparison results show that the proposed method could not only detection all the ships correctly,but also reduce the false alarms effectively,and has high robustness for non-affected by changes of the detection window.