一种快速的SAR影像相干斑抑制算法

A fast SAR image speckle suppression algorithm

提出了一种快速合成孔径雷达(SAR)影像相干斑抑制算法.首先,利用分裂增广拉格朗日收缩法,将总变分正则化去噪模型转换为双参模型.引入交替迭代法将双参模型分解为含非二次保真项和含不可微正则项的两个单参模型.其次,采用牛顿迭代法求解含非二次保真项模型,对偶方法求解含不可微正则项模型.最后,采用快速迭代阈值收缩法(FISTA)优化对偶方法的求解速度.实验选择武汉某地区"高分3号"影像和TerraSAR-X影像.结果表明:提出算法能较好的抑制SAR影像相干斑,比总变分正则化方法提升了约两倍的效率.

A fast speckle suppression algorithm for synthetic aperture radar(SAR) images was proposed. Firstly, using split augmented Lagrange shrinkage method, the total variation regularization denoising model was transformed into a two-parameter model. The alternating iteration method was introduced to decompose the two-parameter model into two single-parameter models with non-quadratic fidelity term and non-differentiable regularity term. Secondly, Newton iteration method was used to solve the model with non-quadratic fidelity term, and dual method was used to solve the model with non-differentiable regularity term. The fast iterative shrinkage-thresholding algorithm(FISTA) was used to optimize the speed of dual method. Experiments on "GF 3" image and TerraSAR-X image of a certain area in Wuhan show that the proposed algorithm can suppress SAR image speckle better, which improves the efficiency by about two times compared with the total variation regularization method.