基于近似观测的加权L1压缩感知SAR成像

Approximated Observation-Based Weighted L1 Compressed Sensing SAR Imaging

近些年,压缩感知(CS)理论已经被应用于合成孔径雷达(SAR)成像。传统的CS—SAR成像需要消耗很高的计算机内存.为了减少计算机内存消耗,基于近似观测的CS-SAR成像模型被提出。已有的基于近似观测的CS-SAR成像模型使用Lq(0≤q≤1)正则化项来稀疏约束成像结果,当q越小时,得到的解越稀疏;但是当q=0时,该优化问题就变成了NP难问题。文中提出了基于近似观测的加权L1-CS-SAR成像模型,加权L1正则化既能够很好地逼近L0正则化,又能够避免NP难问题。进一步,我们针对该成像模型提出了相应的迭代加权阈值算法,仿真结果证明了所提出的成像算法的性能优于已存在的迭代阈值算法。

Recently,the theory of compressed sensing( CS) has been applied in synthetic aperture radar( SAR) imaging. Since the conventional CS-SAR imaging requires high memory cost,the approximated observation-based CS-SAR imaging has been proposed to reduce the computer memory cost. The existing CS-SAR imaging based on approximated observation utilizes Lq(0≤q≤1) regularization which can contribute to generating sparse solutions. The smaller the q,the sparser the solutions yielded by Lq regularization. However,in the case of q = 0,the problem is NP-hard. In this paper,approximated observation-based weighted L1-CS-SAR imaging model is proposed,the regularization of which not only achieves an approximation of L0 regularization,but also avoids an NP-hard problem. Furthermore,the iterative weighted thresholding algorithm is proposed to solve the proposed imaging model. Simulation results demonstrate the superiority of the proposed algorithm over the existing iterative thresholding algorithms.