High performance parallel implementation of compressive sensing SAR imaging

The compressive sensing (CS) theory has been applied to SAR imaging systems in many ways. And it shows a significant reduction in the amount of sampling data at the cost of much longer reconstruction time. In this paper, we investigate the development and optimization of Iterative Shrinkage/Thresholding (IST) algorithm applying to CS reconstruction of SAR images on two parallel architectures, standard vectorized multi-core processors (e.g., quad- core CPUs) and graphics processing units (GPUs). Meanwhile, we modify the IST algorithm according to the characteristic of SAR images to obtain a faster recovery speed. The experiment results show that CS reconstruction of SAR images on parallel architecture has a significant speedup in comparison with implemented on conventional serial architectures.