The Theory And Application Of Fast Compression Sensing Reconstuction Algorithm Based On GPU

Compressed sensing (compressed sensing) is a kind of sparse technique for underdetermined linear system. Compressed sensing is applied to image processing, for acquiring and reconstructing sparse or compressible signals. The breakthrough of compressive sensing technology is that the process can be reconstructed below the sampling frequency of the sampling theorem. But the reconstruction of Compressed Sensing is complex.Nvidia introduced the CUDA-general parallel computing architecture, which allows the GPU to solve complex computational problems. The purpose of CUDA is to help the CPU to carry out the numerical calculation, and the optimization of the original program can be improved by CUDA.The main work of this paper is based on the powerful parallel computing ability of embedded Jetson TK1 NVIDIA to construct the compressed sensing reconstruction algorithm, including the orthogonal matching pursuit algorithm, the two step thresholding iterative algorithm, and the parallel implementation of Bregman algorithm. This paper studies the orthogonal matching pursuit algorithm, two step iterative thresholding algorithm, linear Bregman algorithm, proposes a simplified memory management system for CUDA programming, study parallel technique such as parallel reduction, loop unrolling parallel algorithms and optimization, the results show that the calculation efficiency of our system can significantly improve the algorithm.