Parallelization And GPU Acceleration Of Compressed Sensing Reconstruction Algorithms

The compressed sensing theory effectively solves the problem of sampling frequency in the processing of sampling large bandwidth signals,which is based on the Nyquist sampling theory.In the decoding process,the original signal is rebuilt from the measured data by the reconstruction algorithm.The high computational complexity of the reconstruction algorithm leads to the long computation time,which seriously restricts the practical application of the compressed sensing theory.This paper focuses on the parallel acceleration scheme of parallelization computing,and studies the parallelization acceleration of the reconstruction algorithm.In this paper,we propose a method of accelerating the reconstruction algorithm based on parallel computing.The main innovation points are as follows:(1)In order to solve the problem of low resolution of image,the theory of compressed sensing is introduced into the process of image super-resolution reconstruction.Compressed sensing and super-resolution image reconstruction have the same essence,all of which are the reconstruction of high dimensional signals from low dimensional signals.The original image is taken as the result of measurement,and the result of reconstruction through the compression sensing reconstruction algorithm is the super-resolution image obtained.The quality of the reconstructed image and the selection of the reconstruction algorithm have a great influence.The experimental results show that the compression sensing reconstruction algorithm has achieved good results in the process of super-resolution image reconstruction.(2)Aiming at the long computation time in the process of image reconstruction based on compressed sensing reconstruction algoritlm,a parallel computing method is proposed to improve the computation speed.The image is properly partitioned,and the task based coarse-grained parallel is used.In the process of the reconstruction of each image block,the parallelization method of the reconstruction algorithm is analyzed,and the fine-grained parallel method based on data parallel is used.Under the CUDA platform,the coarse grained parallelism of image reconstruction is realized by using flow,and kernel function is used to realize fine grained parallelism.The experimental results show that the parallel method proposed in this paper has a good acceleration effect on the reconstruction of compressed sensing image.(3)Aiming at the complexity of the image reconstruction process and in order to verify the performance of image reconstruction,a compressed sensing image reconstruction system is designed in this paper.The system realizes the three processes of the observation matrix,sampling and reconstruction of the compressed sensing theory.These three parts can be operated independently.In the reconfiguration part,in order to verify the effect of image reconstruction,a variety of setting parameters can be selected for experiment.The system has a very good human-machine interaction interface.Software users do not need to understand the underlying principles,they can operate skillfully on software,and do image reconstruction experiments.In the process of experiment,you can also check the operation of each part at any time.