The Research On Compressive Sensing Image Super-Resolution Reconstruction Using SRCNN

Compressed sensing(CS)technology has been widely used in real life for its advantages of lower sampling rate than Nyquist's demand in sparse signal processing.The most critical problem is whether the original image signal can be reconstructed accurately by super-resolution.SRCNN(Super-Resolution Convolutional Neural Network)is one of the super-resolution convolution neural networks.Because of its simple end-to-end processing and accurate super-resolution reconstruction ability,it has been greatly welcomed in the field of image processing.Based on the algorithm of compressed sensing image reconstruction,this paper combines the super-resolution convolution neural network SRCNN with the compressed sensing reconstruction algorithm OMP and the block OMP,and uses the MATLAB software to simulate the hybrid algorithm.It is proved that the hybrid algorithm is better than the compressed sensing for image reconstruction.Finally,based on the Vivado software in FPGA,the simulation experiment of the super-resolution convolution neural network SRCNN algorithm is carried out.This paper first introduces the theoretical knowledge of compressive sensing and super-resolution convolution neural network.The research background of compressed sensing and the main algorithms in the present study are briefly analyzed,and the algorithms and structural features of different types of super-resolution convolution neural networks are discussed,which will pave the way for the proposed new algorithm.Then,the idea of combining the super-resolution convolution neural network and the compressed sensing image reconstruction algorithm is proposed,and its feasibility is proved according to the related mathematical theory.The simulation software is used to combine the compressed sensing image reconstruction algorithm OMP,the Block OMP and the super-resolution convolution neural network SRCNN to form a hybrid algorithm model and carry out simulation experiments.The simulation results show that the hybrid algorithm is better than the simple compression process in the image reconstruction.Finally,the hardware implementation of super-resolution convolution neural network SRCNN algorithm is introduced.First of all,the hardware Zedboard is introduced,and the process of using Vivado and Vivado HLS software to compile the RTL circuit diagram and the hardware simulation waveform is carried out.By comparing and analyzing the circuit diagram and the simulation waveform signal flow,the accuracy of the compiled code is verified,and the hardware simulation of the algorithm is completed.