Reserch And Hardware Implementation Of Image Superresolution Reconstruction Based On CNN

Image is the main carrier of information,in order to get more information from the image,image quality is particularly important.Due to the unique advantages of Convolutional Neural Networks in image processing applications,in recent years,Convolutional Neural Networks has started to be fused in the field of image super-resolution reconstruction.Compared with traditional image processing algorithms,it can significantly improve the quality of reconstructed images,but at the same time significantly increase the amount of algorithm calculation.Due to the limited computing capacity of traditional embedded devices,it is difficult to meet the practical application requirements in the scene with high real-time requirements.Aiming at the problem of high-quality image super-resolution reconstruction and real-time calculation on terminals,this paper is devoted to research and obtain a complete set of intelligent terminal solution combining algorithm and hardware,which meets the scene application in the field of image super-resolution reconstruction.The main work and innovation of this paper are as follows:1.Aiming at the quality requirement of image super-resolution reconstruction,this paper makes optimization and improvement on the basis of FSRCNN algorithm,and obtains the lightweight super-resolution neural network algorithm Tiny-FSRCNN.This algorithm can effectively improve the quality of image reconstruction,and simplify the computation and complexity of the algorithm,which is beneficial to the deployment and implementation of the algorithm on hardware.2.Aiming at the real-time requirement of mobile terminal scene for image processing,a solution based on FPGA to realize the computation acceleration of Convolutional Neural Networks is proposed.The algorithm model is realized by digital circuit design.Through the calculation of hardware circuit acceleration algorithm,through experimental comparison,it can realize several times of acceleration compared with CPU,which can meet the requirements of real-time image processing of the actual scene.3.According to the circuit structure characteristics of FPGA,the hardware resources of FPGA are maximized.DSP-Supertile technology is used to realize the high performance convolution computing circuit,which makes the clock frequency of DSP run at twice of the logic clock frequency.Under the condition of consuming the same amount of DSP computing resources,the computing force can be doubled.