Design Of Infrared Night Vision Image Enhancement System Based On FPGA

Navigation image becomes an effective way to aid navigation because of the high risk of night navigation.Navigation images are mainly based on visible light,infrared and remote sensing imaging.Infrared imaging technology can observe invisible energy graph,which has obvious advantages for ship monitoring at night.But the infrared image has many problems such as noise,details are not clear enough.Therefore,how to make the image with high resolution and fast real-time display has become a hot topic in this field.In order to solve the above problems,this thesis adopts super-resolution reconstruction algorithm to enhance low-resolution images and obtain high-resolution images.An image processing system is constructed based on field programmable gate array,which has fast computing speed and high real-time performance.The main research contents are as follows:On the basis of in-depth research of the infrared night vision image characteristic,the Discrete Cosine Transform is used to denoise the image.The traditional super-resolution reconstruction algorithm build on Convolution Neural Network for fuzzy image processing effect is poor because of the movement of the problem,it is the super-resolution reconstruction algorithm build on the Flow Net Corr proposed,the Flow Net Corr is introduced into the super-resolution reconstruction network,and two images respectively before and after the video through three layers of convolution,convolution correlation processing.At the same time,a feature denoising layer is added to the network,and the depth denoising is cascaded through feedback.The network includes optical flow layer,feature extraction layer,denoising layer and reconstruction layer.The proposed algorithm is compared with the traditional super-resolution reconstruction algorithm by using PSNR and SSIM on the same data set.The results show that the proposed algorithm is superior to the traditional algorithm in each evaluation index.In view of the problem that image enhancement system has long processing time and can't effectively transmit display information in real time when there is a large amount of data,it is achieve the hardware system build on FPGA,including image acquisition,storage,processing and display module,and focuses on the transplantation of image enhancement algorithm based on FPGA.Depending on the network structure of the super-resolution reconstruction algorithm designed in this thesis the overall architecture of the image processing system is designed.In the light of the characteristics of FPGA calculation,a quantitative optimization algorithm based on fixed-point calculation is proposed to improve the feasibility of the algorithm.Based on the Caffe framework,the FPGA is trained at a fixed point,and the Linux environment is configured at the same time to complete the experimental verification.Compared with the PC platform,the FPGA's processing speed is increased by 54%,the power consumption is reduced by 56%,and the real-time display of image information is realized while completing the rapid processing of images.