Research On Super-Resolution Reconstruction Algorithm Of Infrared Remote Sensing Image

Infrared remote sensing images play a special role in civil and military applications due to their unique advantages of strong anti-interference ability.However,current infrared remote sensing images still have problems such as low resolution,blurred details and low signal-to-noise ratio,which affect their practical applications.Therefore,improving the resolution of infrared remote sensing images has become an important research direction.Therefore,in order to improve the image resolution,this paper has carried out in-depth study on the super-resolution reconstruction technology of infrared remote sensing images.In this paper,the factors affecting the resolution of infrared remote sensing image are analyzed in detail,and the degradation model of infrared remote sensing image is established.Then,the super-resolution reconstruction technology for improving the degradation model of infrared remote sensing image is studied.The theoretical basis for restoring high-frequency detail information and the commonly used super-resolution reconstruction algorithm based on single frame image is described.Aiming at infrared remote sensing image,this paper proposes a new superresolution reconstruction method of single-frame image based on deep convolution residual learning network.Sub-pixel convolution is used in the convolutional neural network to replace traditional interpolation preprocessing to complete image upsampling and reduce the impact of noise on image reconstruction.The method of combining deep network and small convolution kernel is proposed to increase the sensing field of convolutional neural network,better extract image feature information,improve the ability of detail reconstruction,and reduce the storage requirement of network parameters.A kind of residual network is designed to solve the problem of gradient explosion,gradient disappearance and long-term dependence caused by deep network,so as to improve the network's ability to obtain prior information and accelerate the convergence of the training process.The algorithm proposed in this paper is used for super-resolution reconstruction of infrared images with a resolution of 10 m.In terms of visual effect,the reconstructed images have clear details and strong noise suppression ability.In terms of peak signal-to-noise ratio and structural similarity,they are 5.27%,6.93%,1.06% and 2.48% higher than SRGAN and SRCNN respectively.Moreover,the infrared remote sensing images of 20 m resolution of plain,mountain and village scenes are reconstructed with super resolution respectively,and the algorithm has obvious advantages.In conclusion,the super-resolution reconstruction method of deep convolution residual learning network proposed in this paper can effectively improve the resolution of infrared remote sensing images,which has great application and promotion value.