Research On Super-resolution Image Reconstruction Algorithm Based On Deep Learning

The world has embraced the digital age.As the main medium for information dissemination,images have been widely used in various scenarios.In such field as medical imaging field,satellite remote sensing,high requirements for image quality has been required.Therefore,low-resolution images have been difficult to meet the needs of specific scenes in such a rapidly developed information age.Super-resolution image reconstruction technology reconstructs one or more low-resolution images to a high-resolution image through some algorithms.Although the existing image super-resolution algorithm based on convolutional neural network solves the problems that the algorithm of traditional image super-resolution reconstruction are poor robustness and computational complexity,it still has the problem of more parameters,and unclear images after reconstruction,and it performs poor reconstruction ability,poor visual effects in terms of images with more detailed.In order to solve the above problems,and due to the success in convolutional neural networks and generative adversarial networks in the visual field,this paper deeply studies the convolutional neural networks in deep learning and theoptimization methods of generative adversarial networks in the field of image super-resolution,and proposes many image super-resolution algorithm of scaled convolutional neural networks and Image super-resolution algorithm based on conditional generative adversarial networks.The main work and innovation of this article are as follows:(1)On the basis of reading a large number of chinese and foreign documents,a general summary of the current research status in the field of image super-resolution at home and abroad has been made,and the significance of research in the field of image super-resolution has been demonstrated.A detailed summary and the basic principles of convolutional neural network in deep learning and generative adversarial networks,as well as the advantages and disadvantages have been illustrated.(2)In this paper,the basic principle and basic structure of convolutional neural network are introduced in detail,and the formulas of back propagation and forward propagation are deduced.The theory verifies the superiority of convolutional neural network in image super-resolution reconstruction.Since the image super-resolution based on convolutional neural network has the problems of many parameters,and unclear images after reconstruction,this paper improves them and proposes an image super-resolution reconstruction algorithm based on multi-scale convolutional neural network.The algorithm not only has a certain degree of improvement in subjective vision,but also in structural similarity and peak signal-to-noise ratio,and it reduces the number ofparameters of the network while reconstructing high-resolution images with clear edge and restore more image details.(3)This paper deduces the basic principles of generative adversarial networks,introduces its basic structure and subsequent derivative models,and proposes an image super-resolution algorithm based on conditional generative adversarial networks.After comparing the image super-resolution algorithm based on the conditional generative adversarial networks with the classic image super-resolution algorithm,the paper proposed algorithm that has achieved the effect of the original image super-resolution algorithm based on generative adversarial networks in terms of visual effects and the visual effect is better than that of original image super-resolution algorithm based on generative adversarial networks.