Study On Reconstruction Algorithm For Single Image Super-Resolution

With the continuous progress of information technology and the development of artificial intelligence technology,people’s life has become more convenient and efficient,and the quality and resolution of images have also been constantly improved.How to get high-resolution images under the condition of limited hardware has gradually become the focus of relevant researchers。As an image reconstruction technology based on software algorithm,image super resolution has important applications in many fields,such as security,satellite remote sensing,medical imaging.With the application of deep convolutional neural network in the field of image super-resolution,the quality of super-segmentation image is getting higher and higher due to its super data learning ability and feature expression ability.Therefore,this thesis proposes two image super-resolution reconstruction algorithms based on convolutional neural network.The main research work is as follows:This thesis also designs and implements a residual feature fusion network based on channel attention,which uses a stratified attention to feature learning,and connects a modified channel attention module at the end of each residual block to re-select residual features.Finally,the features selected through the feature channel are fused to the tail of the module,making full use of residual features and real-time selection of features.The experimental results show that the network has a good reconstruction effect.Aiming at the problem that many super-resolution reconstruction algorithms ignore the effective use of residual features and the loss of feature information during network feature transmission,a super-resolution model of spatial and channel attention images based on residual feature fusion was proposed.Through the mechanism of residual feature fusion,the feature information of the middle layer of the network is introduced to the tail of the whole module by jumping connection and feature fusion,which improves the utilization of residual features and the flow of information among the network convolution layers.In addition,the fused features are introduced into the space and channel dual attention module designed in this thesis to select and learn features,which improves the learning ability of the network to image texture,details and other high-frequency information,and enhances the effect of super-segmentation image reconstruction of the network.