| Super-resolution reconstruction method refers to the task of restoring one or more sets of low-resolution image data obtained from the same scene to high-resolution image data.In recent years,the single image super-resolution reconstruction algorithm based on deep learning has made great progress.However,in the process of image reconstruction,there are still many defects in the methods of image feature extraction and nonlinear mapping,which make the network architecture have low utilization rate of different layer information and serious loss of high-frequency details.as a result,the reconstructed image features are too smooth,the texture details are too rough,and the subjective visual effect is poor.In order to solve the above problems,based on the in-depth study of super-resolution reconstruction algorithm based on deep learning,two super-resolution reconstruction algorithms are proposed.The main contents are as follows.(1)An image super-resolution reconstruction method based on deep residual spatial channel attention and densely connected network is proposed.Aiming at the problem of insufficient utilization of deep features in the process of image reconstruction,a dense connection network is introduced to strengthen the utilization of information extracted from each layer,and residual learning is introduced in the overall architecture to improve the integration of shallow features and deep features.In order to improve the ability of the network to process information across feature channels and space,and improve the performance of the deep network,this thesis proposes to combine the channel attention mechanism and the spatial attention mechanism to fully extract and utilize the key high-frequency feature information in the image,so that the network can focus on more useful channel and spatial information,and enhance the ability of discriminative learning.Finally,the effectiveness of this study is proved from both objective and subjective evaluation indicators.(2)This thesis proposes a deep recurrent residual network super-resolution reconstruction method based on Transformer.In the field of deep learning,although there is a positive correlation between the overall performance of the network and the size of the network and the depth,in many scenarios,the storage space and computing power are limited,which makes it difficult to give full play to the performance of the deep network architecture.Therefore,the deep learning network based on the recursive idea is introduced,and the limitations of the convolutional neural network in feature extraction and feature mapping are considered.The Transformer module is introduced into the super-resolution reconstruction network model,and the advantages and advantages of convolutional neural network and Transformer module in the process of super-resolution reconstruction are fully considered,and the two are integrated to a certain extent.The network model architecture and the hardware facilities and parameter Settings used in the verification process are introduced in detail.Finally,the proposed algorithm is compared with the existing classical super-resolution reconstruction model in subjective visual effect and objective evaluation index,which proves the effectiveness of this study.There are 23 figures,7 tables and 81 references in this thesis. |
