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

With the development of society,people have put forward higher requirements for the resolution of images,especially in many fields such as remote sensing and medicine,if the resolution of images is low,it will cause certain difficulties for subsequent processing and application.Super-resolution reconstruction technology can improve the resolution of the image through software,and the use of deep learning methods can significantly improve the effect of image reconstruction,but there are still problems such as low reconstruction image quality and poor visual effects.In order to solve the above problems,the paper proposes three image super-resolution reconstruction algorithms based on deep learning,and the specific work is as follows:(1)Firstly,aiming at the problem of low-resolution image feature information utilization in the process of image super-resolution reconstruction,the image superresolution reconstruction algorithm of multi-scale residual attention network is proposed,which uses different convolutional nuclei to obtain features of different scales,and distinguishes the importance of the characteristics of each channel through the second-order channel attention module,which strengthens the feature extraction ability of the network and makes full use of the information in the low-resolution image.Experimental results on different datasets show that this algorithm has achieved good results in evaluating indicators and visual quality.(2)Then,in view of the problem that the correlation of features at each level of the image super-resolution reconstruction network is not paid enough attention,the image super-resolution reconstruction algorithm of the hierarchical attention network is proposed,and the hierarchical attention module is used to increase the attention to the important level features,which further enhances the feature extraction ability of the network.Experimental results show that this algorithm improves the reconstruction effect of the detailed information in the image and improves the quality of the reconstructed image.(3)Finally,aiming at the difficulty of generating adversarial network training and the problem of more artifacts in the reconstructed image,the image super-resolution reconstruction algorithm of the generative adversarial network is proposed,and the multi-scale residual attention network is used as the generative network,and WGANGP is introduced,which improves the stability of model training.Experimental results show that this algorithm reduces the generation of artifacts in the reconstructed image,stabilizes the training process of the network,and improves the effect of image reconstruction.