Research On Super-resolution Algorithm Of Single Remote Sensing Image Based On Deep Learning

The spatial resolution of remote sensing image is one of the important factors affecting the application effect of remote sensing.On account of the limitation of remote sensing imaging technology,the spatial resolution of remote sensing image will be reduced due to the long distance,low illumination,atmospheric interference and other factors.Therefore,super resolution reconstruction technology becomes an important means to improve the spatial resolution of remote sensing images.In recent years,with the development of deep learning technology,super resolution reconstruction technology based on deep learning has become one of the research hot spots,and has achieved remarkable results in practical applications.It is worth noting that although super resolution reconstruction technology can improve the spatial resolution of remote sensing images,it still has some problems such as single scale and insufficient use of fearure information.To solve the above problems,this paper proposes two super resolution reconstruction methods of remote sensing image,and the specific research contents are as follows:(1)Aiming at single scale problem,a super resolution algorithm of remote sensing image based on multi-scale attention enhancement is proposed.Firstly,dilated convolution with different dilation rates is used to extract image feature information to enlarge the feature receptive field,so as to obtain context information of different scales from the original image.Information of different scales is cross-fused to obtain complementary feature information.Then,channel attention is used to select useful channels to adapt to complex remote sensing images,so as to increase the proportion of useful information.Finally,sub-pixel convolution is used to up-sample the feature information to reconstruct the image.The experimental results show that the reconstructed images are improved by subjective and objective evaluation on the two data sets.(2)Aiming at the problem of insufficient use of local and global feature information of image,a new super resolution algorithm based on Transformer for local-global remote sensing image is proposed.The Swin Transformer backbone network is used to extract the local information of remote sensing image,and the global dependence is obtained through the large kernel attention mechanism,and the complementary advantages of the two are utilized to obtain the local features and global features of the image.In addition,in order to obtain more representative feature information,a large kernel attention tail module is added to effectively gather useful information and improve the quality of image reconstruction.Finally,experiments on two data sets show that the reconstructed results have certain improvements in the two indexes of peak signal-to-noise ratio and structural similarity.