Research On Super-resolution Reconstruction Of Remote Sensing Images

The remote sensing image with high resolution can not only improve and improve the quality of remote sensing image information processing,but also be more effective,accurate and accurate.Clearly used in specific practical applications such as surface object detection,classification and recognition.However,the remote sensing image acquisition process is indispensable due to the remote sensing imaging system’s own hardware conditions,image transmission process degradation and other factors,resulting in low resolution of the acquired image,blurred edge details and other problems.Therefore,under limited hardware conditions,it is of great significance and value to use software to obtain high-resolution remote sensing images.To this end,this article has carried out research based on the super-resolution reconstruction algorithm of remote sensing images.The main contents of the work are as follows.(1)Aiming at the problem of poor natural weather and unstable remote sensing equipment,low resolution of remote sensing images and blurred texture details lead to insufficient utilization of remote sensing data information,based on the original Generative Adversarial Network(GAN),a super-resolution reconstruction of stable GAN is proposed.Algorithm SSRGAN(stable super-resolution generative adversarial networks).Among them,the Spectral Normalization(SN)layer is used instead of the Batch Normalization(BN)layer to prevent the color information of the remote sensing image from being normalized when it passes through the BN layer,reducing the difficulty of reconstruction,and using Earth Mover’s Distance(EMD)calculates the distance between the reconstructed image and the real image,and makes gradient penalties on the discriminant network,which together make the network more stable and help restore more texture details.The experimental results show that compared with other related super-resolution algorithms,the reconstructed remote sensing image has a better visual effect.(2)On the basis of the existing super-resolution algorithm based on deep learning methods,the different scales of ground objects in remote sensing images cause the problems of inaccurate features of reconstructed images and lack of information.Taking the residual dense block as the basic unit,this paper proposes a multi-scale generation countermeasure network remote sensing image super-resolution reconstruction algorithm with prior information.It mainly uses multi-scale cavity convolution to extract multiple features from remote sensing images and generate semantic segmentation maps.Then the semantic segmentation maps are added as prior information to the generation confrontation network to further restrict the instability of the network and improve the quality of reconstruction.