Research On Super-resolution Reconstruction Algorithm Of Remote Sensing Image

The long distance between the satellite remote sensing imaging system and the target object,which results in the problem of reducing the recognition rate and missing some details in image processing,and affects the accurate extraction of the target in remote sensing image.Therefore,the research on super-resolution reconstruction method of remote sensing image is of great significance.On the basis of the sparse representation theory and convolutional neural network super-resolution reconstruction method,a multi-feature joint super-resolution reconstruction algorithm based on non-local self-similarity is proposed to solve the problem of fuzzy local details and missing partial feature information of remote sensing image.In the algorithm,we use the gradient operator and phase consistency method to represent the high-frequency information of remote sensing image,use the idea of a joint dictionary and K-SVD decomposition algorithm for sparse dictionary learning,and a non-local self-similarity constrained has been introduced to optimize the reconstructed image globally.To further solve the complex structure of remote sensing image reconstruction problem.we propose an improved convolution neural network super-resolution reconstruction algorithm,by increasing the network layer to learn deeper image characteristics,deconvolution is used to carry out adaptive up-sampling of remote sensing images,and residual learning structure is added into the network,which solves the problem that it is difficult for artificial feature extraction methods to accurately express the structural details of complex remote sensing images.The simulation results show that the average PSNR value and SSIM value of the multi-feature combined super-resolution reconstruction algorithm based on non-local self-similarity than the SCSR algorithm is improved by 1.17%and 0.05%,respectively.The average PSNR value and SSIM value of the improved convolutional neural network super-resolution reconstruction algorithm are 3%and 0.43%higher than those of the multi-feature joint super-resolution reconstruction algorithm based on non-local self-similarity.The proposed reconstruction algorithm shows a better reconstruction effect than the traditional algorithm and provides a theoretical reference for the super-resolution reconstruction of remote sensing images.