Remote Sensing Image Reconstruction Algorithm Based On Matching Pursuit Algorithm And Convolution Neural Network

In the process of remote sensing image acquisition,it will cause the geometric deformation of the image,which makes the geometric figure in the image different from that of the object in the selected map projection,and then leads to the geometric shape or position distortion.In order to eliminate the above error,the remote sensing image needs to be reconstructed.Compressed sensing technology removes redundant signals by compression,which greatly reduces the burden of information storage and transmission.In addition,it is called a new sampling technology,which separates the sampling technology from the main obstacle of bandwidth,so as to ensure that the lossless recovery can still be carried out at very low sampling frequency.The reconstruction algorithm of compressed sensing technology determines whether the signal can be recovered with high accuracy,so it has been widely studied.Based on this,this paper studies the problem respectively.As the main advantage of greedy algorithm is fast convergence speed and low complexity,it has become the focus of research.However,how to choose a better support set to improve the accuracy of sparse signal reconstruction has become the key of greedy algorithm in compressed sensing.Based on this,this paper uses Dice coefficient and Jaccard coefficient(D-J)matching criteria to optimize the selection of elements in the support set.By combining the D-J compression sampling matching criterion with the matching pursuit algorithm(CoSaMP),the D-J compression sampling matching pursuit algorithm(D-J CoSaMP)is proposed.A large number of simulation and experimental data show that the algorithm is superior to other greedy algorithms in sparse signal reconstruction and remote sensing image restoration.In addition,this paper combines the data-driven deep learning method with the traditional compressed sensing reconstruction algorithm,and proposes a deep compressed sensing network multi-scale deep differential dense network(MS-DRDNet).MS-DRDNet consists of four parts:multi-scale convolutional differential network,sampling network,initial reconstruction network and deep reconstruction network.Specifically,multi-scale convolutional difference network is used to capture sparse high-dimensional information in remote sensing image.Then,the information is sent to the sampling network for compression,so that the measured value is less.Finally,the initial reconstruction network and the depth reconstruction network are used to reconstruct the remote sensing image accurately.A large number of experiments show that the proposed MS-DRDNet is superior to the compressed sensing method based on iterative and deep learning in the accuracy of remote sensing image reconstruction.