Research On Landsat-7 Image Restoration Algorithm Based On Improved Gan

Due to the failure of the scan line corrector of the Landsat-7 satellite,regular fringe occlusions appeared in the scanned images,which seriously affected the application of remote sensing images in radiative transfer calculations,numerical model forecasts,etc.The study of climate still has some value.Therefore,the restoration of this kind of satellite image has become a topic worth studying.At present,some research has been carried out on satellite image restoration,but most of them are for local distortion and small-scale restoration.The research on this kind of image restoration is still relatively weak,and the current restoration of this kind of image mainly relies on the original image information for restoration.This method will result in a high similarity between the repaired pixels and the original image features,and cannot make full use of a large number of Landsat-7 image resources.In this paper,an improved GAN network is proposed based on the generated confrontation network to solve the above problems The model performs restoration research on such images.In this paper,an image data set collected by the Landsat-7 satellite is established.Based on this data set,image restoration experiments are carried out.After comparing various GAN network models,a Cycle Gan network with better image restoration effect is finally obtained.This network can fully the historical image data is used for the restoration of this type of image,and can realize the optimized processing of the edge details of the pixel defect.In order to make better use of the rich reference image resources of the Landsat-7 satellite,the model uses the encoder,conversion The triple structure of the converter and the decoder,in which the form of the convolution block is used to fully extract the image features on the converter,so that the loss of image information is minimized,and the precise repair of the edge details of the pixel defect area is realized.After a large number of iterations,the model shows that The improved Cycle Gan network model has the advantages of fast learning and good repair effect.Finally,the PSNR index and SSIM index are used to compare and evaluate the restoration effect of other network models.The experimental results show that using the method proposed in this paper can show a better restoration effect.