多分支自相似遥感超分辨率生成对抗网络OA

In allusion to the image distortion caused by edge blurring and artifacts in images obtained by the super-resolution reconstruction of the remote sensing imagery,a new remote sensing super-resolution reconstruction algorithm is proposed.In the algorithm,a multi-branch residual dense block(MRDB)that incorporates a joint loss function,and simultaneously employs a self-similarity feature extraction module to repair its high-frequency and edge information is designed.MRDB can improve the multi-branch structure based on RRDB,which can effectively process information of different frequencies,enhance the detail restoration effect and semantic balance of images,and reduce the problem of object edge blurring.The innovative combination of multi-branch structures and dense blocks can stably extract deep features and effectively eliminate artifacts.A joint loss function is designed,which combines L1 content loss,perceptual loss,texture loss,adversarial loss,and self-similarity loss to ensure the overall clarity of the image.The comparative experiments and ablation experiments are conducted on MRDGAN.The experimental results show that in the UC Merced dataset,the qualitative effect of MRDGAN under the categories of highway,airport,and building is closer to original image.Moreover,the average PSNR is 1.13 dB higher than that of the ESRGAN algorithm,the SSIM is 0.028 5 higher,the FID is reduced by 22.13,and the CLIPscore is increased by 0.036 1.This algorithm not only can remove generated artifacts and improve the accuracy of edge reconstruction,but also can demonstrate better results in various evaluation metrics.