Super-Resolution Reconstruction of Remote Sensing Images Using Generative Adversarial Network with Permuted Self-Attention

Image super-resolution (SR) reconstruction plays a key role in meeting the ever-growing demand for high-spatial-resolution remote-sensing imagery. Although generative adversarial networks (GANs) have been widely adopted for SR, their dependence on high-frequency information learned from training data often produces artifacts or distortions in complex remote-sensing scenes. Optical satellite images exhibit more intricate spatial distributions and richer multi-scale ground features than natural images; therefore, directly applying existing SR methods to them usually causes unstable convergence and noticeable visual artifacts, seriously degrading reconstruction quality and usability. To address these issues, we propose an improved GAN-based SR network that embeds a Permuted Self-Attention (PSA) module to strengthen global modeling. The PSA module employs a global-context-aware mechanism to adaptively select useful information and suppress noise, markedly improving the reconstruction of multi-scale objects in remote-sensing images. Extensive experiments on standard remote-sensing datasets demonstrate that the proposed method outperforms state-of-the-art alternatives in both objective metrics (PSNR, SSIM) and subjective visual quality, confirming its robustness and effectiveness in complex remote-sensing scenarios.