RSSRDiff: An Effective Diffusion Probability Model with Attention for Single Remote Sensing Image Super-Resolution

Recently, optical remote-sensing images (RSIs) have found widespread applications in fields such as environmental monitoring and land cover classification. However, these images often suffer from limitations in imaging equipment and other factors, resulting in low-resolution images that are not ideal for image analysis. While existing image super-resolution (SR) algorithms can enhance image resolution, they are not specifically tailored to the characteristics of RSIs and fail to effectively recover high-resolution images. In this study, we propose a novel method called RSSRDiff, specifically designed to address the unique characteristics of RSIs using diffusion probability models. Our approach builds upon the SRDiff network, which has shown promise in the field of remote sensing. Additionally, we incorporate an attention mechanism to enhance the utilization of contextual information in low-resolution images. These improvements enable RSSRDiff to preserve more informative cues, leading to accurate SR results. To evaluate the performance, we conducted extensive experiments on four remote sensing datasets, including both simulated and real-world RSIs. The results demonstrate that RSSRDiff successfully restores high-quality SR images, showcasing its effectiveness in enhancing the resolution of RSIs.