Cross-Dimension Attention Guided Self-Supervised Remote Sensing Single-Image Super-Resolution

In recent years, the application of deep learning has achieved a huge leap in performance remote sensing image super-resolution (SR). However, most existing SR methods employ bicubic downsampling high-resolution (HR) images to obtain low-resolution (LR) and use obtained LR HR as training pairs. This supervised method that uses ideal kernel (bicubic) downsampled train network will significantly degrade when used realistic images, usually resulting blurry images. The main reason is degradation process real more complicated. data cannot reflect problem Inspired by self-supervised methods, this paper proposes cross-dimension attention guided single-image (CASSISR). It does not require pre-training on dataset, only utilizes internal information reproducibility single image, lower-resolution from input (CDAN). module (CDAM) selectively captures useful duplicate modeling interdependence channel spatial features jointly their weights. proposed CASSISR adapts well tasks. A large number experiments show superior current state-of-the-art methods.