Deep Learning Methods for Satellite Image Super-Resolution

Satellite image super-solution with deep learning techniques is a very active field of research. As overviewed in this chapter, most of the work has focused on designing novel and more effective neural network architectures. Comparatively, little work has been done on methods that do not require HR ground truth images and carefully model the degradation process. It is also still an open question what kind of improvements could be obtained from raw images rather than the higher level products that are commonly available.