SAR Super-Resolution Using Capella Satellite Imagery

Abstract. Electro-Optical (EO) images are limited in that they can only be captured during daylight and under clear weather conditions, which restricts their usability in certain environments. In contrast, Synthetic Aperture Radar (SAR) images have the distinct advantage of being able to capture high-quality data regardless of the time of day or weather conditions. This makes SAR images highly valuable across various fields such as national defense, remote sensing, and disaster monitoring. However, despite their advantages, SAR images often suffer from lower resolution due to artifacts like speckle noise, which can significantly degrade image quality.
To address this issue, numerous efforts have been made to enhance the resolution of SAR images through the use of SAR Super- Resolution (SR) techniques. However, research in this area is limited, primarily due to the high cost associated with acquiring real SAR data. Some existing studies in SAR SR have even resorted to using synthetic images that combine speckle noise with regular camera images, which do not accurately represent real SAR data. In response to this, our paper proposes a solution by constructing a dataset based on actual Capella SAR satellite imagery and introducing a novel SAR Super-Resolution model. 
For our experiments, we collected real SAR images from the Capella satellite, used them as high-resolution references, and generated low-resolution counterparts by down-sampling. By modifying state-of-the-art image restoration models for the SR task, we demonstrate through a series of experiments that our proposed model outperforms existing SR methods in both quantitative and qualitative assessments.