SAR Image Super-Resolution via Self-Supervised Learning

Synthetic Aperture Radar (SAR) imaging is a critical technology in remote sensing, known for its ability to operate in all weather and day/night conditions. Despite its benefits, SAR imaging frequently suffers from low spatial resolution due to system constraints, complicating detailed analysis and interpretation. Super-resolution (SR) techniques solve this problem by increasing image resolution, recovering fine details, and improving overall image quality. In this study, we present a self-supervised convolutional neural network (CNN) for single-image SAR SR. The proposed network, trained on natural images to capture general SR features, is tested on SAR datasets that include both noisy and denoised images. The results show that the model is robust and effective at improving spatial resolution and recovering high-frequency details, indicating that it has the potential for practical applications in a variety of operational scenarios.

Keywords:

Super-Resolution , SAR Imaging , Self-Supervised Learning , Convolutional Neural Network , Image Reconstruction