High-Resolution SAR Image Reconstruction Using Deep Unfolding for Disaster Monitoring

This study proposes a deep unfolding approach for high-resolution SAR (Synthetic Aperture Radar) image reconstruction, aiming to improve image quality in disaster monitoring applications. SAR images often suffer from deterioration and resolution degradation due to atmospheric conditions and signal interference, which can reduce the accuracy of disaster analysis. The proposed deep unfolding technique combines the advantages of conventional optimization methods with the capabilities of deep learning to learn better and more accurate image representations. The approach consists of iterative unfolding that adapts data-driven learning with an optimization model to address noise, distortion, and resolution deficiencies in SAR images. The developed deep unfolding model is trained using SAR data from various disaster events, such as floods, earthquakes, and tsunamis, to learn distinctive patterns and structures in SAR images. Experimental results show that this approach successfully improves image quality with significant noise reduction and up to 30% resolution increase compared to conventional reconstruction techniques. Evaluation using the Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM) metrics demonstrated substantial improvements in the quality of recovered imagery, enabling more effective and accurate disaster monitoring. With the ability to recover lost details in SAR imagery, this deep unfolding approach opens up opportunities for broader applications in satellite imagery-based disaster monitoring and emergency response.