Beijing Institute of Technology | China
The research work focuses on the intelligent interpretation of synthetic aperture radar imagery, with particular emphasis on end-to-end understanding of satellite-based SAR data. Core research activities include SAR image pre-processing, Speckle noise suppression in SAR images speckle noise suppression, and robust target detection and recognition under complex imaging conditions. A key scientific contribution lies in addressing the limitations of conventional supervised learning approaches, which typically rely on clean reference images that are rarely available in real-world SAR scenarios. To overcome this challenge, a self-supervised despeckling framework was proposed, enabling effective network training using only intensity SAR images without the need for external ground-truth data. This strategy significantly enhances the practicality and scalability of deep learning methods for operational SAR systems. The research further contributes to improving feature preservation and structural consistency in despeckled images, which directly benefits downstream tasks such as object recognition and scene understanding. In addition, the work actively supports national-level research and development initiatives, fostering collaboration across multidisciplinary teams in remote sensing, signal processing, and artificial intelligence. Overall, these contributions advance the reliability, adaptability, and real-world applicability of intelligent SAR image interpretation, strengthening its role in satellite observation, surveillance, and Earth monitoring applications.