Refocusing Satellite Antenna Motion in Compact Polarimetric ISAR Images Using a Spatial-frequency Feature-guided Brownian Bridge Diffusion Model

Polarimetric Inverse Synthetic Aperture Radar (ISAR) is an important tool for obtaining high-resolution information about satellite targets. Compact polarization provides an optimal balance between system complexity and polarization information capacity and has been widely used in various ground-based ISAR systems. However, during satellite transit, the onboard antenna components often undergo independent mechanical rotation relative to the satellite main body to adjust the observation area. This motion introduces additional frequency modulation into the echo signals, resulting in defocused ISAR images. Existing deep learning-based refocusing methods primarily focus on the global refocusing of ISAR images and often overlook the frequency feature variations induced by component motion, thereby limiting the model’s focusing capabilities. To address these challenges, this study proposes a novel refocusing method for satellite antenna motion in compact polarimetric ISAR imaging based on a spatial-frequency feature-guided Brownian bridge diffusion model. The core idea behind this method is to utilize the Brownian bridge diffusion model to learn the intricate mapping between defocused and focused ISAR images. Simultaneously, spatial and frequency features are extracted from the ISAR images to guide this learning process, thereby achieving precise refocusing of the moving components. To validate this approach, an electromagnetic simulation dataset for polarimetric ISAR satellite targets is established, and comparative experiments are conducted. The results show that the proposed method achieves superior performance in refocusing moving components and exhibits enhanced generalization capabilities.