Topology graph attention mechanism and its application for polarimetric ISAR satellite oriented component recognition

Polarimetric inverse synthetic aperture radar (ISAR) plays an important role in space surveillance. In comparison to the quad-polarization (quad-pol) mode, the compact polarimetric (CP) mode can balance the hardware costs and polarization information. For satellite observation, the recognition of satellite component can offer valuable insights for attitude inversion and the perception of abnormal motion. Comprehending the interrelationships among components across various categories is expected to enhance the recognition performance and refine the spatial position of detection results. In this vein, a topology graph is constructed, and a component-level attention mechanism is introduced. Consequently, a multi-stage recognition framework is employed for feature enhancement. Experimental results indicate that there is a 5.5% improvement for the mean average precision (mAP) index in two scenarios.