Three-dimension polarimetric correlation pattern guided satellite oriented component recognition

Polarimetric inverse synthetic aperture radar (ISAR) offers continuous, all-weather space surveillance capabilities. Identifying satellite components can be beneficial for monitoring their operational status and health. Nevertheless, target orientation relative to the radar line of sight usually exhibits significant influences on the scattering mechanisms. Additionally, ISAR projection introduces orientational characteristics to satellite components in polarimetric ISAR images, which poses challenges for their precise localization and identification. Recently, this diversity in target scattering has been successfully characterized and employed using the three-dimension polarimetric correlation pattern (3-D PCP) interpretation technique, enabling the differentiation of various scattering structures. This study analyzes the relationship between satellite polarimetric responses and both polarimetric orientation angle and polarimetric ellipticity angle based on 3-D PCP. Then a hybrid approach combining 3-D PCP and data-driven model is designed for oriented satellite component recognition. In contrast to using a single polarimetric channel as input for deep neural networks, our approach transforms the data domain and utilizes three independent 3-D PCPs to drive the network. On one hand, the network training is guided by manually extracted features derived from 3D-PCP, including statistical characteristics such as mean, standard deviation, extreme values, and contrast. On the other hand, adaptive feature extraction is performed through a simple convolutional structure and integrated with electromagnetic scattering characteristics. Six satellites are utilized for constructing polarimetric ISAR dataset. Experimental results demonstrate that the proposed method achieves a superior performance, evidenced by a 2.3% improvement in the mean average precision (mAP) index.