Abstract: Space target recognition is a core technology in space situational awareness (SSA). Deep learning-based target recognition techniques have been widely adopted due to their outstanding performance. However, due to differences in system architecture and bandwidth of Radar equipments and pose variations of targets, recognition models face the challenge of distributional discrepancies between training and testing datasets. Additionally, there is insufficient data accumulation in testing scenarios. To address these issues, this paper investigates transfer learning techniques for cross-modal, cross-frequency, and cross-pose adaptation. On the publicly available NASA-3D dataset, the proposed method achieves significant improvements in recognition accuracy and generalization performance compared to baseline methods.