Lunar in situ resource utilization is a core goal in lunar exploration, with accurate lunar rock pose estimation being essential. To address the challenges posed by the lack of texture features and extreme lighting conditions, this study proposes the Simulation-YOLO-Hourglass-Transformer (SYHT) method. The method enhances accuracy and robustness in complex lunar environments, demonstrating strong adaptability and excellent performance, particularly in conditions of extreme lighting and scarce texture. This approach provides valuable insights for object pose estimation in lunar exploration tasks and lays the foundation for lunar resource development. First, the YOLO-Hourglass-Transformer (YHT) network is used to extract keypoint information from each rock and generate the corresponding 3D pose. Then, a lunar surface imaging physics simulation model is employed to generate simulated lunar rock data for testing the method. The experimental results show that the SYHT method performs exceptionally well on simulated lunar rock data, achieving a mean per-joint position error (MPJPE) of 37.93 mm and a percentage of correct keypoints (PCK) of 99.94%, significantly outperforming existing methods. Finally, transfer learning experiments on real-world datasets validate its strong generalization capability, highlighting its effectiveness for lunar rock pose estimation in both simulated and real lunar environments.

Pose estimation of non-cooperative spacecraft remains challenging under feature degradation. Motion blur, self-occlusion, and weak texture can cause structural line disappearance, correspondence ambiguity, and localization drift, which destabilize conventional point- and line-based analytic pose estimation pipelines relying on discrete feature detection and post-hoc 2-D-to-3-D association. To address these issues, we propose a two-stage framework for line-based 6-DoF pose estimation built upon a structure-bound multi-channel spatial distance mapping (SDM), where each SDM channel is uniquely associated with one predefined 3-D model line. By explicitly binding each SDM channel to a predefined 3-D model line, the proposed representation encodes 2-D-to-3-D line correspondence directly in the network output, thereby avoiding unstable line matching after prediction and providing solver-consistent geometric constraints for Perspective-n-Line (PnL) estimation. To reduce localization blur around the SDM zero-level set, a cross-scale self-attention (CSSA) mechanism is introduced to couple high-resolution localization features with low-resolution structural context through window-level cross-scale attention. Based on the predicted SDMs, explicit 2-D structural lines are recovered through weighted robust fitting in narrow bands around the zero-level sets, enabling the completion of partially or fully occluded lines and yielding solver-ready observations for PnL pose recovery. Experiments on a close-range non-cooperative spacecraft dataset with simulated observation distances of 10–30 m show that SDMNet achieves translation/rotation errors of 0.8%/0.0372 rad, 0.91%/0.0394 rad, and 1.38%/0.0579 rad under original, motion-blur, and occlusion conditions, respectively. These results indicate that the proposed framework can robustly recover correspondence-aware structural observations from degraded images and improve the accuracy and stability of spacecraft pose estimation.