基于改进PoseNet的航一天器位姿估计网络  

Spacecraft pose estimation network based on improved PoseNet

作　　者：王煜晶 叶瑞达 段鑫杰 王丽芬 赵越 WANG Yujing;YE Ruida;DUAN Xinjie;WANG Lifen;ZHAO Yue(Space Engineering University,Beijing 101416,China)

机构地区：[1]航天工程大学,北京101416

出　　处：《航天工程大学学报》2025年第2期32-39,共8页

摘　　要：提出了一种改进的PoseNet网络,解决航天器位姿估计中的鲁棒性与精度问题。通过引入注意力机制和测地线损失函数,该网络显著提升了关键特征捕捉能力和估计精度。实验表明,改进后的网络在测试通过率、位置误差和姿态误差等关键指标上均优于原始模型,有效验证了其在复杂空间环境中的适用性。研究证实,测地线损失函数与注意力机制的协同作用能显著增强PoseNet模型的性能。An enhanced PoseNet network is proposed to address robustness and accuracy challenges in spacecraft pose estimation.By incorporating an attention mechanism and a geodesic loss function,the network exhibits a substantial improvement in capturing key features and refining estimation accuracy.Experimental results demonstrate that the improved network outperforms the original model in key metrics,including test pass rate,position error,and orientation error,validating its effectiveness in complex space environments.The research confirms that the synergistic integration of the geodesic loss function and attention mechanism substantially enhances the performance of the PoseNet model.

关 键 词：位姿估计 PoseNet 测地线损失函数 注意力机制 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]