SPTN: Transformer-based spacecraft pose estimation network for space objects tracking

In this work, we presents a novel transformer-based spacecraft pose estimation network, SPTN, for space-object tracking. SPTN consists of a transformer-based backbone with the proposed WBlock module, an innovative neck structure, LBiFPN, and a multitask head. Such a framework will be more effective in feature extraction and fusion while maintaining a lightweight structure compared to CNN-based methods. The proposed WBlock is embedded with window partitioning and hierarchical attention mechanisms to enhance feature extraction. The novel LBiFPN neck module is designed to fuse features at different levels, facilitating a deeper feature integration. Extensive experiments are conducted on the SPEED+ and SHIRT datasets to evaluate the performance of the proposed method. The results show that our SPTN model achieved competitive detection accuracy compared to current state-of-the-art methods while maintaining minimum parameters.