Spacecraft Attitude Estimation Network Based on Deception Training and Space Channel Attention Collaboration

Estimating the pose of the target spacecraft is a necessary technology for achieving space attack and defense, as well as in orbit services. 6D pose estimation in space presents unique challenges that are not commonly encountered in ground environments. Due to the absence of atmospheric scattering, space lighting conditions are more complex, and complex backgrounds such as Earth can also interfere with pose estimation. This article proposes a simple and effective spacecraft pose estimation network DAC-Net to address this issue, which reduces the side effects caused by background interference through a deception training strategy based on anti-background feature extraction. However, due to the large proportion of close range targets in the image and limited background information, their ability to deceive training is limited. In order to further obtain more comprehensive semantic information, this paper adopts a bidirectional feature fusion strategy based on spatially guided channel attention. The validation was conducted on the SwissCube dataset for spatial target pose estimation, and the results showed that the proposed method outperformed existing methods for target pose estimation at different distances and complex backgrounds.