Intelligent Recognition and Integrated Navigation Technology for Typical Spacecraft Components

The demand for on-orbit services—like autonomous spacecraft assembly, on-orbit maintenance, and logistical support—is growing as technology develops. Common spacecraft parts and navigation algorithms form the foundation for the previously described on-orbit services. Due to the influence of light intensity, imaging quality, and spatial interference, standard approaches and deep learning algorithms for spacecraft component identification and localization still require improvement in terms of accuracy and stability. To address these issues, two contributions are made: firstly, the information of three-dimensional point clouds is integrated, based on the application of deep learning algorithms to identify spacecraft targets in low light. Second, to guarantee that highprecision navigation information is supplied autonomously, in real-time, and constantly throughout spacecraft operation, a plug-and-play fusion navigation technique for vision/LIDAR heterogeneous sensitive devices is created based on the factor graph architecture. The work in this paper simplifies the computational resources compared to deep learning algorithms that directly use 3D point cloud recognition and enables the accuracy and stability of spacecraft component recognition and navigation under low light to be improved. In addition, it also improves the implementability of the algorithm in practical engineering applications.