BiGAT: A Model for Recognizing Motion Intentions of Space Noncooperative Targets

This study explores a method based on angular measurement time series to recognize the intentions of space noncooperative targets. By analyzing the relationship between motion intentions and relative orbital dynamics, a feature label sequence for motion intentions is developed, transforming the task from single-label to multifeature recognition. The BiGAT model, integrating bidirectional gated recurrent units (BiGRU), self-attention (SA) mechanisms, and Transformer architectures, is proposed for intention recognition. Training and testing datasets are generated using orbital dynamics models, and BiGAT is trained to achieve recognition accuracies of 98.55957% and 98.90137%, respectively. Experimental results demonstrate that BiGAT significantly outperforms traditional approaches like BiGRU-SA, particularly in recognition accuracy and robustness against interference.