An End-to-End Intent Recognition Method for Combat Drone Swarm

In the field of intent recognition of combat drone swarm, traditional methods are based on the data characteristics which are only from a single target and at a single moment. It is difficult to capture the feature information of the entire swarm on time series. This paper proposes an end-to-end UAV swarm intent recognition method. Firstly, the distance threat coefficient and angle threat coefficient between UAVs are used to model the graph structure data of UAV swarm. Secondly, a novel deep learning method based on graph attention network, graph pooling method and gated recurrent unit (GAT-AP-GRU) is designed. This network can process the graph structure data obtained by modeling and identify the intention of the swarm. Experiments comparing with other methods and ablation experiments demonstrate that GAT-AP-GRU outperforms state-of-the-art methods in terms of accuracy of intent recognition.