Intention Recognition of AAV Swarm Based on GAT-EPool-BiGRU Model
Against the backdrop of rapid advancements in drone intelligence within military applications, the real-time and accurate identification of enemy drone swarm operational intent has become crucial for battlefield decision-making. Addressing the limitations of existing methods—such as the l...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/11079591/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Against the backdrop of rapid advancements in drone intelligence within military applications, the real-time and accurate identification of enemy drone swarm operational intent has become crucial for battlefield decision-making. Addressing the limitations of existing methods—such as the low feature transfer efficiency of stacked autoencoders (SAE) and the tendency of panoramic convolutional long short-term memory networks (PC-LSTM) to lose tactical details—this paper proposes a novel deep learning model called GAT-EPool-BiGRU, which integrates Graph Attention Networks (GAT), Edge Pooling (EPool), and Bidirectional Gated Recurrent Units (BiGRU). The experiments were conducted using 13 features, including altitude, velocity, and acceleration, covering 7 types of tactical intent such as attack, feint, and reconnaissance. With recognition accuracy as the primary evaluation metric, the results demonstrate that the proposed model achieves an intent recognition accuracy of 95.5%, outperforming mainstream models—5.4% higher than Transformer and 8% higher than GCN-LSTM. This performance improvement stems from GAT’s dynamic graph construction capability, EPool’s key feature retention mechanism, and BiGRU’s bidirectional temporal processing, collectively enhancing both robustness and accuracy in intent recognition. Additionally, the study enhances decision-making credibility through interpretability analyses, including attention heatmaps and SHAP values, which reveal critical features and their correlations with tactical intent. This approach not only improves model transparency but also provides actionable insights for real-world military applications. |
|---|---|
| ISSN: | 2169-3536 |