A GRU-Based Method for Predicting Intention of Aerial Targets.

Area of Science:

• Aerospace Engineering
• Artificial Intelligence
• Computational Science

Background:

• Traditional aerial target combat intention recognition methods are limited by relying on single-moment data, failing to capture the dynamic nature of tactical maneuvers.
• The temporal and dynamic changes in a target's state necessitate more sophisticated approaches for accurate intention inference.

Purpose of the Study:

• To develop a novel method for aerial target combat intention recognition that overcomes the limitations of traditional approaches.
• To improve the accuracy and timeliness of intention recognition through deep learning and predictive modeling.

Main Methods:

• A hierarchical approach was used to construct an air combat intention characteristic set, encoded into numeric time-series data.
• A deep learning model based on bidirectional gated recurrent units (BiGRU) was employed for characteristic analysis.
• An attention mechanism was integrated to adaptively assign weights to characteristics, enhancing recognition accuracy.
• A characteristic prediction module was introduced to enable prediction of future intentions.

Main Results:

• The proposed method achieved 89.7% accuracy in recognizing aerial target combat intentions.
• The model demonstrated the capability to predict enemy intentions one sampling point ahead of time.
• The integration of BiGRU and attention mechanisms significantly improved the deep learning of combat characteristics.

Conclusions:

• The developed aerial target combat intention recognition method, incorporating BiGRU and attention mechanisms with predictive capabilities, offers a scientific and effective solution.
• The model's ability to predict intentions in advance provides significant value for real-time decision-making assistance in air combat scenarios.