Attack Intent Recognition for Incoming Vehicles Based on Deep Learning

Attack Intent Recognition (AIR) for incoming vehicles is crucial for the effective execution of defense strategies and enhances the overall air defense system performance. The problem is addressed in this paper by leveraging the Artificial Neural Network (ANN), Joint Probability Calculation (JPC), and Interacting Multiple Model (IMM). First, the swarm engagement scenario is decoupled into multiple one-on-one pairs, with each pair treated as an independent event. An ANN is employed to compute the probabilities of attack and non-attack for the vehicle in each pair. Subsequently, the joint probabilities of each pair are calculated based on the principle of mutually exclusive events. Finally, an interacting operation is adopted to normalize these joint probabilities. The proposed scheme decouples the multi-class problem into multiple binary classifications, simplifying the network's fitting challenge. By introducing the non-attack probabilities, the method is able to deal with the non-preset conditions. Numerical simulation results demonstrate the advantages of the proposed method over traditional Gated Recurrent Unit (GRU), Kalman Filter (KF), and eXtreme Gradient Boosting (XGBoost) methods in terms of convergence speed and recognition accuracy. Specifically, it achieves a mean accuracy of 96.16% in 1.5 seconds, outperforming the GRU (3 seconds, 93.11%), KF (3 seconds, 85.82%), and XGBoost(1.5 seconds, 77.75%). These advantages enhance its practicality for real-world applications, especially in the dynamic and unpredictable environments.