V3I7P45

An Adaptive Transformer-Based Trajectory Prediction Framework for Highly Maneuvering UAV Targets in Air Defense Radar Systems

Nguyen Manh Trung*¹

Abstract

Accurate trajectory prediction of highly maneuvering unmanned aerial vehicle (UAV) targets is a critical task in modern air defense radar systems, directly affecting target tracking, threat assessment, and missile guidance performance. However, conventional prediction methods based on Kalman filtering and recurrent neural networks often struggle to maintain high accuracy when UAVs perform abrupt maneuvers, rapid heading changes, or variable-speed flight due to their limited ability to capture long-range temporal dependencies and complex nonlinear motion patterns. To address these challenges, this paper proposes an Adaptive Transformer-Based Trajectory Prediction Framework for highly maneuvering UAV targets. The proposed framework employs a Transformer architecture to model long-term temporal correlations in radar observation sequences while introducing an adaptive maneuver-awareness mechanism that dynamically adjusts the attention weights according to the estimated maneuver intensity of the target. Furthermore, a physics-guided prediction strategy is incorporated to constrain the predicted trajectories within realistic UAV kinematic limits, thereby improving prediction stability and physical consistency. The proposed method is evaluated using simulated air defense radar scenarios involving straight flight, coordinated turns, acceleration, deceleration, zigzag maneuvers, and high-g evasive actions. Its performance is compared with several representative trajectory prediction algorithms, including the Kalman Filter (KF), Extended Kalman Filter (EKF), Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and the standard Transformer model. Experimental results demonstrate that the proposed framework consistently achieves lower prediction errors, faster convergence, and greater robustness under highly maneuvering conditions while maintaining computational efficiency suitable for real-time air defense applications. These results indicate that the proposed framework provides an effective solution for next-generation intelligent radar target prediction and can significantly enhance the performance of air defense surveillance and interception systems.

Keywords:

UAV trajectory prediction; Transformer network; adaptive attention; maneuver-aware prediction; air defense radar; target tracking; deep learning.