Bayesian Filtering Enhanced Causal-Attention-LSTM for Tracking Highly Maneuvering Aerial Targets

With the increasing complexity of battlefield environments and increasing maneuverability of aerial targets, the motion uncertainty of battlefield targets necessitates novel target-tracking methods. In this study, we focused on the tracking problem of airborne multiple sensors for highly maneuvering targets. To address the problems of low tracking accuracy and low efficiency of the classical interactive multiple model tracking algorithm with respect to tracking highly maneuvering aerial targets, we propose a long short-term memory (LSTM) recursive tracking method for causal attention, based on the Bayesian filtering (BF) principle and deep learning. This method comprehensively considers the different characteristics of highly maneuvering aerial targets in terms of spatial and temporal dimensions, uses a causal convolution operation and an LSTM recursive network to extract the temporal and spatial characteristics of the target motion, and introduces a self-attention mechanism to capture the importance of the target trajectory at different times, to track maneuvering targets. Compared with conventional target-tracking methods, the proposed causal attention LSTM-based tracking method does not require prior information, such as the target’s motion model and noise covariance. Simulations show that the proposed algorithm tracks more accurately than existing target-tracking algorithms with respect to complex maneuvers of aerial battlefield targets.