A Maneuver Detection LSTM and Fuzzy Neural Network-Based Approach for High-Maneuver Target Tracking

Tracking highly maneuverable targets is challenging due to model mismatches and motion uncertainties. This paper proposes a neural network-based framework for highly maneuverable target tracking, named Maneuver Detection Tracker (MD-Tracker), which consists of a prediction module based on Long Short-Term Memory and a fuzzy neural network update module. The prediction module extracts maneuvering features from residual information via the Maneuver Detection Unit (MDU) to adaptively predict the target state, while the update module employs fuzzy rules to adjust the Kalman gain and correct the estimation. To validate the effectiveness of the proposed method, a three-dimensional dataset of maneuvering target trajectories encompassing multiple motion models is constructed. Ablation studies and comparative experiments are conducted to evaluate the contribution of the MDU module and the overall estimation performance of the model. In highly maneuverable target scenarios, MD-Tracker demonstrates superior performance compared to current state-of-the-art algorithms.