Space-based deep learning multi-model tracking of hypersonic targets

Hypersonic glide vehicles (HGVs) may adopt S-shaped or C-shaped maneuvering trajectories to avoid no-fly zones (NFZs) during the approach to the target, leading to complex motion trajectory. This renders a single filtering model unable to stably track the HGVs. To this end, a deep learning-based multi-model selection (DL-MMS) filtering algorithm is proposed to improve the matching accuracy between kinematic models and the HGVs’ motion states in multi-model filtering methods. First, multiple kinematic models are used for tracking and filtering different motion states of the HGV. Second, error analysis is conducted to determine the optimal mapping relationship between the target’s motion states and the filtering kinematic models. Finally, a neural network is utilized to learn and memorize the optimal mapping relationship. Simulation results show that the proposed method can achieve high-precision tracking of HGVs, and the filtering error is significantly smaller than that of the interactive multi-model (IMM) filtering method under the condition of using the same sub-model.