Maneuvering Targets Fusion Filtering for Multi-Source Heterogeneous Sensor Using GRU-Fusion Net

The ISAR imaging principle requires continuous time observation of maneuvering targets. The premise of continuous time observation is to track the target robustly and continuously. The data fusion technology of multi-source heterogeneous sensors is the key to improving the robustness of maneuvering target tracking. However, multi-source heterogeneous sensors face challenges in maneuvering target fusion tracking, including errors from time registration of data from different sources and incomplete accuracy in model representation. To address these issues, this paper proposes a GRU-Fusion Net method that combines multi-source heterogeneous sensor maneuvering target tracking algorithms with GRU networks and FC layers. This method applies Kalman filtering in each local filter to obtain the state estimation and error covariance matrix of the maneuvering target. Based on multi-source heterogeneous sensor fusion tracking algorithms and the maneuvering target state model, different network modules are designed to adapt to the cases where varying numbers of sensors have observations at the sampling points, thus processing local filter results to obtain the tracking results of the maneuvering target. Simulation results show that, compared to traditional multi-source heterogeneous sensor fusion tracking algorithms, the GRU-Fusion Net method effectively overcomes errors arising from time registration of data from different sources and inaccuracies in model representation, and improves the tracking accuracy of maneuvering targets.