A model-and-data-driven method for radar highly maneuvering target detection

This article addresses the coherent integration problem for detecting a highly maneuvering radar target with the range migration (RM) and Doppler frequency migration (DFM). Although many research efforts have been devoted toward this problem, how to strike a good balance between the detection performance and the computational complexity is still a challenge yet. In the algorithm, proposed in this article, a neural network is first developed to directly infer the target trajectory from the radar echo, and then the target energy distributed along the inferred trajectory is accumulated via the dechirp technique for detection. Since the proposed algorithm corrects the RM and compensates the DFM via the data-driven and model-driven approaches, respectively, we argue that the proposed algorithm operates in a model-and-data-driven approach. Besides, we fully integrate the domain knowledge into the development of the neural network, and simulation results suggest that this practice helps improve the detection performance of the proposed algorithm. Finally, numerical experiments are provided to show the high detection performance and computational efficiency of the proposed algorithm. Furthermore, we visualize the learned information of the neural network and find that it accords with our domain knowledge, demonstrating the rationality of the neural network's predictions.