Generalized Radon-Fourier transform (GRFT) is a classical long-time coherent integration method for radar maneuvering target detection. GRFT, whose core is to achieve motion parameter estimation via searching, can almost reach the optimal detection performance but heavily suffers from the high computational cost. Motivated by the fact that motion parameter estimation is essentially a non-linear mapping from the radar echo to the target’s motion parameters, one can use a deep neural network (DNN), a kind of modeling tool that can learn complex mappings from training data, to directly realize this mapping, thus alleviating the heavy computational burden brought by the searching efforts. Based on this idea, a DNN-aided long-time coherent integration algorithm, which can be viewed as a fast implementation of GRFT, is proposed in this paper. More specifically, we first use a pre-trained DNN to roughly estimate the motion parameters of the target to be detected from the radar echo, and then accomplish the coherent integration of the target for detection via a fine grid search in the neighborhood of the obtained rough estimation results. Finally, simulation results demonstrate that the proposed algorithm can achieve the detection performance close to that of GRFT but with a much lower computational cost.

通用Radon-Fourier变换（GRFT）是用于雷达机动目标检测的经典长时间相干积分方法。GRFT的核心是通过搜索实现运动参数估计，它几乎可以达到最佳检测性能，但是却遭受了高昂的计算成本的困扰。由于运动参数估计本质上是从雷达回波到目标运动参数的非线性映射，因此可以使用深度神经网络（DNN），该模型工具可以从训练数据中学习复杂的映射，直接实现这种映射，从而减轻了搜索工作带来的沉重计算负担。基于此思想，本文提出了一种基于DNN的长时间相干集成算法，可以看作GRFT的快速实现。更具体地说，我们首先使用预训练的DNN从雷达回波粗略估计待检测目标的运动参数，然后通过在获得的附近进行精细网格搜索来完成待检测目标的相干积分粗略估算结果。最后，仿真结果表明，所提算法可以达到与GRFT接近的检测性能，但计算成本较低。