A Deep Learning-Based Radar Track Initiation Algorithm with Spatiotemporal Feature Fusion

Track initiation is a critical issue in multi-target tracking tasks, which can establish the initial motion trajectory of the target through radar scanning. With the advancement of radar technology and the increasing complexity of environments, track initiation often occurs in conditions with strong clutter. This situation frequently leads to difficulties in initiating true tracks while false tracks are more readily initiated. To address this issue, this paper proposes a deep learning-based track initi-ation algorithm for radar measurement spatiotemporal feature fusion, designed for application in strong clutter environments. Firstly, the candidate set is determined from the combination of radar measurements. The spatiotemporal vectors are derived from the candidate set, which is used as the input of the gated recurrent unit (G RU) model to obtain the temporal and spatial features, and the fusion features are obtained by merging them. The self-attention module assigns importance weights to the fusion features, and finally, the softmax classifier is used to distinguish between true and false tracks to complete the track initiation. Simulation results show that the proposed algorithm can effectively improve the true track initiation rate and suppress the formation of false tracks with low time loss in a strong clutter environment.