A novel track initiation method based on rule knowledge and deep detection network

Radar target tracking plays an important role in reconnaissance, monitoring and defense. Clutter caused by complex electronic and natural environments poses many challenges to radar target tracking. The current track initiation method, which heavily relies on empirical knowledge, struggles with the increasing problem of false tracks and missed detections. To address this, a track initiation method based on rule knowledge and a deep detection network is proposed. In this method, radar echoes are converted into images that comprehensively reflect the different frames and target shapes, facilitating processing. With the help of the excellent detection, positioning and recognition ability of the deep learning network FasterRCNN, the candidate boxes where might exist target track in the echo image are extracted candidate boxes where targets might exist in the echo image are extracted. Additionally, the corresponding angle and wave gate are set as initiation rules. The combination of radar data learning and partial experiential rules reduces the need for accurate prior parameters in traditional methods. Furthermore, by considering the spatial and temporal information of the target comprehensively, it achieves a good balance between the correct initiation rate and false initiation rate. Simulation environments under different observation trajectories are constructed, and Monte Carlo simulation is used to compare the proposed method with the traditional track initiation method and other track initiation methods combined with machine learning and prior rules. Results show that our method can ensure the correct initiation rate of the target track and effectively suppress the false initiation rate.