Multi-target cross localization method for distributed range-Doppler only high frequency surface radar

The multi-target cross-localization using only range-Doppler measurements in a radar system is quite challenging due to the always unknown associations between the measurements, which is commonly referred to as the data association problem. Meanwhile, false associations can lead to the generation of ghosts. In this paper, we propose a multi-target cross localization method to address the data association problem in a distributed high-frequency surface wave radar system that relies solely on range-Doppler measurements. Firstly, a multi-start heuristic Hungarian algorithm is employed to obtain the initial candidate targets. Then, a unique tracking strategy is utilized to further eliminate false tracks by leveraging motion features. Furthermore, to handle associations under ambiguous conditions, we introduce an LSTM-based deep neural network to distinguish false tracks within each ambiguous association group. The proposed method is validated across multiple scenarios, taking into account the influence of target numbers, target densities, and detection conditions. The effectiveness of the proposed localization method is compared to traditional methods based on metrics such as the probability of detection, the optimal subpattern assignment (OSPA) error, and the cardinality error. Additionally, the tracking strategy is evaluated in terms of its ability to generate true target tracks and suppress false tracks throughout the tracking process.