Transformer Architecture Based Multi-frame TBD Algorithms for Maneuvering Targets

The multi-frame track-before-detect (MF-TBD) algorithm can effectively improve the tracking performance of the target in low signal-to-noise ratio (SNR) scenarios by considering all reasonable paths. However, when the target undergoes maneuvering motion, there will be a mismatch between the actual model and the assumed model. It leads to a sharp decline in algorithm performance. In this paper, a model-free trajectory sequence prediction network based on transformer architecture is studied for maneuvering targets. The network is trained using trajectory sequences from multiple types of target models to estimate the state of the target in different maneuvering characteristics. The trajectory sequence prediction network is integrated into MFTBD, replacing the target model, and the specific process of the improved MF-TBD algorithm is provided. The simulation results demonstrate the superior performance and flexible adaptability of the proposed algorithm.