Abstract: Abstract: The development and application of stealth technology have brought great challenges to radar target detec-tion. As one of the important technical directions for weak target detection, tracking-before-detection (TBD) technology is often used in radar detection. However, traditional TBD require a large amount of computation and are limited by target motion models, making it difficult to effectively detect maneuvering targets in various motion states. To address this issue, a TransUNet based intelligent TBD algorithm for weak targets is proposed. The algorithm is not limited by any motion model and can effectively identify maneuvering targets in various motion states. Firstly, utilizing the spatio-temporal correlation characteristics of the target motion trajectory, a rough detection algorithm for maneuvering weak targets in different motion states was designed based on the TransUNet model to achieve preliminary detection of weak maneuvering targets. Then, the preliminary detection results are used to perform regional search on the target measurement, and achieve backtracking of weak maneuvering target points. Finally, in response to the problem of missed detection in preliminary backtracking points under low signal-to-noise ratio (SNR), the auxiliary information of partially successfully detected points is utilized, combined with the optimal parabolic (OP) value function to achieve precise detection of target points and complete backtracking of tracks. Simulation experiments have verified that this method can effectively detect weak maneuvering targets in various motion states.

Key words: TBD, Weak target, TransUNet network, Spatiotemporal correlation characteristics, Target detection, Optimal parabolic value function