A directional-aware attention mechanism-based approach for target trajectory extraction from bearing-time records

To address the issues of blurring and fragmentation in weak target bearing trajectories within bearing-time records for passive underwater acoustic detection in complex marine environments, this paper proposes a directional-aware attention U-Net framework for weak target trajectory extraction. The framework adopts a hierarchical encoder-decoder U-shaped architecture and incorporates the physical prior of the spatiotemporal extension of underwater acoustic target trajectories. On one hand, it introduces a dynamic direction sensitive detection module in the decoding stage, which decouples feature responses along the time and bearing dimensions through orthogonal anisotropic convolutions and employs a dynamic gating mechanism to adaptively enhance structures consistent with the target trajectory's extension direction. On the other hand, an adaptive feature enhancement module is introduced to suppress strong interference in the spatiotemporal domain, improve the semantic consistency of skip connection features, and focus on the continuous structure of weak trajectories. Experiments on both simulated and real-world data demonstrate that the proposed method significantly outperforms existing comparison methods in terms of trajectory continuity and distinguishability.