Haifeng Zhang

As global maritime security threats become increasingly stealthy, traditional methods such as Automatic Identification Systems (AIS), Synthetic Aperture Radar (SAR), and optical remote sensing reveal prominent limitations in coping with non-cooperative targets and extreme weather conditions. Distributed Acoustic Sensing (DAS) technology leverages submarine optical cables to construct large-scale passive sonar arrays, offering a novel, covert, and all-weather paradigm for wide-area maritime surveillance. However, constrained by the one-dimensional linear topology of submarine optical cables and the high nonlinearity and anisotropy of DAS spatiotemporal signals (waterfall plots), precisely resolving the four-quadrant headings of cargo ships under strong background noise remains a formidable challenge. To address these issues, this paper proposes a Spatiotemporal Anisotropic Decoupling Detection Transformer framework (STAD-DETR). The main contributions are threefold: First, a dual-branch preprocessing strategy combining Multidimensional Variational Mode Decomposition (MDVMD) and Frangi filtering is introduced to achieve the synergy of ocean wave noise suppression and ship wake ridge feature enhancement. Second, a Spatiotemporal Anisotropic Decoupling (STAD) module inspired by Cross Stage Partial (CSP) concepts is constructed to enhance the representation capability of asymmetrical spatiotemporal features through architectural innovation. Finally, a Direction-Aware Varifocal Loss (DAVFLoss) is designed, incorporating a directional penalty term to effectively resolve the issue of false alarms caused by heading reversals. Experimental results demonstrate that STAD-DETR achieves an $mAP_{50}$ of 95.91\% on the DAShip dataset, outperforming the baseline RT-DETR model by 1.35\%. This research provides high-precision technical support for tactical analyses, such as anti-smuggling operations and anomalous navigation monitoring.