Sea-State-Aware Wake Segmentation for Maritime Surveillance

Vessels that disable or falsify Automatic Identification System (AIS) remain a central challenge for maritime security, especially in the case of small, fast craft. Automatically detecting these so-called ’dark’ ships in the complex environment of the sea surface is a task that cannot be accomplished by traditional methodologies. UEIKAP addresses this with a wake-centric, multimodal framework that exploits Sentinel-1’s day-night, all-weather imaging and fuses it with optical data, meteo-marine fields, AIS for validation, and physics-based synthetic wakes. A SAR dataset was assembled with polygonal annotations of wakes and hulls, and co-registered environmental descriptors to condition inference. Within the pipeline, SAR supports wake/hull segmentation and motion retrieval from wake geometry and azimuth-shift cues, while meteo-marine variables gate detections and explain visibility. This paper presents the SAR dataset and pre-processing, details the role of SAR within the workflow, and reports baseline results from SAR-only training. It also establishes the logic for integrating meteo-marine measurements with SAR and frames a comparative analysis between SAR-only and SAR with meteo-marine setups. Validation campaigns in Venice (2024) and in the Gulfs of Naples and Genoa (2025) are used to contextualize performance across sea states and to motivate environment-aware fusion for robust, non-cooperative vessel monitoring.