Title: DarkVesselNet: Leveraging Multi-Modal Remote Sensing and Trajectory Logic for Identifying Dark Vessels

Abstract:

Detecting "dark vessels" necessitates the integration of self-reported Automatic Identification System (AIS) data with satellite-based observations captured via radar and optical sensors. To address this, DarkVesselNet introduces a comprehensive multi-modal remote sensing framework. This system synergizes Sentinel-1 Synthetic Aperture Radar (SAR) data, Sentinel-2 optical imagery, geospatial foundation model backbones, AIS trajectory analysis, TGARD-style gap detection mechanisms, and an anomaly detection head inspired by Pi-DPM.

The project makes the system accessible as a validated Python package and hosts it on a public Hugging Face Space. The accompanying paper details the sensor integration, backbone abstraction, data fusion pathways, the anomaly head architecture, and preliminary validation results. Current evidence supporting the system is primarily software-based, encompassing tests for SAR speckle filtering, optical band ratio calculations, Haversine distance measurements, TGARD gap emissions, sensor coregistration, backbone token dimensions, and differentiable anomaly scoring.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC