Dark Vessel Detection Agent: Evidence-Driven Screening and Verification Task Generation Using Satellite Imagery and AIS
Year: 2026
Type: article
Abstract: Dark vessels that deactivate their Automatic Identification System (AIS) or conceal their identities pose substantial challenges to maritime monitoring. A central scientific problem is whether satellite–AIS mismatch can be interpreted as direct evidence of non-cooperative vessel activity, or whether its evidential meaning must be resolved through spatiotemporal AIS support and maritime geographic context. To address this problem, we develop an evidence-driven Dark Vessel Detection Agent that formulates suspected dark-vessel screening as a target-level evidence-interpretation process. The framework integrates Gaofen-6 vessel-like target detection using YOLOv11 oriented bounding boxes, historical AIS-derived navigation context, rule-based geospatial reasoning, DeepSeek-V4-assisted evidence-consistency review, and image-time AIS trajectory rechecking. In the Pearl River Estuary, China, 825 AIS-unexplained targets were reduced to 361 reviewed candidates. Image-time trajectory assessment subsequently identified 114 candidates as AIS-spatiotemporally explainable, while 247 remained unresolved and were retained as prioritized verification tasks, reducing the initial candidate pool by 70.06%. Selective AIS querying reduced query volume by 32.41% relative to exhaustive rechecking and identified 65.14% of the cases meeting the closure criterion. These results show that satellite–AIS mismatch is not a self-sufficient indicator of non-cooperative vessel activity. Its evidential meaning emerges from the joint interpretation of remote-sensing observations, image-time AIS trajectory support, and maritime geographic context.
Source: Remote Sensing
DOI: 10.3390/rs18223942
Authors: Yuehan Wang(Conceptualization, Formal analysis, Investigation, Methodology, Software, Visualization, Writing – original draft), Fenzhen Su(Funding acquisition, Project administration, Supervision, Writing – review & editing), Yinxia Cao(Writing – review & editing), Nan Xu(Writing – review & editing), Fengqin Yan(Conceptualization, Formal analysis, Methodology, Project administration, Software, Supervision, Visualization, Writing – original draft, Writing – review & editing)
Language: English
Subfield: Ocean Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $3,258
OpenAlex ID: W7235324399
