Temporal fusion wins for dark vessels
Experts say identifying dark vessels comes from joining weak signals over time—SAR/EO detections, AIS gaps, RF and behavioral priors—rather than single images, with systems building and scoring vessel hypotheses as new observations arrive. Recommended architectures split real‑time candidate generation, stateful correlation for track maintenance, and analyst feedback loops to update confidence over revisits. (x.com)
A “dark vessel” is a ship that stops telling the world where it is, and experts say finding one usually takes many weak clues stitched together over time, not one decisive image. (fisheries.noaa.gov) The basic signal most systems start with is the Automatic Identification System, a radio beacon that broadcasts a vessel’s identity, position, speed, and course to other ships and shore stations. The European Space Agency says gaps in that signal leave “limitations” that create blind spots in maritime surveillance. (earth.esa.int) One way to fill those gaps is synthetic aperture radar, a type of satellite imaging that can spot metal targets at sea at night and through cloud cover. ICEYE says combining synthetic aperture radar with Automatic Identification System data can flag vessels detected by radar that were not visible to Automatic Identification System receivers. (iceye.com) Another layer is radio frequency sensing, which listens for emissions from shipboard systems such as radar, very high frequency radio, or satellite communications. The European Space Agency said in March 2026 that Unseenlabs’ 15-satellite constellation was built to add radio frequency detections that complement Automatic Identification System, synthetic aperture radar, and vessel monitoring data. (earth.esa.int) That is why practitioners talk about “fusion”: a system keeps a running hypothesis that a radar hit, a radio emission, and a missing beacon may all belong to the same vessel. Spire said its Maritime Custody Service pairs radio frequency detections with follow-up imagery and artificial intelligence analysis to monitor change over time. (spire.com) The same logic applies to behavior. NOAA said researchers identified more than 55,000 examples of fishing vessels disabling their Automatic Identification System transponders, and estimated that disabled beacons obscure about 6 percent of global fishing vessel activity. (fisheries.noaa.gov) Those historical patterns become priors, or starting odds, for the next pass: where vessels usually go dark, how long they stay silent, and what routes or fisheries are associated with that behavior. NOAA said its study broke beacon disabling down by location, fishery, and home country, which is the kind of context analysts use to score whether a new contact is ordinary or suspicious. (fisheries.noaa.gov) Governments and companies have been moving toward this multi-sensor approach for years. ICEYE and Spire announced a joint dark-vessel monitoring service in January 2019 built around satellite Automatic Identification System plus synthetic aperture radar, and newer services now add radio frequency collection and automated retasking of imaging satellites. (iceye.com) The pressure to improve detection is not limited to smuggling or sanctions evasion. The Food and Agriculture Organization says illegal, unreported, and unregulated fishing remains one of the greatest threats to marine ecosystems, and the United Nations Office on Drugs and Crime says the fisheries sector is also exposed to corruption, document fraud, tax evasion, and human trafficking. (fao.org) (unodc.org) Automatic Identification System data can also be manipulated, not just switched off. The European Maritime Safety Agency said this month that European Union member states have set up a dedicated working group on Automatic Identification System spoofing, a sign that the problem has moved from isolated anomaly to formal security issue. (emsa.europa.eu) So the current consensus is less about a single perfect sensor than about a workflow: generate candidates quickly, keep state on every possible track, and update confidence every time a satellite revisits the scene. In maritime surveillance, the vessel that disappears in one frame is often the one that becomes visible only after many frames are connected. (windward.ai)