AI Case Study
SkyTruth researchers identify transshipment vessels to aid monitoring of suspicious nautical activity using CNNs
Researchers develop a public database of transshipment vessels which indicate potential illegal nautical activity including trafficking and illegal fishing. This was done by analysing ship automatic identification system messages using convolutional neural networks to identify suspicious movement patterns.
Freight And Logistics
From Global Fishing Watch: "Previously, no public, global database of transshipment vessels existed. So, as a first step to understand global transshipment activity, we developed one, combining data from vessel registries, hard-nosed internet investigations, and applying machine learning techniques to identify potential transshipment vessels." Researchers from SkyTruth, Google and Global Fishing Watch used convolutional neural networks to identify vessels which could not be identified through fishing vessel databases. This was done based on movement pattern analysis, as explained by Frontiers in Marine Science: "Vessels that were identified as likely transshipment vessels by the neural network were manually validated through web searches and RFMO registries. Vessel identities were further corroborated via the IMO [International Maritime Organization] as nearly all vessels could be matched to an IMO registry number."
From Frontiers in Marine Science: "A transshipment occurs when two vessels meet to exchange cargo, supplies, or personnel, often between vessels at sea and far from a home port. By allowing fishing vessels to offload catch at sea and remain on the fishing grounds, transshipment consolidates fuel costs within a fleet and moves product to market more efficiently. However, transshipment also introduces concerns over traceability and transparency in the seafood industry. The Food and Agriculture Organization (FAO) of the United Nations estimates over 15% of annual global catch is illegal, unreported, or unregulated (IUU). Transshipment has also been linked to human trafficking and can allow captains to keep their crew at sea indefinitely, resulting in de facto slavery."
This has resulted in the "first public, carrier vessel database [which] includes roughly 680 vessels, predominated by large vessels operating within Russian waters or the high seas tuna/squid fleets" according to Global Fishing Watch.
The vessel identification technique, as described in Science: "We processed 22 billion global AIS positions from 2012 to 2016 and trained two convolutional neural networks (CNNs): one to identify vessel characteristics and a second to detect AIS positions indicative of fishing activity. The vessel characterization CNN was trained on 45,441 marine vessels (both fishing and nonfishing) that were matched to official fleet registries. The resulting model identifies six classes of fishing vessels and six classes of nonfishing vessels with 95% accuracy... The fishing detection model was trained on AIS data from 503 vessels and identified fishing activity with >90% accuracy."
According to the Science research paper, "22 billion global AIS [automatic identification system] positions from 2012 to 2016".