Researchers have developed what they say is the first AI algorithm dedicated to detecting trafficked dead marine wildlife from 3D X-ray images. The system was most effective at finding species with idiosyncratic shapes, like shark fins and seahorses, but also detected sea cucumbers with 86% accuracy.
—
By Daniel Shailer
On Sunday, April 26, Argentine officials stopped an unusual shipment arriving at an airport near Buenos Aires. Inside, they found so many dead and dying fish, octopuses and crabs that a national rescue center had to install 10 new emergency tanks to support the survivors. It was the third time in a year authorities had seized an illegal shipment of sea life at the same airport, the Associated Press reported.
Marine wildlife trafficking is a growing global business, driven by demand for ornamental fish, luxury foods and traditional medicines. Much of that trade is routed through airplane luggage or airmail, where the vast majority of animals, dead or alive, go undetected.
The combined use of artificial intelligence (AI) and 3D X-ray machines could change that, according to an international team of researchers. Training an algorithm on samples of seahorses, shark fins and sea cucumbers, the scientists achieved successful detection rates between 86% and 96%, according to a research paper published last week.
“As it stands, our methods of detecting something that shouldn’t be in our bags on the front line is reliant on human inspection and biosecurity dogs,” Vanessa Pirotta, a marine biologist at Macquarie University in Australia and the paper’s lead author, told Mongabay. “AI could be used to complement that. It’s not a silver bullet, but an assistant and a tool.”
Other researchers and enforcement agencies said they welcomed the research, while stressing that it would only work as a complementary tool. “Detection is the first link in a longer chain, not the whole answer,” a spokesperson from the United Nations’ Office on Drugs and Crime told Mongabay in a written statement. “Technology can flag a bag. People, forensics and prosecutors turn a flagged bag into a sentence.”
The research follows a 2022 paper from the same team training an algorithm to spot terrestrial species.
Michelle Anagnostou, a research fellow with the University of Oxford’s illegal wildlife trade program who was not involved in the study, said innovations in detection like these were “exciting,” especially if they contribute to broader attempts to address trafficking systems at large. “We’ve been arresting people for decades and it hasn’t gotten us very far,” Anagnostou told Mongabay. She recommended thinking about it from a systems perspective, including resourcing enforcement agencies, educating the public and combating corruption and poverty in source countries.
In addition to driving biodiversity loss, illegal wildlife trafficking spreads infectious diseases, introduces invasive species and has been connected to other forms of organized crime and labor abuse. Trafficking of marine wildlife represents a growing and likely underreported share of the total illegal animal trade, according to Operation Thunder, a global crackdown effort coordinated by Interpol and the World Customs Organization (WCO). Last year, 91,000 pieces of trafficked marine life were seized, almost double the number of reptiles, birds and primates combined.
Despite this, the trade of marine animals goes underrecognized, according to Sarah Foster, a fisheries researcher at Canada’s University of British Columbia and member of the IUCN’s group protecting seahorses, pipefish and seadragons. “Our biggest challenge in ocean conservation writ large is getting people to recognize fish as wildlife, the way they care about elephant ivory or rhino horn,” she told Mongabay. “Marine [species] haven’t gotten the same attention that terrestrial illegal wildlife trade has — although I am pleased to see papers like this try and shift that, and come up with solutions.”
To train their algorithm – the first dedicated to detecting marine wildlife trafficking – Pirotta’s team collected 68 individual samples of seahorses, shark fins and sea cucumbers. Most of these were borrowed from the Australian Museum Collections, initially seized during real-world trafficking busts. One fin was collected fresh from a beached bull shark (Carcharhinus leucas). Another, of unknown species, was bought from an East Asian grocer in Sydney, along with some of the dried sea cucumber samples used in the study.
“I was just like every other tourist in Sydney going into Chinatown; it was a non-event,” Pirotta recalled. Both dried fins and sea cucumbers were easy to find. “But for me as a whale biologist it was just foreign — it was my out-of-water experience. The idea was not to support this whole market but to use a real-life example of what this could look like.”
Researchers spent six months “thinking how traffickers think,” Pirotta said, to create just under 6,000 “bags”: 3,500 with the samples hidden among and inside items typically used to disguise smuggled animal parts, including toys, clothes and tin foil. Another 2,400 bags were prepared without animal parts. These were then scanned by 3D X-ray machines produced by Rapiscan, a U.S.-based security company that funded the research.
Trained on those images, the algorithm had 95-96% success detecting shark fins and seahorses. It identified sea cucumbers 86% of the time. It threw false alarms for seahorses most often (9%), and for shark fins and sea cucumbers just 1-2% of the time.
The algorithm’s greater difficulty identifying sea cucumbers illustrates one potential limitation of this approach, according to Pirotta. “Fins and seahorses look the same, but there’s more variation [with] sea cucumbers,” she said. For that reason, an algorithm might have more trouble distinguishing between legal and illegal samples within the same species.
The algorithm also focused exclusively on small airborne luggage. “We know we’re missing a lot” in sea freight, said Anagnostou, who is currently working with AI to screen for suspect maritime shipments.
For now, the algorithm has only been tested on dead, mostly dried samples, and can only work with X-ray machines able to create 3D images in real time. “Not everyone around the world has access” to this technology, Pirotta acknowledged, but added that it is becoming more common.
Toby Breckon, a computer science professor specializing in analyzing X-ray images at the University of Durham in the U.K., told Mongabay that Pirotta was right to be optimistic about the spread of 3D scanning. “Globally all major airports have them for hold luggage now; hand luggage is coming a little more slowly but the tech is being driven by safety requirements, and this [work] could piggy back onto that.”
For Pirotta, the next step is to share the algorithm’s “recipe” to help detect additional species in other regions. “When I first started this work I never thought AI would be such an instrumental part of what I do as a scientist. Here I am talking about AI, knowing that it’s not going to be the answer to everything, but in an optimistic, progressive way.”
Featured image: Hamadi Mwamlavya via iNaturalist.
This article was originally published on Mongabay, written by Daniel Shailer, and is republished here as part of an editorial partnership with Earth.Org.
