Conservationists want to deploy seafaring artificial intelligence to fight against unlawful commercial fishing.
Fisheries observers often work 24/7, spending weeks at a time on vessels like longline tuna boats hundreds of miles offshore in the Pacific Ocean. It can be a thankless job, especially when fishermen disdain hosting observers who track every fish caught by a boat’s crew. That includes unintended hauls known as bycatch, which can be illegal and must be reported to authorities.
It’s estimated that nearly 20 percent of the seafood served throughout the world is harvested illegally. And most of this theft occurs on licensed boats, not renegade vessels.
A new artificial intelligence (AI) monitoring system — think facial recognition for fish — could soon aid fisheries observers in their quest to keep fish populations healthy and thriving in the world’s oceans.
[Read related story about AI: Does Whale Snot Hold the Secret to Ocean Health?]
AI on the High Seas
Tech-savvy conservationists from The Nature Conservancy (TNC) are advancing the use of AI-equipped video monitors on commercial fishing vessels. Machine learning software can determine the species of each fish as it is brought aboard, based on its size, shape and color.
Rather than assigning human observers to fishing boats, fisheries regulators could rely on a digital camera and software to review footage of a boat’s haul to determine its contents and legality.
The idea was initially met with great skepticism, said Mark Zimring, who directs TNC’s Indo-Pacific Tuna Program. Understandably so, since AI image recognition technology is still evolving. It’s challenging to operate reliably in controlled, static environments using white screen backgrounds or a consistent flow of objects down a conveyor belt.
“[But] we’re seeking to deploy it in an incredibly dynamic environment with salt spray on the lens, and nighttime and daytime [lighting], and fish moving all over the place,” said Zimring.
Yet it’s a big advance from the current monitoring model where international regulatory bodies rely on a small band of hardworking onboard observers to report illegal fishing activity.
Onboard video cameras, also called electronic monitors (EM), could take the place of human observers. EM cameras are linked to a vessel’s communications system. If someone tampers with or covers the lens, an alert is sent to local fishing regulators.
At the end of a fishing trip, it can take hundreds of hours to review the footage collected from a single outing, so TNC began developing AI algorithms that essentially operate like facial recognition for fish.
Teaching Machines to Identify Fish
TNC launched a contest on Kaggle — a crowdsourcing site that uses competitions to advance data science — and figured if 100 teams submitted code, the contest would be successful. Instead, nearly 2,300 teams entered, making it one of the most popular contests Kaggle has hosted. TNC split a $150,000 purse across the top five winning teams. TNC then partnered with EM manufacturer Satlink to test the effectiveness of the top Kaggle entries.
“Our goal is to have clear examples and a set of web services that current electronic monitor vendors can hook into their existing workflows,” said Matt Merrifield, who leads TNC’s Geographic Information Systems (GIS) group. “We don’t want to reinvent the EM workflow, but make a tool that augments it.”
The software will learn from false identifications and improve its accuracy over time, said Chris Rodley, CEO of Snap Information Technologies, a New Zealand-based company that helped TNC refine the AI algorithms.
The machine learning algorithms use archival EM footage to learn what each fish species looks like from a variety of angles and under a wide range of lighting conditions. The program then relies on human observers to confirm or correct how well the system identifies fish in the video.
Rodley and others believe this system could cut EM reviewing time from 40 hours down to a handful of hours. That could free up time for observers to manage other aspects of their work.
Using AI for Sustainable Fishing
Zimring said that TNC’s goal is to improve transparency in the fishing industry and bring accountability to fleets breaking the law. If AI software proves effective in reducing labor costs and identifying catch with high accuracy, its use could spread.
It’s something large commercial fishing companies could choose to use if they want to boost their transparency in order to gain a competitive advantage. Or large retailers could begin to require the use of EM systems on all their vendors’ fleets to verify lawful fishing practices.
AI has great potential but it faces significant practical hurdles, according to Francisco Blaha, a former fisherman who now is a consultant to governments and regulatory groups. He said technology providers will need to work closely with regulators to agree on a legal framework in which AI-derived video evidence will be placed. This ranges from “the validity of the electronic evidence to the margin of error that is acceptable.”
To make it work, developers will need to spend considerable resources honing the AI, especially given the highly unpredictable lighting and movements aboard fishing boats, which will constantly shift the content the software is analyzing, said Blaha.
Plus, since the push for the technology is coming from TNC and EM vendors, rather than government regulators, there is no set timeline in which it must be deployed, if at all. Still, Blaha believes it is worth the effort because seafood is a critical protein source around the world, and the industry needs more transparency and accountability.
Putting human observers onboard boats simply does not scale, said TNC’s Zimring. Facial recognition software for fish could emerge as a proficient robo-cop that never gets seasick and may, in the end, help stop illegal fishing.
Feature image photo credit: Mary Catherine O’Connor.