EXCLUSIVE: AI sorts deformed fish earlier

Aquaticode has launched an artificial intelligence system designed to automate the detection and removal of deformed juvenile fish, targeting a task that remains largely manual in sea bass and sea bream hatcheries.

The Norwegian aquaculture technology company says its AquaLens system can inspect and sort up to 350,000 fish a day, identifying deformities at a much earlier stage than conventional manual grading.

Deformities are a significant cost for Mediterranean fish hatcheries. A 2023 study cited by Aquaticode estimates that skeletal deformities affect 7% to 20% of juveniles on average, with severe deformities costing Mediterranean hatcheries between €23 million and €116 million annually.

“Deformities are a well-known cost in bass and bream,” Stian Rognlid, chief executive at Aquaticode, told World Fishing & Aquaculture. “Buyers also set hard limits on deformities, including on any single type. To stay inside them and cover transport losses, producers tell us they grow around 20 percent more fish than they need.”

That makes sorting unavoidable. Hatcheries typically employ teams of 15 to 30 people to inspect fish by eye, but Rognlid said manual grading cannot provide the same level of consistency at commercial scale.

“A good grader can spot a deformity if they get time with each fish. What nobody can do is give every fish that attention at commercial scale, shift after shift, to the same standard at every site,” he said.

AquaLens is designed to inspect juveniles weighing between one and 10 grams. Cameras capture multiple views of each fish while deep-learning models assess characteristics including the operculum, jaw, nose, spine and dorsal and caudal fins. Fish are then directed into commercial or non-commercial groups using a soft-touch sorting process.

The system can process up to 44,000 fish an hour, according to Aquaticode. Because it operates on smaller juveniles, producers can remove fish unlikely to meet buyers’ specifications before they consume significant amounts of feed or occupy valuable tank space.

“We don’t publish a single survival or growth figure, because it depends on how many deformed fish a batch has and how early they come out,” Rognlid said. “What farmers see is fewer surplus fish, less feed and tank space spent on fish that won’t make grade, and staying inside the buyer’s limits without over-producing.”

The company says AquaLens is intended to improve labour efficiency as well as economics. Rather than requiring hatcheries to invest in equipment upfront, the system is offered as a service priced per fish sorted, allowing the cost to scale with production.

The technology also aims to address concerns around consistency and AI performance. Aquaticode measures accuracy using each customer’s own fish, comparing the system’s results by trait, species and fish size against fish assessed manually.

“The models are only as good as the data behind them,” Rognlid said. “We also routinely add to the models, for example when a hatchery brings in a new genetic line or a batch that looks different.”

AquaLens runs locally at hatcheries rather than relying on cloud processing, with dedicated lighting and optics designed to maintain consistent conditions. Aquaticode says new deformity traits can also be introduced through software updates rather than hardware changes.

The system builds on infrastructure developed for Aquaticode’s SORTpro salmon sorting technology, launched in 2023. The company has agreements with Cooke ESPAÑA and Turkey’s Ilknak covering the phenotyping and sorting of 300 million sea bass and sea bream.

“We’ve done this in salmon. Now we’re doing it in bass and bream, and other species will follow,” Rognlid said.