In the cold, clear waters of Norway's fjords, a tiny parasite called the salmon louse has been a headache for fish farmers and wild salmon for years. These lice feed on the skin and blood of salmon, and when fish farms pack millions of salmon into net-pens, they create ideal conditions for lice to multiply and spread into the surrounding waters. Every year, between 400 million and 450 million salmon and rainbow trout fry are released into Norwegian fish farms — and a single farm can release millions of louse larvae into the fjords every single day. Fish farmers must count the lice on their fish regularly and report the numbers, with facilities rated on a "traffic light" system designed to protect wild fish from dangerous levels of infection.
Now, researchers have developed an AI tool that could make monitoring these parasites much faster and more accurate. A team from the Norwegian University of Science and Technology (NTNU) and Wageningen University in the Netherlands collected over 120,000 real images of salmon louse larvae floating in seawater. They also created synthetic images by editing and combining real data to expand their training dataset. Then they trained AI models to recognize the larvae among the ocean's plankton and particles — where, according to researcher Lars Christian Gansel, there can be hundreds of thousands or even millions of other organisms for every single louse larva.
The results were striking. In one large, complex seawater sample, experienced biologists took more than 30 hours spread over several days to identify just 82% of the salmon louse larvae present. The AI model did the same job in just 30 minutes — and correctly identified 97.5% of the larvae. The findings were published in the journal Computers and Electronics in Agriculture.
Gansel, who heads NTNU's Department of Biological Sciences in Ålesund, said better detection methods are essential for protecting both farmed and wild salmon. "If we are to succeed in eradicating salmon lice, the best approach is to prevent contact between the parasite and the fish," he said. "Our model makes it possible to obtain this information." The hope is that faster, more accurate counting will help farmers and regulators develop and evaluate better ways to keep lice under control — ultimately giving wild salmon populations a fighting chance.