The salmon farming industry has increased host fish availability, allowing salmon lice populations to grow rapidly. Researchers captured over 120,000 images of louse larvae in seawater to train artificial intelligence models. These models identified the parasites more quickly and accurately than trained biologists.
Lepeophtheirus salmonis has long affected wild salmonids, but expanded aquaculture has created favorable conditions for the parasite. Despite extensive efforts in research and regulation, the resilient louse continues to threaten wild fish and pose challenges for the industry. The findings appear in Computers and Electronics in Agriculture.
Scientists from the Norwegian University of Science and Technology and Wageningen University created a new detection approach. They combined real images of larvae with synthetic data generated by editing and recombining visuals to expand training sets for AI systems. The resulting models improve efficiency in locating lice at sea.
In tests, biologists required more than 30 hours across multiple days to find 82 percent of larvae in a complex sample. The AI completed the same task in 30 minutes while detecting 97.5 percent of the organisms.
Researcher Lars Christian Gansel noted that multiple control measures are often needed together. Additional data on larval distribution can help assess current methods and support new strategies. The model enables collection of such information.
Norway releases hundreds of millions of salmon and trout into net pens annually. A single farm can release millions of larvae daily. The traffic light system regulates production based on lice counts to limit mortality in wild populations below 10 percent.
Preventing contact between parasites and fish remains the most effective approach. Detecting drifting larvae supports evaluation of preventive techniques. Salmon lice larvae are rare relative to other plankton, requiring large water volumes for accurate monitoring. Previous counting methods have often proved slow, imprecise, and costly.
