Researchers from NTNU and Wageningen University have developed an artificial intelligence system capable of detecting salmon lice larvae in seawater far faster and more accurately than experienced biologists. The technology could improve parasite monitoring and support efforts to protect both farmed and wild salmon populations.
The researchers created a dataset of more than 120,000 high-resolution images of salmon lice larvae collected from real seawater samples and combined it with synthetic data to train AI models. The approach addresses one of the biggest challenges in automated detection – the lack of large, high-quality image datasets for training.
In trials, trained biologists required more than 30 hours over several days to identify 82% of salmon lice larvae in a large and complex seawater sample. The AI model completed the same task in just 30 minutes, correctly identifying 97.5% of the larvae.
Salmon lice remain one of the biggest threats to both the aquaculture industry and wild salmon stocks. Although the parasite naturally occurs in the ocean, the expansion of fish farming has created ideal conditions for its spread. A single fish farm can release millions of salmon lice larvae into surrounding waters every day, making reliable monitoring essential.
According to the researchers, current methods for counting salmon lice larvae are labour-intensive, expensive and often lack the accuracy needed for continuous monitoring. The new AI-based approach could provide more precise information on where larvae are present, how they spread and how effective different control measures are.
To build the training dataset, the research team collected and filtered several thousand cubic metres of seawater from fish farms and coastal areas near Ålesund, Norway. Because salmon lice larvae are relatively rare among countless other marine organisms, researchers also hatched lice in laboratory conditions and introduced them into seawater samples to create additional training material.
Using a custom-built video microscope, they recorded larvae at different developmental stages. Individual images were then processed to generate synthetic variations by rotating, resizing and combining them with images of other plankton species, allowing the AI to recognise lice under a wide range of conditions.
The researchers say the technology could improve forecasting models, provide more accurate assessments of parasite pressure in coastal waters and reduce uncertainty in Norway’s current salmon lice monitoring system, which largely relies on counting parasites found on farmed fish rather than measuring larvae directly in the sea. Better monitoring could also help fish farmers plan production more effectively and support decisions on where additional measures are needed to reduce the impact of salmon lice on wild fish populations.
