Advanced machine learning for management of salmon farms

Production from wild-capture fisheries has remained at approximately 90m tons/year globally since 1990\. There is no significant scope for growth of fishing, and the FOA reports that a third of global stocks are already being fished beyond sustainable limits.

During this same time frame, aquaculture output has increased from less than 10m tons in 1990, to now providing half of all fish eaten around the world. The only feasible and sustainable option to meet growing global demand for fish is through continued growth of aquaculture.

Salmon farming is arguably the most advanced form of aquaculture, with 73% of the world's salmon production being farmed. Farming takes place in large nets in sheltered bays, and most farmed salmon comes from Norway, Scotland, Chile and Canada.

There are significant concerns over environmental contamination from salmon farms, in particular due to wasted feed contaminating the local environment. Moreover, especially with UK salmon farming, almost all of the suitable locations are already exploited. Salmon farming output is therefore largely limited primarily by the availability of licenses, which is in turn limited by the environmental impact of salmon farming. Thus the only practical measures to increase production are to improve productivity, and to minimise environmental impact, which can be achieved by reducing the quantity of wasted feed.

This project uses state-of-the-art computer vision and machine learning methods to optimise the feeding process. The aims are to improve productivity through optimal feeding (i.e., faster growth of the fish), and to reduce the amount of feed that is wasted, which is a major cost for fish farms as well as having environmental impact. Reduced environmental impact through improved farm management could potentially enable further growth of the aquaculture industry by allowing regulators to grant further licenses for new farm sites, or to increase the capacity of existing sites.