In recent years, a high proportion of undersized saithe has caused concern both in the industry and among management authorities. Catching unwanted species and undersized fish leads to economic losses for fishers and weakens the long-term sustainability of the stock. There is therefore an urgent need for technology that can provide reliable estimates of species composition and size in a school before the net is set. Echosounders and sonar are currently key tools used by the purse seine fleet to detect and quantify fish in individual schools, but existing systems for species identification and size estimation are not sufficiently accurate.
The project will develop AI models for species and size estimation of saithe based on echosounder data. The assumption is that school characteristics, such as relative frequency response and school shape, will vary between size groups and can thus be exploited to estimate individual size within the schools. Where individual fish can be detected, size can be estimated directly from target strength. The models will be integrated into a robust system architecture for real-time operation on board without requiring continuous internet access.
