Using machine learning for detection and classification of fish-farm equipment measured using acoustics

Aquaculture food production grows faster than other major food production sectors, and in the Faroe Islands, salmon accounts for nearly half of the countries export value. In order to to keep up width the global trend, fish-farming in the Faroe Islands has moved from sheltered locations, to more exposed farming sites. Understanding the behaviour of fish-farming equipment and their inhabitants at exposed sites is important for ensuring fish welfare, and by extension, profits. Measurement equipment and methods make used today make it difficult to obtain an accurate description of the spatial extent and dynamics of the cage and the fish distribution and movement within, since the cages are very large and in exposed sites are subject to large deflections and deformations. Sonars have a comparatively high range, compared to optical cameras, but lack the ability to measure in different direction. Multibeam sonars allow for spatial information of its surroundings to be gathered, both in the spatial and temporal domain. The aim of this project is to develop methods to collect and extract spatial information about the extent of the cage, and the distribution of the biomass within. These methods can be used to get a better understanding of the behaviour of fish farming equipment and its inhabitants. Preliminary experimental results show that neural networks are able to detect and classify the surface and net in a single ping configuration.