Abstract
Cage based salmon aquaculture has grown substantially over the last decades, however it is still, to a large degree, relying on experience-based production regime today. Advances in the digital transformation of the aquaculture industry will improve the ability to monitor, control and document the production systems, and facilitate knowledge-based decision making. In this paper, a combined instrumentation and interpreting solution is proposed and tested for monitoring fish distributions in aquaculture net cage. Farmed salmon in a full-scale sea cage are simulated by an individual-based fish model. And real-time behavioural changes are introduced and determined by data-driven parameter identification, thereby reflecting observed fish distributions from a set of single-beam echosounders. This forms a hybrid approach to combine the interpretability of physics-based models with the automatic pattern-identification capabilities of advanced deep learning algorithms. The performance of a tentative instrumental and model setup is evaluated by comparing with measured fish density data, while providing more detailed information such as the number and swimming speed of the fish, which are notoriously difficult to quantify using conventional solutions. The proposed hybrid approach is considered to be suitable for developing a more comprehensive fish monitoring system to be used on a daily basis in marine aquaculture.