Fish Monitoring in Aquaculture Using Multibeam Echosounders and Machine Learning

Abstract

Offshore salmon aquaculture is a growing industry that faces challenges such as sea lice infestations and varying environmental conditions, necessitating the development of new monitoring systems to improve fish welfare and sustainability. In this paper, we propose and test a machine learning based method for underwater detection and localisation using multibeam echosounders (MBES) in fish farming applications. We demonstrate a three-stage process involving data acquisition, pre-processing, and object detection. We then compare the performance of four different vision based deep learning object detection algorithms in different signal-to-noise scenarios by artificially adding noise to the pre-beamformed signals. This method successfully detects fish in MBES images, which has potential applications in optimising feeding schedules, behaviour analysis, and fish health monitoring. Furthermore, this method holds potential for the detection and tracking of other objects within fish farms, such as cages and mooring lines. This study paves the way for further development of MBES data being used as a non-invasive and automated monitoring method in aquaculture.