Mouth Opening Frequency of Salmon from Underwater Video Exploiting Computer Vision

Abstract

Maintaining fish welfare is one of the most prevailing and challenging tasks in aquaculture production, in particular as there can be more than 200,000 fish present within a single sea cage. Observing this amount of individual animals is a very challenging task and requires a high degree of automated computer vision-based monitoring along with the use of advanced machine learning and AI algorithms. This paper introduces a computer vision based approach for determining the ventilation frequency in terms of the mouth opening frequency of salmon in aquaculture environments through underwater video analysis. We exploit an object detection network to detect specific salmon features including head, eye, snout, mouth and then apply optical flow analysis to the snout region to assess their ventilation rates, serving as a potential indicator of fish health and well-being. This aims towards analyzing salmon behavior on individual level in real-time, enabling an in-depth view of the population's status. Our methodology represents a significant progress towards a more automated salmon monitoring providing an objective measure to determine an important aspect offish behaviour in an effective way which will subsequently help to enhance efficiency in aquaculture operations. The results, obtained from analyzing annotated videos are very promising and validate the usability of our approach. This study also paves the way for further exploration into utilizing computer vision and machine learning for comprehensive fish status assessment, contributing valuable insights into sustainable aquaculture practices.