Irregularity Detection in Net Pens Exploiting Computer Vision

Abstract

Protecting the remaining wild salmon stock in Norway is of utmost importance and requires that farmed salmon cannot escape from aquaculture sites. As holes in net-cages are responsible for a large fraction of the escaped salmon the industry has to perform frequent inspections of the _sh cage integrity. In this paper we propose an image processing and computer vision based attention mechanism towards a more automated _sh-cage inspection. The presented algorithm allows to indicate areas in videos showing net-pen locations where potential holes are present. We show the e_ectivity of the approach on video-recordings of holes also in commercial _sh-cages.