Detection of fishes in turbulent waters based on image analysis

Abstract

This paper analyses the automatic fish segmentation problem in turbulent waters. To this end, a SOM neural network is used to detect fishes in images from an underwater camera system built in a vertical slot fishway, an hydraulic structure built in obstructions in rivers to allow the upstream migration of fishes. This technique allows the study of real fish behavior and may help to understand biological variables and swimming limitations of the fish species in high speed environments. This knowledge, may be used to replace traditional techniques such as direct observation or placement of sensors on the specimens, which are impractical or affect the fish behavior. To test the proposed technique, a ground true dataset was designed with experts and a series of assays have been performed where the results obtained with the proposed technique were compared with different segmentation techniques. © 2013 Springer-Verlag.