This paper analyses the automatic fish segmentation problema in turbulent waters. To this end, a SOM neural network is used to detect fishes in images from an underwater built in obstructions in rivers to allow the upstream migration of fishes.
This technique allows the stury of real fish behavior and may help to undertad biological variable 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 teh speciments, which are impractical or affect the fish behavior.
To test the proposed technique a ground true dataset was designed with expersts and a series of assays have been performed where the results obtained with the proposed technique were compared with different segmentation techniques.