Automatic detection of behavioral patterns in herring from sonar data in response to external stimulus
Updated: 18 June 2016 - 2:57pm by J.C. Castillo
Start:2014 / End:2015
Principal investigator: José Carlos Castillo Montoya
NILS Science and Sustainability programme
Automatic detection of behavioral patterns in herring from sonar data in response to external stimulus

The problem of how fish
receive and collectively respond to stimuli presents a challenge for the
research community. Under conditions of particularly high density it becomes
intractable to maintain individual tracks and under such situations ”optical
flow” technique have proven to be highly suitable for quantifying fish behavior.
This project pursues a global aim, which is the
detection of behavioral patterns in herring from sonar data in response to
external stimulus
. This goal can be divided into three main tasks: a) To analyse
an actual fish species (herring) behavior; b) To adapt computer vision
algorithms to sonar for herring detection and tracking; and c) To develop a classifier
able to detect different behavioral patterns in herring. The use of computer
vision techniques implies the need for mathematical models such as statistical
approaches for classifiers and the multidisciplinary team involved includes
expertise in engineering, biology, computer science and mathematics.

The project includes several related objectives. The analysis
of collective behavior in fish is a challenging issue. For example, the
community is dealing wit the problem of understanding how an individual action
can scale into a collective phenomenon. An area of particular interest is to
understand how information is transferred in animals groups, and how stimuli is
amplified or dampened through social interactions. The project aims at creating an automated approach to detect
fish behavior
using sonar data
instead of traditional techniques, which mostly relied on human monitoring.
This work proposes the adaptation of computer vision techniques to a
challenging domain: acoustic sensing. This sensor will be used to detect behavior
patterns related to the collective information transfer in herring.


The project uses sonar recordings of herrings being exposed to a
wide range of stressors, including killer whale playbacks, a major predator for
herring, vessel noise, important for understanding how anthropogenic pressures
affects fish populations, range of tones and sweeps, for understanding how the
various properties of the sound affect fish, and, finally, predator models
mimicking direct attacks on the herring school. Sonar recording are being made
available through projects running in the host institution (the Norwegian
Institute of Marine Research, IMR). Similar recording from in situ experiment
conducted in Jupiter, Florida, will be available through IMR collaboration with
Florida International University.

The data is collected from a real
: the testbed developed for project “Collective behavior of
penned herring (CollPen)” at Austevoll Aquaculture station.


Finally, the results of this project are addressed to validate some of the hypotheses established
in CollPen project, such as
“to assess the impact of noise regulation on herring
population, both from an ecological and a commercial point of view”. This
project will also shed light on the impact of the different stressors on
herring welfare, which is closely related to the sustainability of the fishery.

More info can be found in

Journal Publications

Conference Publications



Doctoral Thesis