Development of an AI-based bioacoustic wolf monitoring system

Autor/innen

  • Olivier Stähli College of Management of Technology EPFL
  • Thomas Ost Life Science Dep. Supercomputing Systems AG Technoparkstrasse
  • Thomas Studer University of Bern

DOI:

https://doi.org/10.32473/flairs.v35i.130552

Schlagworte:

Convolutional neural network, edge computing, bioacoustics, wolf monitoring

Abstract

Wolves are spreading in the Alpine region at an increasing rate, which leads to human--wolf conflicts. In order to reduce those and to perform an active wolf management, solid information about the presence of wolves is required. Getting this information is challenging since wolves are nocturnal, have sharp senses, and large territories. A monitoring method is to detect wolf howling, which can be heard over several kilometres and therefore simplifies finding them. Current acoustic methods, however, are very labour-intensive as a memory card has to be fetched from the device in the field and then the recordings have to be checked for howling manually. We present a novel approach to acoustic wolf monitoring using a convolutional neural network that runs on an embedded system in the wolf territory. Thus, we obtain accurate real-time information about the presence of wolves. On our data set, we achieve an F1-score of 0.61, thus outperforming previous systems by far. We develop prototypes and conduct two field test: first in a zoo, where we even achieve an F1-score of 0.8, and then in a wolf territory, where we successfully detect wolf howling.

Downloads

Veröffentlicht

2022-05-04

Zitationsvorschlag

Stähli, O., Ost, T., & Studer, T. (2022). Development of an AI-based bioacoustic wolf monitoring system. The International FLAIRS Conference Proceedings, 35. https://doi.org/10.32473/flairs.v35i.130552

Ausgabe

Rubrik

Main Track Proceedings