Session: | A02-05: Terrestrial bio-acoustics - Part II |
Date: | Thursday 14 September 2023 |
Time: | 08:20 - 08:40 |
Title: | Automatic detection of indrisÂ’ songs using convolutional neural networks |
Author(s): |
D. Valente, University of Turin, Dept. Life Sciences and Systems Biology, Via Accademia Albertina 13, 10123 Torino, Italy
D. Ravaglia, University of Turin, Dept. Life Sciences and Systems Biology, Via Accademia Albertina 13, 10123 Torino, Italy
V. Ferrario, University of Turin, Dept. Life Sciences and Systems Biology, Via Accademia Albertina 13, 10123 Torino, Italy
C. De Gregorio, University of Turin, Dept. Life Sciences and Systems Biology, Via Accademia Albertina 13, 10123 Torino, Italy
F. Carugati, University of Turin, Dept. Life Sciences and Systems Biology, Via Accademia Albertina 13, 10123 Torino, Italy
T. Raimondi, University of Turin, Dept. Life Sciences and Systems Biology, Via Accademia Albertina 13, 10123 Torino, Italy
W. Cristiano, University of Turin, Dept. Life Sciences and Systems Biology, Via Accademia Albertina 13, 10123 Torino, Italy
V. Torti, University of Turin, Dept. Life Sciences and Systems Biology, Via Accademia Albertina 13, 10123 Torino, Italy
A. Von Hardenberg, Conservation Biology Research Group, University of Chester, Chester 4BJ, UK, CH1 Chester, UK
J. Ratsimbazafy, GERP, Fort Duchesne, Antananarivo 101, Madagascar, 101 Antananarivo, Madagascar
C. Giacoma, University of Turin, Dept. Life Sciences and Systems Biology, Via Accademia Albertina 13, 10123 Torino, Italy
M. Gamba, University of Turin, Dept. Life Sciences and Systems Biology, Via Accademia Albertina 13, 10123 Torino, Italy
|
Pages: | 4339-4342 |
DOI: | https://www.doi.org/10.61782/fa.2023.1138 |
PDF: | https://dael.euracoustics.org/confs/fa2023/data/articles/001138.pdf |
|
Conference proceedings
|
|
Abstract
The combination of bioacoustic monitoring with machine learning algorithms is
increasingly used to gain information on species distribution or activity. Among
these methods, convolutional neural networks (CNN) for the automatic
classification of both environmental and animal sounds proved particularly
effective. We employed an automated classifier based on a CNN aimed at detecting
the presence of Indri indri songs recorded in Maromizaha Forest from 2019 to 2021
via passive acoustic monitoring. The network achieved high accuracy (>90%) and
recall (>80%) values in assessing the songs presence while the use of data
augmentation and transfer learning was able to generalize to unsampled periods.
Lastly, our process was able to correctly describe both daily and annual pattern
of indris’ singing behavior, critical piece of information to plan data collection
and conservation practices.
|