Bird call recognition using deep convolutional neural network, ResNet-50

Conference Publication ResearchOnline@JCU
Sankupellay, Mangalam;Konovalov, Dmitry
Abstract

Birds are an important group of animal that ecologist monitor using autonomous recordings units as an crucial indicator of health of an environment. There is not yet an adequate method for automated bird call recognition in acoustic recordings due to high variations in bird calls and the challenges associated with bird call recognition. In this paper, we use ResNet-50, a deep convolutional neural network architecture for automated bird call recogni- tion. We used a publicly available dataset consisting of calls from 46 different bird species. Spectrograms (visual features) extracted from the bird calls were used as input for ResNet-50. We were able to achieve 60%-72% accuracy of bird call recognition using ResNet-50.

Journal

N/A

Publication Name

AAS2018: Acoustics 2018: hear to listen

Volume

N/A

ISBN/ISSN

N/A

Edition

N/A

Issue

N/A

Pages Count

8

Location

Adelaide, SA, Australia

Publisher

Australian Acoustical Society

Publisher Url

N/A

Publisher Location

Adelaide, SA, Australia

Publish Date

N/A

Url

N/A

Date

N/A

EISSN

N/A

DOI

N/A