Optimising Convolutional Neural Network Architectures for Fin Whale Pulse Detection in Spectrograms

Deep neural networks are widely used for image classification in different fields, although selecting an appropriate architecture often remains a trial-and-error process. The purpose of this work is to investigate a convolutional neural network architecture used to detect whale pulses in spectrogram...

ver descrição completa

Detalhes bibliográficos
Autores: Román Ruiz, Marta, Rossi, Claudio
Formato: artículo
Estado:Versión publicada
Fecha de publicación:2026
País:España
Recursos:Consejo Superior de Investigaciones Científicas (CSIC)
Repositorio:DIGITAL.CSIC. Repositorio Institucional del CSIC
OAI Identifier:oai:dnet:digitalcsic_::c57e35e43eace3ae96caa545fa4f6214
Acesso em linha:http://hdl.handle.net/10261/429805
Access Level:acceso abierto
Palavra-chave:layer analysis
model optimisation
bioacoustics
spectrogram classification
id ES_0d56ff1e4fbda3c9a11162a1fef3cb95
oai_identifier_str oai:dnet:digitalcsic_::c57e35e43eace3ae96caa545fa4f6214
network_acronym_str ES
network_name_str España
repository_id_str
spelling Optimising Convolutional Neural Network Architectures for Fin Whale Pulse Detection in SpectrogramsRomán Ruiz, MartaRossi, Claudiolayer analysismodel optimisationbioacousticsspectrogram classificationDeep neural networks are widely used for image classification in different fields, although selecting an appropriate architecture often remains a trial-and-error process. The purpose of this work is to investigate a convolutional neural network architecture used to detect whale pulses in spectrograms in order to better understand the causes of its underperformance. By examining the behaviour of its internal layers, we show that the early convolutional blocks capture the most informative acoustic features, while deeper layers provide limited additional benefit and, under the considered training conditions, may even degrade classification accuracy. Based on these observations, we derive a simplified architecture consisting of only the first two convolutional layers followed by a lightweight classifier. This network achieves near-optimal performance, improving accuracy from 87% to 98%, and exhibits substantially lower variability between repetitions compared to the original model.The authors acknowledge the funding of the TRIDENT project (Grant Agreement No. 101091959) of the HE Framework Programme.Peer reviewedMultidisciplinary Digital Publishing InstituteEuropean CommissionRomán Ruiz, Marta [0000-0002-1067-8865]Rossi, Claudio [0000-0002-8740-2453]Consejo Superior de Investigaciones Científicas [https://ror.org/02gfc7t72]202620262026info:eu-repo/semantics/articlehttp://purl.org/coar/resource_type/c_6501Publisher's versioninfo:eu-repo/semantics/publishedVersionhttp://hdl.handle.net/10261/429805reponame:DIGITAL.CSIC. Repositorio Institucional del CSICinstname:Consejo Superior de Investigaciones Científicas (CSIC)Inglés#PLACEHOLDER_PARENT_METADATA_VALUE#info:eu-repo/grantAgreement/EC/HE/101091959https://doi.org/10.3390/app16052345Síinfo:eu-repo/semantics/openAccessoai:dnet:digitalcsic_::c57e35e43eace3ae96caa545fa4f62142026-05-22T06:33:51Z
dc.title.none.fl_str_mv Optimising Convolutional Neural Network Architectures for Fin Whale Pulse Detection in Spectrograms
title Optimising Convolutional Neural Network Architectures for Fin Whale Pulse Detection in Spectrograms
spellingShingle Optimising Convolutional Neural Network Architectures for Fin Whale Pulse Detection in Spectrograms
Román Ruiz, Marta
layer analysis
model optimisation
bioacoustics
spectrogram classification
title_short Optimising Convolutional Neural Network Architectures for Fin Whale Pulse Detection in Spectrograms
title_full Optimising Convolutional Neural Network Architectures for Fin Whale Pulse Detection in Spectrograms
title_fullStr Optimising Convolutional Neural Network Architectures for Fin Whale Pulse Detection in Spectrograms
title_full_unstemmed Optimising Convolutional Neural Network Architectures for Fin Whale Pulse Detection in Spectrograms
title_sort Optimising Convolutional Neural Network Architectures for Fin Whale Pulse Detection in Spectrograms
dc.creator.none.fl_str_mv Román Ruiz, Marta
Rossi, Claudio
author Román Ruiz, Marta
author_facet Román Ruiz, Marta
Rossi, Claudio
author_role author
author2 Rossi, Claudio
author2_role author
dc.contributor.none.fl_str_mv European Commission
Román Ruiz, Marta [0000-0002-1067-8865]
Rossi, Claudio [0000-0002-8740-2453]
Consejo Superior de Investigaciones Científicas [https://ror.org/02gfc7t72]
dc.subject.none.fl_str_mv layer analysis
model optimisation
bioacoustics
spectrogram classification
topic layer analysis
model optimisation
bioacoustics
spectrogram classification
description Deep neural networks are widely used for image classification in different fields, although selecting an appropriate architecture often remains a trial-and-error process. The purpose of this work is to investigate a convolutional neural network architecture used to detect whale pulses in spectrograms in order to better understand the causes of its underperformance. By examining the behaviour of its internal layers, we show that the early convolutional blocks capture the most informative acoustic features, while deeper layers provide limited additional benefit and, under the considered training conditions, may even degrade classification accuracy. Based on these observations, we derive a simplified architecture consisting of only the first two convolutional layers followed by a lightweight classifier. This network achieves near-optimal performance, improving accuracy from 87% to 98%, and exhibits substantially lower variability between repetitions compared to the original model.
publishDate 2026
dc.date.none.fl_str_mv 2026
2026
2026
dc.type.none.fl_str_mv info:eu-repo/semantics/article
http://purl.org/coar/resource_type/c_6501
Publisher's version
info:eu-repo/semantics/publishedVersion
format article
status_str publishedVersion
dc.identifier.none.fl_str_mv http://hdl.handle.net/10261/429805
url http://hdl.handle.net/10261/429805
dc.language.none.fl_str_mv Inglés
language_invalid_str_mv Inglés
dc.relation.none.fl_str_mv #PLACEHOLDER_PARENT_METADATA_VALUE#
info:eu-repo/grantAgreement/EC/HE/101091959
https://doi.org/10.3390/app16052345
Sí
dc.rights.none.fl_str_mv info:eu-repo/semantics/openAccess
eu_rights_str_mv openAccess
dc.publisher.none.fl_str_mv Multidisciplinary Digital Publishing Institute
publisher.none.fl_str_mv Multidisciplinary Digital Publishing Institute
dc.source.none.fl_str_mv reponame:DIGITAL.CSIC. Repositorio Institucional del CSIC
instname:Consejo Superior de Investigaciones Científicas (CSIC)
instname_str Consejo Superior de Investigaciones Científicas (CSIC)
reponame_str DIGITAL.CSIC. Repositorio Institucional del CSIC
collection DIGITAL.CSIC. Repositorio Institucional del CSIC
repository.name.fl_str_mv
repository.mail.fl_str_mv
_version_ 1869403334423609344
score 15,812455

Registos relacionados