Detecção automática dos sinais acústicos de delfinídeos obtidos por monitoramento acústico passivo

Authors

  • Celso Jose Munaro Departamento de Engenharia Elétrica, UniversidadeFederal do Espírito Santo/UFES, Vitória 29075-900, Brasil
  • Lilian Sander Hoffmann Laboratório de Nectologia, Departamento de Oceanografiae Ecologia, Universidade Federal do Espírito Santo/UFES, Vitória 29075-900, Brasil
  • Agnaldo Martins Laboratório de Nectologia, Departamento de Oceanografiae Ecologia, Universidade Federal do Espírito Santo/UFES, Vitória 29075-900, Brasil
  • Beatriz Araujo Farias Laboratório de Nectologia, Departamento de Oceanografiae Ecologia, Universidade Federal do Espírito Santo/UFES, Vitória 29075-900, Brasil
  • Helder Antonio Medina Avila Laboratório de Nectologia, Departamento de Oceanografiae Ecologia, Universidade Federal do Espírito Santo/UFES, Vitória 29075-900, Brasil

DOI:

https://doi.org/10.20906/CBA2024/4594

Keywords:

Bioacoustics, Dolphins, Whistles, Echolocation, Spectrogram, Classification, Patterns Recognition

Abstract

Cetaceans,comprisingwhales,dolphins,andporpoises,areintegralcomponentsofmarine ecosystems,playingpivotalrolesinmaintainingecologicalbalance.Acousticmonitoringemergesasa crucialtoolinunderstandingcetaceanbehavior,distribution,andpopulationdynamics.Equipmentthat capturesandrecordsthesesignalsoverdaysgeneratesalargeamountofdatathatrequiresagreateffort toextractonlytheintervalsinwhichtherearesignalstobeanalyzed.Thisworkaimstopresentanew methodologyforautomaticallydetectingthepresenceofdelphinidvocalizationsinaudiosignals.Atest statisticisapplieddirectlytotheaudiosignalsgeneratingtheintervalsinwhichtherearewhistlesor clicksandburstpulses.Inthecaseofwhistles,thespectrogramiscomputedandthecontourshowingthe frequencyovertimeisobtained,allowingitsuseinclassifierstoindicatedolphinspeciesandactivities. Themethodologywasappliedtotwocasestudiesandfordetectingwhistles,anaccuracygreaterthan 89%andprecisiongreaterthan97%wereachieved.Whendetectingclicksorburstpulses,the performancewas86%foraccuracyand81%forprecision.Thealgorithmsrequirelowcomputational effortandcansignificantlyreducetheeffortrequiredtoextractcapturedsignals,thusallowingan increase in the number of collection points and their periods.

Downloads

Published

2024-10-18

Issue

Section

Articles