Marineterpreter:Automatic Identification of Marine Mammals using Acoustic Vocal Calls

July 13, 2021

While there are several ways of understanding the human impact on marine mammals andtheir habitats, recent studies in marine ecology reveal that whales directly contribute to thereduction of CO2level. However, not many works focus on understanding and interpretingtheir communication. What is generally known, is that cetaceans are capable to shift theirvocal call bands, surpassing the generated noise pollution caused by the human presence.However, interpretation of marine mammals’ vocal calls still remains a challenge, wherecurrent literature remains mostly oriented on recognizing such species through acoustics.This provides an opportunity for more robust algorithms to crack the whale, dolphin andseal language.In effort to apply more technologies to such challenge, this dissertation describes thedesign of a software apparatus, capable of detecting and classifying the marine mammalspecies based on their acoustic vocal calls. Using the existing machine and deep learningtechniques for audio and image analysis and existing datasets of marine mammals vocalcalls, contribution of the thesis is in created models for for predicting marine mammalfamily classes, where extensive comparisons are provided.While tailored for IoT devices, proposed software supports the acoustic surveys fromthe sea vessels, where marine ecologists participate in recording and labeling the species,contributing to the model evolution. Two experimental studies are performed describinghow marine mammals are identified using machine and deep learning.

Authors in the community:

• 

  J

  João Filipe Correia Gomes de Gouveia

  ist190861

Supervisors of this institution:

• 

  D

  Duarte Nuno Jardim Nunes

  ist132720

electrical-engineering-electronic-engineering-information-engineering - Electrical engineering, electronic engineering, information engineering

eng - English

Instituto Superior Técnico