Détection et classification dans un contexte acoustique passive : application à la détection des signaux basse-fréquences des baleines bleues

The analysis of the large volumes of data resulting from continuous and long-term monitoring efforts of blue whales (BWs) benefits from the automated detection of target signals. This thesis investigates the challenging problem of the detection and classification of stereotyped signals in a low-freq...

Full description

Bibliographic Details
Main Author: Bouffaut, Léa
Other Authors: Institut de Recherche de l'Ecole Navale (IRENAV), Université de Bordeaux (UB)-Institut Polytechnique de Bordeaux-Centre National de la Recherche Scientifique (CNRS)-Institut National de Recherche pour l’Agriculture, l’Alimentation et l’Environnement (INRAE)-Arts et Métiers Sciences et Technologies, HESAM Université - Communauté d'universités et d'établissements Hautes écoles Sorbonne Arts et métiers université (HESAM)-HESAM Université - Communauté d'universités et d'établissements Hautes écoles Sorbonne Arts et métiers université (HESAM), Université de Bretagne occidentale - Brest, Abdel-Ouahab Boudraa
Format: Doctoral or Postdoctoral Thesis
Language:English
Published: HAL CCSD 2019
Subjects:
Online Access:https://theses.hal.science/tel-02475688
https://theses.hal.science/tel-02475688/document
https://theses.hal.science/tel-02475688/file/These-2019-SML-Acoustique_sous_marine_et_traitement_du_signal-BOUFFAUT_Lea.pdf
Description
Summary:The analysis of the large volumes of data resulting from continuous and long-term monitoring efforts of blue whales (BWs) benefits from the automated detection of target signals. This thesis investigates the challenging problem of the detection and classification of stereotyped signals in a low-frequency passive acoustic context where (1) signals traveling long distances are deteriorated by the propagation channel, (2) overlapping noises interfere and, (3) SNRs vary continuously. Developed methods are applied to recordings from ocean bottom seismometers deployed in the western Indian Ocean.First, the stochastic matched filter (SMF) is adapted to the passive context by overcoming noise estimation and estimating the SNR automatically. This filter is successfully applied to the detection of Antarctic blue whales calls and is compared to the MF and the Z-detector on an annotated ground-truth dataset exhibiting various SNRs and noises. The passive SMF showed better performances, increasing the detection range up to 100 km in the presence of ship noise.The problematic of the detection of concurrently calling species is addressed based on a pattern recognition development for the automatic transcription of BW songs where, tonal signals are extracted, characterized, and classified. The hence identified signals are then reconstructed as separate waveforms reconstructing of the underlying songs. The success of the reconstruction relies on the quality of the tonal detector: the ridge detector was chosen for its efficiency. Training and unsupervised application revealed promising results of the proposed transcription method and its utility for multi-species analysis. L’analyse des grands volumes de données générés par la surveillance par acoustique passive long-terme et continue des baleines bleues (BW) est améliorée par la détection automatisée des signaux d’intérêt. Le travail présenté dans cette thèse s’attaque au problème de la détection et classification de signaux stéréotypés dans un contexte passif basse fréquence ...