2025_programme: Realtime unsupervised object analysis on glider for whale monitoring applications

• Day: June 17, Tuesday
    Location / Time: B. ERATO at 08:50 - 09:10
• Last minutes changes: -
• Session: 25. Unmanned vehicles for underwater acoustic surveillance and monitoring
  Organiser(s): Alain Maguer
  Chairperson(s): Alain Maguer
• Lecture: Realtime unsupervised object analysis on glider for whale monitoring applications
  Paper ID: 2203
  Author(s): Marc Michau, Rémi Emmetière
  Presenter: Marc Michau
  Abstract: The growing need for 24/7 real-time monitoring of marine life, driven by applications such as anti-collision or bycatch reduction systems, positions unmanned vehicles as key solutions to enhance the effectiveness of monitoring efforts. New generation of gliders and acoustic payloads allows for increased real-time embedded processing such as detection, classification or localization of underwater acoustic signals. Most of the time, detection and classification tasks are based on time-frequency representation. On specific annotated dataset, AI based supervised algorithms have shown interesting results. However, robustness of such approaches is highly related to the quality of the database and may depends on the acquisition system and the environmental noises. An alternative approach is described here, based on unsupervised time-frequency object extraction and description with physical parameters. Such embedded information can be transmitted via satellite communication, paving the way for combined embedded/decentralized processing or expert validation. Results are presented for two use cases of whale detection applications : (i) when the targeted signal is highly stereotypical, in which case a comparison between CNN model predictions is performed, and (ii) when the acoustic environment is unknown prior to the mission. For both cases, a scenario combining embedded processing, data transmission, and decentralized processing is outlined, along with the dynamic adaptation of the navigation plan.
• Corresponding author: Dr Marc Michau
  Affiliation: ALSEAMAR
  Country: France