2025_programme: Comparison of Detection Methods for Fin Whale Calls from Distributed Acoustic Sensing

Back


  • Day: June 20, Friday
      Location / Time: A. TERPSIHORI at 09:10 - 09:30
  • Last minutes changes: -
  • Session: 05. Distributed Fiber-Optic Sensing technology for underwater acoustical monitoring
    Organiser(s): Alexander Gavrilov, Evgenii Sidenko, Hefeng Dong
    Chairperson(s): Hefeng Dong, Evgenii Sidenko
  • Lecture: Comparison of Detection Methods for Fin Whale Calls from Distributed Acoustic Sensing
    Paper ID: 2330
    Author(s): Khanh Truong, Jo Eidsvik, Robin Andre Rørstadbotnen, Jan Petter Morten
    Presenter: Khanh Truong
    Abstract: Distributed Acoustic Sensing (DAS) technology offers a novel approach for detecting marine mammals, such as fin whales, by utilizing long-range fiber optic cables as acoustic sensors. This study compares the performance of various machine learning methods for detecting fin whale calls in DAS data. The dataset consists of thousands of calls captured over twelve hours of DAS data recorded in August 2022 on a 130 km section of a submarine cable near Svalbard, Norway, with a temporal sampling rate of 625 Hz, spatial sampling period of 4 meters, and a gauge length of 8 meters. After applying FK filtering with a frequency range of 15-25 Hz and velocity range of 1,500-30,000 m/s, we explored three detection approaches: traditional computer vision (Hough transform line and hyperbola detection, template matching), unsupervised learning (DBSCAN), and deep learning-based object detection (YOLO). Performance metrics include F1-score, average precision (AP), and computational efficiency. The performance is evaluated for potential deployment on edge computing devices, specifically the DAS interrogator, for real-time processing. Our findings provide insights into the relative strengths of each detection method for marine mammal monitoring in DAS data, aiming to enhance automated, large-scale tracking of fin whales with an efficient and adaptable approach.
  • Corresponding author: Mr Khanh Truong
    Affiliation: Norwegian University of Science and Technology
    Country: Norway

Back