2025_programme: Impact of ambient noise and water sound speed on deep learning seabed classification from shipping noise

• Day: June 17, Tuesday
    Location / Time: B. ERATO at 11:20-11:40
• Last minutes changes: -
• Session: 19. Uncertainty quantification and machine learning in signal processing
  Organiser(s): Angeliki Xenaki, Zoi-Heleni Michalopoulou
  Chairperson(s): Angeliki Xenaki, Zoi-Heleni Michalopoulou
• Lecture: Impact of ambient noise and water sound speed on deep learning seabed classification from shipping noise [Invited]
  Paper ID: 2239
  Author(s): Tracianne Neilsen, Alexandra McDaniel, Ginger Lau, William Hodgkiss, David Knobles
  Presenter: Traci Neilsen
  Abstract: For ocean application of machine learning, the temporal variations in the ocean environment introduce uncertainty. Specifically, the impacts of the temporal variations in ambient noise and the water sound speed profile have been observed in recent work using ship noise in a ResNet-18 model trained for seabed classification. The input data samples are 20 minute spectrograms of shipping noise in the 360-1100 Hz band. Synthetic data are used for training and validation. The trained models are then applied to ship noise data samples from the 2017 and 2022 Seabed Characterization Experiments in the New England mud patch. For the measured 2017 data samples, the seabed classifier output has a correlation with the ambient noise statistics close in time to when the measured ship noise was recorded. For both datasets, the collection of water sound speed profiles included in the synthetic training dataset affects the seabed classification results on the measured data samples. A review of these observations is provided along with a discussion of how such temporal variability may be addressed in future applications.
• Corresponding author: Prof Traci Neilsen
  Affiliation: Brigham Young University
  Country: United States