2023_programme: Multitaper Spectral Processing for Acoustic Detection of UUVs

Back


  • Session: 15. Towards Automatic Target Recognition. Detection, Classification and Modelling
    Organiser(s): Johannes Groen, Yan Pailhas, Roy Edgar Hansen, Jessica Topple and Narada Warakagoda
  • Lecture: Multitaper Spectral Processing for Acoustic Detection of UUVs [invited]
    Paper ID: 2070
    Author(s): Prisbrey Milo, Teti Michael, Orescanin Marko, Migliori Benjamin
    Presenter: Migliori Ben
    Abstract: The detection of unmanned undersea vehicles (UUVs) represents a prominent area of application for underwater acoustic methods. Successful vehicle tagging requires an understanding of the background noise, the vehicle parameters and configuration, and the acoustic propagation environment. Here, we present a novel non-parametric method for preprocessing hydrophone data for UUV detection based on multitaper analysis. Multitaper methods use multiple overlapping tapers, each an orthogonal function that is localized in both time and frequency, to compute part of a spectral estimate. The full estimate is the weighted sum of the individual taper estimates, where the weights are chosen to optimize the spectral resolution and for variance reduction. The multitaper method has been shown to be particularly effective in cases where the signal is contaminated with noise or other sources of interference, as it can suppress such sources more effectively than traditional methods. In this work, we show how to configure multitaper analysis for UUV detection, and show spectral comparisons with more traditional methods. Finally, we show how to use spectral estimates as inputs for neural networks, and describe suitable architectures that match the physics of the acoustic data. We demonstrate the efficacy of our methods on UUV data acquired in Lake Crescent, Washington. We show the the use of physics-informed features, on the form of multitaper spectral estimates, provide a computationally efficient and high performance data view on otherwise normally acquired hydrophone data. While we demonstrate our method for the desirable case of detection of UUV acoustics, this technique is expected to be applicable to shipping traffic, marine biome acoustics, and natural undersea phenomena.
      Download the full paper
  • Corresponding author: Dr Benjamin Migliori
    Affiliation: Los Alamos National Lab
    Country: United States
    e-mail:

Back