Underwater Source Localization and Classification using Hybrid Machine Learning: an integrative approach

Abstract

We propose an interdisciplinary 3 year research effort to develop an underwater source localization and classification framework th"at couples acoustic propagation modelling with a new Hybrid Machine Learning (HML) algorithm. The proposed HML algorithm will integrate model-based and data-driven (provided by blind deconvolution of shipping sources of opportunity) learning at various spatial and temporal scales governing the acoustic wave propagation in the underwater environment of interest. Our HML approach seamlessly integrates training data directly with a model for the acoustic wave propagation (using best-fit environmental parameters) by trusting the former in parts of the domain that are data-rich, and the latter in parts that are only sparsely populated by sources of opportunity. This coupled-HML platform will identify ~through-the-sensor~ the in-situ physical parameters governing the wave propagation and thus improve model prediction with a significant reduction in computationalcost compared to classical high-resolution inversion and data-assimilation procedures. Ultimately we expect this approach will enhance the source localization and classification performance by inherently accounting for the unresolved scales and ocean variability affecting the data but not reproduced by the baseline mo"del.

Document Details

Document Type
DoD Grant Award
Publication Date
Aug 20, 2019
Source ID
N000141912639

Entities

People

  • Karim G. Sabra

Organizations

  • Georgia Tech Research Corporation
  • Office of Naval Research
  • United States Navy

Tags

Readers

  • Acoustical Oceanography.
  • Distributed Systems and Data Platform Development
  • Wave Propagation and Nonlinear Chaotic Dynamics.

Technology Areas

  • AI & ML
  • AI & ML - Neural Networks