Physics-Informed Neural Networks for Ocean Acoustic Field Prediction with Envelope Smoothing

Yongsung Park; Peter Gerstoft; Seunghyun Yoon; Woojae Seong

ICASSP 2025accepted0 citations

Physics-Informed Neural Networks for Ocean Acoustic Field Prediction with Envelope Smoothing

Yongsung Park, Peter Gerstoft, Seunghyun Yoon, Woojae Seong

Abstract

Predicting ocean acoustic fields in shallow water is challenging due to high spatial variability, with depth scales of 100 m and range scales of 1 km. Limited acoustic data further complicates this task. We propose a physics-informed neural network (PINN) with the Helmholtz equation as a physics constraint, enhancing prediction accuracy with scarce data. A preprocessing step using an envelope smoothing technique is introduced. This reduces the spatial field variability, enabling more accurate training of the PINN than purely data-driven approaches. Our method is validated through ocean data, demonstrating substantial improvements in PINN performance for complex ocean acoustic predictions.

BibTeX

@inproceedings{icassp2025_physicsinformedn,
  title = {Physics-Informed Neural Networks for Ocean Acoustic Field Prediction with Envelope Smoothing},
  author = {Yongsung Park and Peter Gerstoft and Seunghyun Yoon and Woojae Seong},
  booktitle = {ICASSP 2025},
  year = {2025}
}