Seabed and range estimation of impulsive time series using a convolutional neural network.

Area of Science:

• Ocean acoustics
• Deep learning applications
• Geophysical signal processing

Background:

• Traditional methods for seabed characterization and acoustic source localization face limitations.
• The application of deep learning in ocean acoustics is underexplored due to limited labeled field data.
• Convolutional Neural Networks (CNNs) offer potential for analyzing complex acoustic signals.

Purpose of the Study:

• To investigate the use of a CNN for simultaneous seabed type identification and acoustic source range estimation.
• To train a CNN using simulated ocean acoustic data.
• To evaluate the CNN's performance on real-world hydrophone data.

Main Methods:

• A Convolutional Neural Network (CNN) was developed and trained on simulated impulsive sound pressure time series.
• The training dataset encompassed four distinct seabed types and a broad spectrum of source parameters.
• The trained CNN was applied to single hydrophone data from the 2017 Seabed Characterization Experiment.

Main Results:

• The CNN successfully predicted seabed types from measured acoustic data.
• Accurate source range estimations within 0.5 km were achieved for ranges exceeding 5 km.
• The study demonstrates the feasibility of using deep learning for ocean acoustic inverse problems.

Conclusions:

• Deep learning, specifically CNNs, can effectively analyze ocean acoustic data for seabed characterization and source localization.
• Simulated data is a viable approach for training models when labeled field data is scarce.
• This research highlights the potential of CNNs to advance ocean acoustic sensing technologies.