Deep learning-based high-frequency source depth estimation using a single sensor.

Area of Science:

• Acoustic propagation modeling
• Underwater acoustics
• Signal processing
• Machine learning

Background:

• Underwater acoustic propagation models are highly sensitive to environmental variability, especially at higher frequencies.
• This sensitivity makes accurate acoustic propagation predictions and high-frequency source localization challenging using traditional model-based methods.
• Existing methods struggle with the mismatch between model assumptions and real-world underwater acoustic scenarios.

Purpose of the Study:

• To develop and evaluate a deep learning approach for accurate underwater source depth estimation.
• To address the limitations of model-based methods in high-frequency acoustic localization.
• To assess the performance of the proposed method across varying receiver depths and environmental conditions.

Main Methods:

• Utilized several 18-layer residual neural networks (ResNets) trained on normalized log-scaled spectrograms.
• Data was acquired using a single hydrophone and processed from linear frequency modulation chirp signals (11-31 kHz) during the Shallow-water Acoustic Variability Experiment 2015.
• The ResNet algorithm was applied to data from two vertical line arrays (VLAs), with performance evaluated against receiver depth and compared with frequency-difference matched field processing.

Main Results:

• The ResNet-based method successfully determined complex features of high-frequency acoustic signals.
• Accurate depth estimations were achieved irrespective of the receiver's depth within the vertical line array.
• The proposed deep learning approach demonstrated robustness against environmental and positional variability, outperforming traditional methods in complex scenarios.

Conclusions:

• Deep learning, specifically ResNet, offers a powerful solution for accurate high-frequency underwater source depth estimation.
• The trained algorithm effectively handles the complexities of real-world underwater acoustic environments.
• This approach significantly advances the capability for reliable acoustic localization in challenging shallow-water conditions.