Deep-learning geoacoustic inversion using multi-range vertical array data in shallow water.

Area of Science:

• Ocean Acoustics
• Geophysical Signal Processing
• Machine Learning Applications in Marine Science

Background:

• Accurate estimation of shallow water geoacoustic parameters is crucial for underwater acoustic applications.
• Conventional methods like matched-field inversion (MFI) can be sensitive to environmental uncertainties and source positioning.
• Developing robust and efficient methods for geoacoustic parameter inversion remains an active research area.

Purpose of the Study:

• To propose and validate a novel multi-range vertical array data processing (MRP) method based on convolutional neural networks (CNNs) for estimating shallow water geoacoustic parameters.
• To assess the performance of the MRP method against conventional matched-field inversion (MFI) and single-range processing techniques.
• To demonstrate the effectiveness of the MRP method in improving underwater localization accuracy in shallow water environments.

Main Methods:

• Development of a CNN-based multi-range vertical array data processing (MRP) method.
• Utilizing normalized sample covariance matrices from broadband multi-range data as network input.
• Employing multi-task learning to simultaneously estimate geoacoustic parameters (sound speed, density, attenuation) with varying scales.
• Training and validation using simulation data with varied source depths to mitigate source position uncertainty.

Main Results:

• The CNN with MRP effectively alleviates geoacoustic parameter coupling and demonstrates robustness to source depth variations in shallow water.
• Geoacoustic parameter inversion using MRP leads to improved localization performance, even when assuming a range-independent model for a range-dependent environment.
• Validation with real data from the East China Sea confirms that MRP outperforms MFI and single-range CNN processing in achieving better localization accuracy.

Conclusions:

• The proposed CNN-based MRP method offers a significant advancement in estimating shallow water geoacoustic parameters.
• MRP provides a more robust and accurate approach compared to traditional MFI and single-range processing methods.
• The enhanced geoacoustic parameter estimation using MRP directly translates to improved underwater localization capabilities.