Seabed classification using physics-based modeling and machine learning.

Area of Science:

• Oceanography
• Geophysics
• Acoustics

Background:

• Seabed sediment classification is crucial for marine geological surveys and resource exploration.
• Understanding geoacoustic properties of layered seabeds is key to accurate remote sensing.
• Existing methods for sediment classification face challenges with noise and model complexity.

Purpose of the Study:

• To develop and evaluate machine learning (ML) techniques for classifying oceanic sediments based on geoacoustic properties.
• To compare the performance of ML methods against traditional matched-field approaches in different acoustic scenarios.
• To assess the robustness of ML classifiers to environmental noise and model uncertainties.

Main Methods:

• Model-based acoustic field modeling using normal modes for low-frequency scenarios.
• Application of various machine learning techniques, including one-dimensional convolutional neural networks (1D CNNs), for sediment classification.
• Investigation of a high-frequency scattering model for a rough, two-layer seafloor.

Main Results:

• Machine learning methods consistently outperformed the matched-field approach in classifying seafloor sediments across different noise levels.
• 1D CNNs demonstrated high accuracy in sediment classification for the high-frequency, rough seafloor scenario.
• Both simple and complex ML formulations proved effective for robust sediment characterization.

Conclusions:

• Machine learning offers a powerful and effective framework for oceanic sediment classification.
• ML techniques provide superior performance and robustness compared to traditional methods, especially in noisy environments.
• The study highlights the potential of ML for advancing marine geological and geophysical investigations.