Source localization in an ocean waveguide using supervised machine learning.

Area of Science:

• Ocean acoustics
• Machine learning
• Signal processing

Background:

• Underwater source localization is crucial for marine research and operations.
• Traditional methods like matched-field processing have limitations.
• Data-driven machine learning offers a novel approach to acoustic data analysis.

Purpose of the Study:

• To investigate the application of machine learning algorithms for underwater source localization.
• To compare the performance of different machine learning models against conventional techniques.
• To assess the efficacy of solving range estimation as both classification and regression problems.

Main Methods:

• Utilized acoustic pressure data from a vertical linear array.
• Preprocessed data by constructing a normalized sample covariance matrix.
• Applied three machine learning algorithms: feed-forward neural networks (FNN), support vector machines (SVM), and random forests (RF).
• Trained and tested models for range estimation as both classification and regression tasks.

Main Results:

• FNN, SVM, and RF models demonstrated capability in estimating source ranges.
• Performance was evaluated against conventional matched-field processing using Noise09 experimental data.
• Machine learning approaches showed potential for accurate underwater source localization.

Conclusions:

• Machine learning techniques offer a viable and potentially superior alternative for underwater source localization.
• The data-driven nature of these algorithms allows direct learning from acoustic observations.
• Further research can refine these methods for enhanced underwater acoustic sensing.