Bayesian Optimization-Tuned machine learning for underwater acoustic target localization.

Area of Science:

• Oceanography
• Acoustics
• Machine Learning

Background:

• Underwater acoustic target localization (UATL) faces challenges with existing methods like matched field processing (MFP) due to environmental noise and processing inefficiency.
• Accurate and real-time UATL is crucial for various applications but remains difficult to achieve.

Purpose of the Study:

• To develop and evaluate a novel Bayesian optimization-tuned machine learning approach for enhanced UATL.
• To compare the performance of this new approach against traditional MFP and other machine learning methods with different hyperparameter tuning strategies.

Main Methods:

• Generated simulated training data using the KRAKEN propagation code for a specific shallow sea environment.
• Employed two machine learning models: k-nearest neighbor and support vector regression.
• Utilized Bayesian optimization for hyperparameter tuning and compared it with alternative methods.

Main Results:

• Machine learning approaches demonstrated superior localization accuracy compared to MFP.
• Successfully identified an underwater target at a 5.6 km range (error < 0.1 km) and 79 m depth (error < 0.5 m).
• Bayesian optimization proved more efficient for hyperparameter tuning than other methods.

Conclusions:

• The proposed Bayesian optimization-tuned machine learning method offers a more accurate and efficient solution for UATL.
• This approach effectively addresses limitations of traditional methods in noisy and large-scale data scenarios.