Alleviating negative transfer using unsupervised adversarial partial domain adaptation for underwater source range...

Area of Science:

• Ocean acoustics
• Signal processing
• Machine learning

Background:

• Underwater source localization is crucial for various applications.
• Existing unsupervised domain adaptation (UDA) methods struggle with practical data where measured and simulated ranges differ.
• This performance degradation is known as negative transfer.

Purpose of the Study:

• To propose an improved underwater source range estimation method using unsupervised partial domain adaptation (PDA).
• To address the negative transfer issue in UDA when measured and simulated data range spaces are not identical.
• To enhance the accuracy of underwater source localization in challenging, real-world conditions.

Main Methods:

• Developed a partial domain adaptation (PDA) framework for underwater source range estimation.
• Incorporated a simulated data weighting mechanism into the UDA framework based on PDA.
• Utilized a deep neural network to process measured data and guide the weighting of simulated data for domain discrimination.

Main Results:

• The proposed PDA-based method significantly outperforms UDA when measured and simulated data range spaces differ.
• The method demonstrates superior performance compared to traditional matched-field processing in various scenarios.
• Effectiveness was validated across different adaptation data sampling strategies and environmental parameter mismatches.

Conclusions:

• The PDA-based weighting mechanism effectively mitigates negative transfer in unsupervised domain adaptation for range estimation.
• This method offers a robust solution for underwater source localization, especially when dealing with domain shift.
• The approach shows significant promise for improving the accuracy and reliability of acoustic source localization in ocean environments.