Hybrid data- and model-driven three-dimensional ocean sound speed field super-resolution: Diffusion model meets...

Area of Science:

• Oceanography
• Acoustics
• Signal Processing

Background:

• High-resolution 3D sound speed fields (SSFs) are vital for underwater acoustics.
• Current SSF observation systems yield low-resolution, noisy data, limiting applications.
• Existing super-resolution (SR) methods struggle with subtle variations in sparse, noisy SSF data.

Purpose of the Study:

• To develop an advanced SR method for accurate 3D SSF reconstruction.
• To overcome limitations of purely model-based or data-driven SR techniques.
• To improve characterization of ocean sound propagation and acoustic transmission losses.

Main Methods:

• Proposed Hybrid-DOT, a hybrid SR algorithm integrating data-driven and model-driven approaches.
• Employed a pre-trained diffusion model for data distribution insights.
• Utilized low-rank tensor (LRT) modeling for domain knowledge integration and computational efficiency.

Main Results:

• Hybrid-DOT demonstrated superior performance over state-of-the-art methods.
• The algorithm achieved high accuracy across various SR factors and noise levels.
• Enhanced SSF reconstruction enabled precise characterization of fine-grained acoustic transmission losses.

Conclusions:

• Hybrid-DOT effectively reconstructs high-resolution 3D SSFs from limited data.
• The hybrid approach offers improved accuracy and efficiency for underwater acoustic applications.
• This method advances the capability to analyze complex ocean sound propagation environments.