SSFDiffusion: A Degradation-Aware Diffusion Model for Ocean Sound Speed Field Super-Resolution
Open Access
- 23 March 2026
- journal article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
- Vol. 19 (19391404) , 12222-12233
Abstract
Ocean sound speed field (SSF) is a fundamental geophysical parameter in underwater acoustics and ocean observation, as it plays a critical role in applications such as sonar imaging, underwater communication, and navigation. However, its low spatial resolution limits the performance of these applications in practical scenarios. Super-resolution (SR) techniques provide an effective means to reconstruct high-resolution (HR) SSFs from low-resolution (LR) observations. Nevertheless, most existing SR methods assume a simple, static degradation process, which overlooks the complex and unknown degradation and noise patterns inherent in real marine environments. This oversimplification often results in limited generalization and suboptimal performance in real-world applications. To overcome this limitation, we propose SSFDiffusion, a degradation-aware diffusion model that explicitly models the coupled effects of global degradation and local noise through three core components: (1) a dynamic denoising network adaptive to varying noise levels, (2) a directional multi-scale Transformer for modeling global degradation and local noise, and (3) an alternating optimization strategy that balances degradation correction and noise suppression to prevent artifact accumulation. Extensive experiments on the South China Sea and GDCSM Argo datasets demonstrate that SSFDiffusion consistently outperforms seven state-of-the-art super-resolution methods across five key quantitative metrics. Notably, our method achieves up to 48.52 dB PSNR and an order-of-magnitude RMSE reduction. These experimental results indicate that SSFDiffusion enhances both the spatial fidelity and practical applicability of reconstructed SSF data, benefiting oceanographic research and underwater acoustic applications.Keywords
Funding Information
- National Natural Science Foundation of China (12402349)
- Natural Science Foundation of Hunan Province (2024JJ6468)
- Youth Foundation of the National University of Defense Technology (ZK2023-11)
- National Key Research and Development Program of China (2021YFB0300101)
This publication has 43 references indexed in Scilit:
- MH-DCNet: An improved flow field prediction framework coupling neural network with physics solverComputers & Fluids, 2024
- Introduction to Sound Propagation Under Water2022
- Influence of internal wave induced sound speed variability on acoustic propagation in shallow waters of North West Bay of BengalApplied Acoustics, 2022
- SRDiff: Single image super-resolution with diffusion probabilistic modelsNeurocomputing, 2022
- A Gaussian process regression-based sea surface temperature interpolation algorithmJournal of Oceanology and Limnology, 2021
- Investigation on total adjustment of the transducer and seafloor transponder for GNSS/Acoustic precise underwater point positioningOcean Engineering, 2021
- ESRGAN: Enhanced Super-Resolution Generative Adversarial Networks2019
- The Interpolation of Sparse Geophysical DataSurveys in Geophysics, 2018
- Spatial Variation of Diapycnal Diffusivity Estimated From Seismic Imaging of Internal Wave Field, Gulf of MexicoJournal of Geophysical Research: Oceans, 2017
- The use and abuse of image analysis in geophysical potential field interpretationSurveys in Geophysics, 1991