Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Physics-Informed Diffusion Models for SAR Ship Wake Generation from Text Prompts

Kamirul Kamirul, Odysseas Pappas, Alin Achim

TL;DR

This work tackles the paucity of annotated SAR wake data by training a text-conditioned latent diffusion model on wake images produced by a physics-based simulator. By pairing simulator-generated images with parameter-derived prompts, the model learns to generate Kelvin wakes with controllable attributes and substantially faster inference than the physics-based simulator. Visual and spectral analyses show the diffusion model captures major wake structures and their dependence on wind, ship speed, and heading, though limited training data introduces some mid-frequency noise. The approach enables fast, end-to-end wake synthesis suitable for downstream maritime SAR tasks and scalable data generation for detection and analysis.

Abstract

Detecting ship presence via wake signatures in SAR imagery is attracting considerable research interest, but limited annotated data availability poses significant challenges for supervised learning. Physics-based simulations are commonly used to address this data scarcity, although they are slow and constrain end-to-end learning. In this work, we explore a new direction for more efficient and end-to-end SAR ship wake simulation using a diffusion model trained on data generated by a physics-based simulator. The training dataset is built by pairing images produced by the simulator with text prompts derived from simulation parameters. Experimental result show that the model generates realistic Kelvin wake patterns and achieves significantly faster inference than the physics-based simulator. These results highlight the potential of diffusion models for fast and controllable wake image generation, opening new possibilities for end-to-end downstream tasks in maritime SAR analysis.

Physics-Informed Diffusion Models for SAR Ship Wake Generation from Text Prompts

TL;DR

This work tackles the paucity of annotated SAR wake data by training a text-conditioned latent diffusion model on wake images produced by a physics-based simulator. By pairing simulator-generated images with parameter-derived prompts, the model learns to generate Kelvin wakes with controllable attributes and substantially faster inference than the physics-based simulator. Visual and spectral analyses show the diffusion model captures major wake structures and their dependence on wind, ship speed, and heading, though limited training data introduces some mid-frequency noise. The approach enables fast, end-to-end wake synthesis suitable for downstream maritime SAR tasks and scalable data generation for detection and analysis.

Abstract

Detecting ship presence via wake signatures in SAR imagery is attracting considerable research interest, but limited annotated data availability poses significant challenges for supervised learning. Physics-based simulations are commonly used to address this data scarcity, although they are slow and constrain end-to-end learning. In this work, we explore a new direction for more efficient and end-to-end SAR ship wake simulation using a diffusion model trained on data generated by a physics-based simulator. The training dataset is built by pairing images produced by the simulator with text prompts derived from simulation parameters. Experimental result show that the model generates realistic Kelvin wake patterns and achieves significantly faster inference than the physics-based simulator. These results highlight the potential of diffusion models for fast and controllable wake image generation, opening new possibilities for end-to-end downstream tasks in maritime SAR analysis.
Paper Structure (13 sections, 4 equations, 4 figures, 3 tables)

This paper contains 13 sections, 4 equations, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Theoretical Preliminaries
  3. Physics-based Wake Simulation
  4. Diffusion Models
  5. Methodology
  6. Training images
  7. C. Parameter-to-Text Conversion
  8. Model Training
  9. Results and Discussion
  10. Visual Quality
  11. Quantitative Evaluation
  12. Inference Speed
  13. Conclusions

Figures (4)

  • Figure 1: Training of a text-conditioned diffusion model for SAR ship wake generation. Synthetic images from a physics-based simulator enable a physics-informed learning process.
  • Figure 2: Effect of sea state and ship parameters on wake patterns generated by the LDM. Top row: Variation of wind speed ($v_w = 3$ m/s, $5$ m/s, $7$ m/s), illustrating how increasing wind strength influences the surface wave characteristics. Middle row: Variation of ship speed ($v_s = 8$ m/s, $10$ m/s, $12$ m/s), showing how changes in vessel velocity affect the wake structure. Bottom row: Variation of ship orientation ($\theta_s = 0^\circ$, $45^\circ$, $90^\circ$), demonstrating the impact of heading angle on wake symmetry and direction.
  • Figure 3: Comparison of SAR wake simulations for two prompt conditions, with PBM results on the left and LDM results on the right in each row.
  • Figure 4: Spectral comparison between images generated by physics-based simulator (PBS) and latent diffusion model (LDM).