Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

From Spaceborne to Airborne: SAR Image Synthesis Using Foundation Models for Multi-Scale Adaptation

Solene Debuysere, Nicolas Trouve, Nathan Letheule, Olivier Leveque, Elise Colin

TL;DR

This work tackles the data bottleneck for airborne SAR by assembling a large ONERA-based SAR dataset and SAR–optical pairs to train a stable diffusion foundation model (SDXL) for translating satellite SAR to airborne SAR and for multi-resolution upscaling. It introduces a latent-space diffusion pipeline with two ControlNet modules and LoRA fine-tuning to enhance TerraSAR-X imagery to a 40 cm ground resolution, while also increasing realism of EMPRISE-simulated images. Key contributions include the dataset construction (≈110k SAR images and ≈100k SAR–optical pairs), the latent-upscaling strategy across 160–40 cm resolutions, and the integration of spatial conditioning to preserve structural fidelity. The work identifies SAR-specific challenges for ControlNet (originating from optical conditioning) and points to future directions in SAR-tailored conditioning and multimodal, multi-resolution conditioning for remote sensing data augmentation.

Abstract

The availability of Synthetic Aperture Radar (SAR) satellite imagery has increased considerably in recent years, with datasets commercially available. However, the acquisition of high-resolution SAR images in airborne configurations, remains costly and limited. Thus, the lack of open source, well-labeled, or easily exploitable SAR text-image datasets is a barrier to the use of existing foundation models in remote sensing applications. In this context, synthetic image generation is a promising solution to augment this scarce data, enabling a broader range of applications. Leveraging over 15 years of ONERA's extensive archival airborn data from acquisition campaigns, we created a comprehensive training dataset of 110 thousands SAR images to exploit a 3.5 billion parameters pre-trained latent diffusion model \cite{Baqu2019SethiR}. In this work, we present a novel approach utilizing spatial conditioning techniques within a foundation model to transform satellite SAR imagery into airborne SAR representations. Additionally, we demonstrate that our pipeline is effective for bridging the realism of simulated images generated by ONERA's physics-based simulator EMPRISE \cite{empriseem_ai_images}. Our method explores a key application of AI in advancing SAR imaging technology. To the best of our knowledge, we are the first to introduce this approach in the literature.

From Spaceborne to Airborne: SAR Image Synthesis Using Foundation Models for Multi-Scale Adaptation

TL;DR

This work tackles the data bottleneck for airborne SAR by assembling a large ONERA-based SAR dataset and SAR–optical pairs to train a stable diffusion foundation model (SDXL) for translating satellite SAR to airborne SAR and for multi-resolution upscaling. It introduces a latent-space diffusion pipeline with two ControlNet modules and LoRA fine-tuning to enhance TerraSAR-X imagery to a 40 cm ground resolution, while also increasing realism of EMPRISE-simulated images. Key contributions include the dataset construction (≈110k SAR images and ≈100k SAR–optical pairs), the latent-upscaling strategy across 160–40 cm resolutions, and the integration of spatial conditioning to preserve structural fidelity. The work identifies SAR-specific challenges for ControlNet (originating from optical conditioning) and points to future directions in SAR-tailored conditioning and multimodal, multi-resolution conditioning for remote sensing data augmentation.

Abstract

The availability of Synthetic Aperture Radar (SAR) satellite imagery has increased considerably in recent years, with datasets commercially available. However, the acquisition of high-resolution SAR images in airborne configurations, remains costly and limited. Thus, the lack of open source, well-labeled, or easily exploitable SAR text-image datasets is a barrier to the use of existing foundation models in remote sensing applications. In this context, synthetic image generation is a promising solution to augment this scarce data, enabling a broader range of applications. Leveraging over 15 years of ONERA's extensive archival airborn data from acquisition campaigns, we created a comprehensive training dataset of 110 thousands SAR images to exploit a 3.5 billion parameters pre-trained latent diffusion model \cite{Baqu2019SethiR}. In this work, we present a novel approach utilizing spatial conditioning techniques within a foundation model to transform satellite SAR imagery into airborne SAR representations. Additionally, we demonstrate that our pipeline is effective for bridging the realism of simulated images generated by ONERA's physics-based simulator EMPRISE \cite{empriseem_ai_images}. Our method explores a key application of AI in advancing SAR imaging technology. To the best of our knowledge, we are the first to introduce this approach in the literature.
Paper Structure (13 sections, 6 equations, 3 figures)

This paper contains 13 sections, 6 equations, 3 figures.

Table of Contents

  1. Introduction
  2. Methodology
  3. Objectives
  4. Architecture: Stable Diffusion Foundation Model
  5. Inputs
  6. Training Process
  7. Strategy: Transfer learning in Radar domain
  8. Spatial conditioning: ControlNet Module
  9. Applications
  10. Objectives
  11. Pipeline architecture
  12. Discussion
  13. Conclusion

Figures (3)

  • Figure 1: Creative upscaling pipeline
  • Figure 2: Enhanced TerraSAR-X satellite images to aerial resolution Top row: TerraSAR-X images (512x512, 80cm). Bottom row: AI-enhanced TerraSAR-X images (2048x2048, 40cm).
  • Figure 3: Enhanced simulated images from ONERA's Radar simulator with creative content Top row: simulated images (512x512, 40cm). Bottom row: AI-enhanced simulated images (2048x2048, 40cm).