Abstract
We proposed UNSB (Unpaired Image-to-Image Translation via Neural Schrödinger Bridge)-based image translation and Pix2pix-based colorization to simulate cloud occlusion areas in optical images. When converting the image, topographic information about the simulated area was referenced from the high-resolution SAR image. In addition, we generated a simulated image based on CycleGAN (Cycle Generative Adversarial Network), which has been widely used in existing SAR image simulation studies, and compared and analyzed the results of the proposed method.The KOMPSAT-3A Panchromatic image was Pan-sharpened as the target image to be simulated, and the COSMO-SkyMed 4 image with 1m resolution was used as the reference SAR image. As a result of the experiment, the RMSE of the UNSB-based simulated image in the experimental area was calculated as 9.475, the PSNR 28.671, and the SSIM 0.870, and it was confirmed that it was improved by 4.67 % compared to the CycleGAN-based simulated image that was compared.
| Translated title of the contribution | Optical Image Simulation of SAR Image Based on UNSB |
|---|---|
| Original language | Korean |
| Pages (from-to) | 33-39 |
| Number of pages | 7 |
| Journal | Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography |
| Volume | 42 |
| Issue number | 1 |
| DOIs | |
| State | Published - 2024 |
Keywords
- CycleGAN
- High-Resolution SAR Image
- Image Restoration
- UNSB
Fingerprint
Dive into the research topics of 'Optical Image Simulation of SAR Image Based on UNSB'. Together they form a unique fingerprint.Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver