REVERSIBLE MULTI-STEP DIFFUSION MODEL FOR IMAGE TRANSLATION BETWEEN RADAR AND OPTICAL DATA

In the context of expanding Earth remote sensing capabilities, increasing availability of diverse satellite data, and the need for efficient geospatial analysis, effective transformation and integration of synthetic aperture radar (SAR) and optical (RGB) imagery has become highly relevant. We propose a reversible diffusion model based on the Schrodinger bridge for bidirectional transformation of unpaired SAR and RGB images. A multi-step stochastic process progressively perturbs the data with noise, while a U-Net-based neural network denoises each step. A bidirectional scheme ensures reversibility: the forward generator converts optical images into radar, and the backward generator does the opposite. Both components are trained iteratively under an entropy-regularized stochastic control framework, aligning intermediate distributions and preserving key scene structures. The model was tested on the SEN1-2 and SN6-SAROPT datasets. According to PSNR, SSIM, and FID metrics, it surpasses traditional GAN-based approaches (e.g., CycleGAN) and one-way diffusion models. Moreover, for the complete RGB→SAR→RGB cycle, the discrepancy from the original image remains under 5–10% (SSIM). The approach can be applied to generate missing imagery, support multisensor data analysis, and mitigate cloud coverage in optical domains.

1. Prokofiev K.V., Ivanov S.V. Generatsiya radiolokatsionnykh snimkov mestnosti iz opticheskikh [Generation of radar imagery from optical]. Informatika, telekommunikatsii i upravlenie, 2022. Vol.15 no.4. pp.7–21(In Russian).

2. Zhu J., Park T., Isola P., Efros A.A. Unpaired image-to-image translation using cycle-consistent adversarial networks. Proceedings of IEEE International Conference on Computer Vision (ICCV), 2017. pp. 2242–2251.

3. Fu S., Xu F., Jin Y.Q. Reciprocal translation between SAR and optical remote sensing images with cascaded-residual adversarial networks. Science China Information Sciences, 2021. Vol. 64, 112401. Doi: 10.1007/s11432-020-3077-5

4. Ho J., Jain A., Abbeel P. Denoising diffusion probabilistic models. Advances in Neural Information Processing Systems (NeurIPS), 2020. Vol. 33. pp. 6840–6851.

5. Song Y., Sohl-Dickstein J., Kingma D.P., Kumar M., Ermon S., Poole B. Score-based generative modeling through stochastic differential equations. Proceedings of International Conference on Learning Representations (ICLR),2021. 22 p. DOI:10.48550/arXiv.2011.13456

6. De Bortoli V., Thornton J., Heng J., Doucet A. Diffusion Schrodinger bridge with applications to score-based generative modeling. Advances in Neural Information Processing Systems, 2021. V. 34. P. 17695–17709.

7. Su X., Song J., Meng C., Wu Y., Zhu J. Dual diffusion implicit bridges for image-to-image translation. Proceedings of Eleventh International Conference on Learning Representations (ICLR), 2023. 18 p. doi: 10.48550/arXiv.2203.08382

8. Li B., Xue K., Liu B., Zhao J., Tao D. BBDM: Image-to-image translation with Brownian bridge diffusion models. Proc. IEEE/CVF Conf. Computer Vision and Pattern Recognition (CVPR), 2023. P. 1952–1961.

9. Liu G.H., Vahdat A., Huang D.A., et al. I2SB: Image-to-image Schrodinger bridge. International Conference on Machine Learning (ICML), 2023. 18 p. doi:10.48550/arXiv.2302.05872

10. Kim B., Kwon G., Kim K., et al. Unpaired image-to-image translation via neural Schrodinger bridge. Proceedings of International Conference on Learning Representations (ICLR), 2024. Doi:10.48550/arXiv.2305.15086

11. Shi H., Wang Y., Xu M., et al. A brain-inspired approach for SAR-to-optical image translation based on diffusion models. Frontiers in Neuroscience, 2024. Vol. 18, 1352841. 14 p.

12. Seo M., Oh Y., Kim D., et al. Improved flood insights: Diffusion-based SAR to EO image translation. arXiv:2307.07123, 2023.

13. Wang X., Liao H., Yang Z., et al. Optical-ISAR image translation via denoising diffusion implicit model. Proc. IEEE Int. Conf. Image Processing and Computer Vision Applications (ICIPCA), 2023. pp. 1437–1441.

14. Schmitt M., Hughes L.H., Zhu X.X. The SEN1-2 dataset for deep learning in SAR-optical data fusion. ISPRS Annals of Photogrammetry, Remote Sensing and Spatial Information Sciences, 2018. Vol. IV-1. pp. 141–146.

15. Liu M.-Y., Breuel T., Kautz J. Unsupervised image-to-image translation networks. Proceedings of 31st Conference on Neural Information Processing Systems (NIPS), 2017, pp. 700–708.