Recent Advancements in Deep Learning for SAR-to-Optical Image Translation: A Systematic Literature Review (2021–2025)
Trevor Dove, Zhixiao Xie, Meng Tang, Yu Zhang, Xinyue Ye, Nathan Jacobs, Gongbo Liang · IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing · 2026
The fusion of Synthetic Aperture Radar (SAR) and optical remote sensing data is a cornerstone of modern Earth observation. SAR-to-Optical (S2O) image translation has emerged as a pivotal research area, aiming to leverage the all-weather, day-and-night imaging capabilities of SAR with the high interpretability and rich spectral information of optical sensors. This translation addresses critical challenges such as data gaps from cloud cover and enhances the utility of SAR data for a wide range of downstream applications. However, the task is inherently ill-posed due to the significant domain gap between the two modalities, stemming from different imaging physics, speckle noise, and geometric distortions. This paper presents a systematic literature review of the advancements in deep learning-based S2O translation from 2021 to mid-2025. We analyze the rapid architectural evolution from specialized Generative Adversarial Networks (GANs) to high-fidelity Denoising Diffusion Probabilistic Models (DDPMs) and context-aware Vision Transformers (ViTs). We chart the corresponding maturation of public datasets, from low-resolution benchmarks to high-resolution, task-specific, and multi-temporal collections. Furthermore, we examine the shift in evaluation paradigms from simple pixel-wise metrics to perceptual and, increasingly, task-based performance measures. Key downstream applications, including cloud removal, wildfire assessment, and Automatic Target Recognition (ATR), are reviewed as primary drivers of innovation. Finally, we synthesize the field's persistent challenges, such as model generalization, the fidelity-efficiency trade-off, and the data bottleneck, as well as outline promising future trajectories, including sequential translation, integration with foundation models, and physics-informed AI.