Translation from dual-polarization complex-valued SAR to optical satellite images

Synthetic Aperture Radar (SAR) delivers all-weather, high-resolution imagery, yet its unique imaging mechanism significantly reduces visualization and interpretability. Transforming SAR images into optical representations enhances their interpretability and compensates for data gaps in optical imagery caused by adverse weather. While existing SAR-to-optical translation algorithms use only the intensity data and abandon the phase information in translating into optical images, this work suggests the Single Look Complex (SLC) SAR and challenges the SLC-to-optical image translation. To address the issue, the S1SLCS2 dataset is constructed, which pairs SLC, Ground Range Detected (GRD), and optical images. A conditional diffusion model is designed, which incorporates spatial-frequency domain feature extraction and operates on the covariance matrix derived from complex-valued dual-polarization SLC images. The proposed method is evaluated on the new dataset and compared against five state-of-the-art methods by adapting them to SLC-to-optical translation. Quantitative evaluations confirm that the translation from dual-polarization SLC can compete with the GRD-to-optical translation when the covariance matrix and suitable networks are provided.