Abstract: Optical remote sensing imagery is often affected by cloud cover, leading to the loss of ground object information. Therefore, cloud removal from remote sensing imagery is crucial for improving the utilization of these images. Synthetic Aperture Radar (SAR) imagery, which can penetrate clouds and capture surface scattering information, provides effective information for recovering optical data obscured by clouds. However, due to significant differences in the imaging modes of optical and SAR data, existing cloud removal methods exhibit weak recovery capabilities in areas with complex surface details, such as urban regions. To address this issue, this study proposes a multi-source, multi-temporal cloud removal method for remote sensing imagery based on a dual adversarial network. First, a cloud removal dataset (UCR-1.0) is constructed, covering major cities worldwide and including multi-temporal optical imagery and SAR data from Sentinel. Then, a Dual Adversarial Cloud Removal Network (DGAN) is introduced. Through dual adversarial collaborative optimization and an adaptive attention mechanism, the proposed method enhances the spatial consistency and semantic fidelity of cross-modal remote sensing image generation, while accurately repairing the cloud-contaminated areas in optical imagery. Finally, experiments are conducted using UCR-1.0 to train and test multiple models. The results show that DGAN outperforms other models in terms of SSIM and PSNR metrics, demonstrating the effectiveness of the proposed method in urban region restoration.