Synthetic Aperture Radar (SAR) is a widely applied and important remote sensing technology capable of acquiring high-resolution ground information. However, in practical operations, moving targets in SAR images often cause azimuthal defocusing, presenting a challenge to image interpretation and analysis. To address this issue, this paper proposes a deep learning-based method for moving target detection. This method combines diffusion models with an improved Unet network to generate segmentation masks for moving target detection. In terms of network architecture, we utilize a dual encoding structure that allows SAR images to be effectively introduced into the network through a conditional encoder. This mechanism largely constrains the generation of segmentation masks, thereby enhancing detection accuracy. Experimental results ultimately demonstrate that our method can effectively detect moving targets.