The Synthetic Aperture Radar (SAR) images have a wide range of applications in agriculture, military and other fields, but there are also some problems such as noisy, poor visibility and difficulty in interpretation etc., this paper proposes a translation method based on Conditional-DCGAN with similarity loss function, after summarizing the development of Generative Adversarial Networks at the present stage. This method realizes the transfer from SAR images to optical images, so that the generated image has the gray feature, which reduces image noise and improves image resolution and interpretability., The experimental result shows that the transferred images with high resolution and obvious features have high similarity with the optical image of the same scene, which improves the interpretation of the SAR images.