In deep learning based image processing, the number of dataset is important to train the designed model. However, it is hard to secure SAR images, because satellite-based SAR is limited and high-resolution images are very expensive. Generative adversarial network (GAN) supplements this problem by learning two models, generator and discriminator, in an adversarial process at the same time. In this paper, we take one dataset as input data, and compare its accuracy using GAN models. CycleGAN is used to generate images among GAN models. Optical images are used for dataset and Chinese cities are selected for SAR images. The lack of dataset, a drawback of SAR images, is supplemented using data augmentation. SSIM, MSE, and PSNR of fake and original images are calculated using dataset and show that CycleGAN has the most lower MSE with 639.4379 and highest PSNR with 20.0728. Pix2pix has the most highest SSIM with 0.7842.
ACKNOWLEDGMENT This research was supported by the MSIT(Ministry of Science and ICT), Korea, under the ITRC(Information Technology Research Center) support program(IITP-2021-2018-0-01424) supervised by the IITP(Institute for Information & communications Technology Promotion)