Feature-guided multi-stage generative adversarial network for SAR image generation

At present, relatively few image generations focus on the characteristics of synthetic aperture radar (SAR) images. Focusing on the problem of SAR image generation, a feature-guided multi-stage generative adversarial network (FGMS-GAN) algorithm is proposed in this paper, which mainly consists of three stages: the background stage, the shape stage, and the target stage. In the background stage, the statistical characteristics of SAR images are modeled by using the G0 distribution to ensure that the generated images conform to the statistical properties of real SAR images. In the shape stage, the azimuth angles of SAR images are used to achieve higher similarity in shape and geometry between the generated and real SAR images. In the target stage, the scattering center characteristics of SAR images are extracted and applied to the generative network of target stage, which can enhance the detailed representation of the target and improve recognition and clarity in the target region. The moving and stationary target recognition (MSTAR) dataset is utilized to train and generate SAR images. Experimental results have verified the effectiveness of the proposed algorithm.