An Autofocus Method Based on Generative Adversarial Networks for Harmonic SAR

Harmonic synthetic aperture radar (SAR) is a new type of radar that has the ability to resist interference from fundamental frequency clutter. Due to the environment and dynamics, the harmonic SAR platform is unable to maintain uniform linear motion, which causes motion errors in the echo signals. Motion errors greatly affect the quality of harmonic SAR images and cause the image to appear defocused. In order to address this issue, a two-dimensional error phase model of harmonic SAR based on non-uniform motion is first constructed. Second, an autofocus generative adversarial network (GAN) for the harmonic SAR is constructed by using the deep learning method. Subsequently, the network is trained and tested using harmonic SAR imaging results. Finally, the focus results of the proposed method and other autofocus methods are given, and the focusing results are evaluated by using image entropy. The focusing results demonstrates the feasibility of using deep learning to solve the problem of SAR autofocus.