Generating SAR Images Based on Neural Network

Abstract

Compared with measurement, electromagnetic simulation can greatly reduce time and funding cost in SAR imaging. But there are still many differences between simulated and measured SAR images since the simulation is hard to take stochastic environments into account. In this paper, a cycle generative adversarial neural network, which can generate SAR images by learning the mapping between simulated SAR images and measured SAR images (MSTAR datasets), is constructed. The generated SAR images can be purely similar with measured SAR images.