Bridging a Gap in the Synthetic and Measured SAR Images Based on Data Augmentation Methods

Creating simulated synthetic aperture radar (SAR) images using electromagnetic algorithms is an effective way to augment insufficient measured SAR data in automatic target recognition applications. However, discrepancies in the modeling process and simplifications of the electromagnetic characteristics inevitably cause some differences between synthetic and mea-sured images. Various other data augmentation methods also exist, such as noise addition, rotation, and numerous generative adversarial network (GAN) techniques. This paper quantitatively discusses the effects of different augmentation methods and provides recommendations for SAR image augmentation.