Exploring deep transfer learning interference classification on neural style transfer generated synthetic SAR datasets

Synthetic aperture radar (SAR) imagery has received a great deal of attention in recent years due to the deployment of many cutting edge spaceborne radar systems providing high resolution imagery. However, severe image distortion is a critical problem, and this is often a result of radio frequency interference (RFI) and noise. Issues that arise from distortion include missing detection and inaccurate height maps. SAR images are particularly important for classification and automatic target recognition (ATR) tasks. For such applications, access to comprehensive databases of SAR images as well as SAR images contaminated with RFI and noise is critical to enable the effective training and optimisation of classification algorithms and to provide a common baseline for benchmarking purposes. Given these challenges, the purpose of this paper is to show that neural style transfer can be used to induce RFI and noise into SAR images. We can also further classify the type of contamination using image classification techniques. The experimental data is shown to verify the efficiency of our approach.