Generating Radar Video Data Using Generative Adversarial Nets

Abstract

In development of radar tools, for example machine learning for target classification and simulation of processing chains, there is a need for large amounts of complexvalued data recorded under real conditions by a radar. However, large and properly labelled data sets of this kind are time consuming and expensive to collect. Currently such applications instead use simulated data, or smaller sets of real recorded data, limiting the development of new applications. In this thesis, we investigate the possibility to utilise Generative Adversarial Nets (GANs) to extend existing radar data sets in order to get out of this low data regime. Existing techniques are combined and further developed for generating complex-valued radar data, with analysis of the quality of the generated data and its relevance. We conclude that in this context, GANs could be used to extend existing radar data sets, though more work is needed to make it perfectly realistic, which is non-trivial.