Range-Doppler Map Augmentation by Generative Adversarial Network for Deep UAV Classification

In this paper, we report a data augmentation technique to alleviate the data shortage for UAV (Unmanned Aerial Vehi-cle) classification problem. The UAV classification problem is modeled on CNN (Deep Convolutional Neural Network), which is prevalent in artificial intelligence, and the training data consists of an RD (Range-Doppler) map of a FMCW (Frequency-Modulated Continuous-Wave) radar for a moving UAV. Getting more training data usually helps a deep CNN better generalize to test data or the real world. Therefore, we introduce a Generative Adversarial Network (GAN)-based data augmentation technique to generate synthetic RD maps used for training of UAV classifiers. By doing so, the UAV classifier was able to achieve better performance on the test dataset, especially when the classifier was trained on a smaller dataset.