DP-GAN: A Novel Generative Adversarial Network-Based Drone Pilot Identification Scheme

With the development of the Industrial Internet of Things, drones have been applied to many application scenarios for providing reliable communication services. However, due to being deployed in an open environment, drones often suffer from impersonation attacks, posing significant threats to flight safety. Therefore, designing an effective pilot identification scheme is crucial for the safe flight of drones. Currently, some pioneer works have been devoted to pilot identification. Limited by the insufficient training samples, the performance of pilot identification still needs to be improved. For this reason, a novel generative adversarial network (GAN)based drone pilot identification scheme named DP-GAN has been proposed to improve pilot identification performance. Specifically, we first construct a long short-term memory (LSTM)-based generator to estimate the distribution of the collected dataset and then utilize the temporal and spatial relationships among the received control commands and drone attitudes to produce realistic flight data. Moreover, we also designed a three-stage adversarial training strategy to optimize the generator and discriminator simultaneously. Since the well-constructed generator could produce realistic flight data, the pilot identification performance of the discriminator could be further enhanced. After being verified by systematic experiments, the proposed scheme has achieved 94.42% and 97.02% accuracy under natural and constrained environments on S500. Thanks to the lightweight system overhead, this scheme holds the potential to be deployed on the drone platform for real-time pilot identification.