Federated learning for drone authentication
Department
Software Engineering and Game Development
Document Type
Article
Publication Date
9-1-2021
Abstract
The ever-rising applications of drones in the Internet of Things (IoT) era is offering many opportunities and challenges. Owing to drone abilities (silent flying, capturing photos and videos, etc.), there is widespread concern about drone authentication and which drones allow to fly. In this regard, there are several machine learning (ML) proposals for authentication in IoT networks. Such ML-based models have drawbacks in data security, privacy-preserving, and scalability when applied in drone authentication. ML-based methods collect all data and centrally train the authentication model, exposing the model to adversarial situations. This paper proposes a federated learning-based drone authentication model with drones’ Radio Frequency (RF) features in IoT networks. In the proposed model, the Deep Neural Network (DNN) architecture is implemented for drone authentication with Stochastic Gradient Descent (SGD) optimization performed locally on drones. Also, Homomorphic encryption and the secure aggregation method are applied to secure model parameters. Experimental results show that the federated drone authentication model gains a high true positive rate (TPR) during drone authentication and better performance compared to other ML-based models.
Journal Title
Ad Hoc Networks
Journal ISSN
15708705
Volume
120
Digital Object Identifier (DOI)
10.1016/j.adhoc.2021.102574