Public Blockchain-Based Data Integrity Protection for Federated Learning in UAV Networks Using MAVLink Protocol

The utilization of federated learning in unmanned aerial vehicle (UAV) networks facilitates collaborative training of machine learning models by multiple UAVs while ensuring privacy preservation. However, the existing solutions for securing local model updates, such as heavy computation homomorphic encryption, secure multiparty computation, and differential privacy, are not feasible for UAV networks with limited computational resources and data capacity. To address this issue, a new lightweight protocol has been proposed, which protects the integrity of non-privacy sensitive local model updates in UAV networks using the MAVLink protocol over WiFi and LoRa communication technologies. The lightweight protocol has been designed using the SHA256 hash function and integrated with a public blockchain for integrity verification purposes. A proof of concept has been presented to demonstrate the proposed protocol’s capability of protecting the integrity of local model updates in UAV networks. Furthermore, the security of the proposed protocol has been analyzed and shown to be secure against adversary-in-the-middle and replay attacks. The computation cost of the proposed protocol has also been evaluated and found to be supported by UAV networks.