Hybrid Blockchain-Assisted Federated Learning Intrusion Detection System for DoS Attacks in UAV Sensor Network

Unmanned Aerial Vehicles (UAVs) are increasingly deployed in mission-critical domains, raising significant security concerns. This paper presents a Federated Learning-based Intrusion Detection System (PureChain-FL) for UAV clients, focusing on Denial-of-Service (DoS) attack detection. To enhance trust and auditability, the proposed architecture integrates PureChain, a lightweight blockchain framework based on the Proof of Authority Association ($\text{PoA}^{2}$) consensus algorithm for secure logging of detected threats. The system features client-based localized model training using the WSN-DS dataset and a global FL server for model aggregation. A smart contract governs audit logging, enabling verifiable and tamper-resistant record-keeping. Evaluation of multilayer perceptron (MLP), one-dimensional convolutional neural network (1D CNN), and CNN combined with long-short-term memory (CNN-LSTM) models over 20 FL rounds with five clients using the Flower framework and TensorFlow in a Google collaborative environment revealed CNN-LSTM as the best-performing model with 99.3 % accuracy. Figures and diagrams illustrating system architecture, communication flow, and model training are integrated into the discussion for clarity. The results show promise in the system's effectiveness in privacypreserving and trustworthy intrusion detection for unmanned aerial wireless sensor networks.