Intrusion Detection System for Drones

Drones are vulnerable to cyber-attacks due to their reliance on wireless networks for communication and control. This dissertation addresses this critical need by exploring novel methodologies for drone security through advanced anomaly detection systems and enhancing communication protocols. The research is organized around three main objectives: (1) detecting abnormalities in network-side operations of drones using machine learning (ML) algorithms, (2) developing control-side anomaly detection systems using recurrent neural networks (RNNs) and long short-term memory (LSTM) models, and (3) improving the security of the MAVLink protocol without altering its core structure. The study introduces DUDE-IDS, an intrusion detection system specifically designed for drone networks. The network-side IDS utilizes supervised ML algorithms such as Gradient Boosting, Linear SVC, Decision Tree, K-NN, and Random Forest, while the control-side IDS leverages LSTM model to detect deviations from normal operational patterns. A significant contribution of this research is the creation of labeled datasets specifically tailored for network-related and control-related cyber-attacks. These datasets are instrumental in developing and evaluating the effectiveness of the proposed detection mechanisms. The dissertation further demonstrates the practical application of DUDE-IDS in a real-time drone testbed which shows its suitability for resource-constrained environments. To address MAVLink protocol vulnerabilities, this research investigates advanced symmetric authenticated encryption (AEAD) techniques, such as ChaCha20-Poly1305, AES-GCM-SIV, AES-OCB3, and AES-CCM, into the protocol without modifying its lightweight structure. The performance of these encryption schemes is validated through real-time implementation on a custom-built drone platform that ensures a balance between security and computational efficiency. This dissertation makes important contributions to drone cybersecurity by providing robust detection mechanisms for both network-side and control-side anomalies and enhancing communication protocol security. The findings of this research lay the foundation for future work on lightweight IDS and secure communication protocols tailored to drone systems.