Intrusion Detection Model of UAV System Based on Machine Learning and Neural Network

Wengang Zhou¹Bo Li^2,3*Xue Zeng³

• ¹Civil Aviation Flight University of China, Guanghan, China
• ²Information Department of Pengzhou People's Hospital, Chengdu, China
• ³Advanced Cryptography and System Security Key Laboratory of Sichuan Province, Chengdu, China

Abstract—With the increasing complexity and scale of cyber attacks, the intrusion detection of unmanned aerial vehicle systems has become a major challenge in the field of modern network security. Traditional traffic detection methods lack efficient modeling of local and global features, making it difficult to capture complex data patterns. Our research focuses on Unmanned Aerial Vehicle intrusion detection, which aims to identify unauthorized access, malicious data injections, or normal - operation disruptions in Unmanned Aerial Vehicle systems. Unmanned Aerial Vehicles have unique features like limited battery life, restricted data - transmission distance, and small data - storage capacity. Malicious activities can disrupt power usage, communication, and data storage, highlighting the need for dedicated intrusion - detection research. To address this issue, we propose an intrusion detection model for unmanned aerial vehicle systems that integrates machine learning and neural networks. Firstly, the drone data is cleaned up, and traditional feature selection techniques are used to select key and non key features. The non key features are mapped to the key features through CNN+LSTM, and the fused features are used as inputs for the model. Then, machine learning and neural networks are combined to detect the traffic of the drone network. Through testing on public datasets ISCXVPN2016, CICIDS2018, TON IoT and CIC IoT 2023, and comparing with three major feature selection techniques (filtering, packaging, and embedding), our method can improve accuracy by up to 3%, F1 by up to 4%, and recall by up to 3%.