Is this project an undergraduate, graduate, or faculty project?
Graduate
group
What campus are you from?
Daytona Beach
Authors' Class Standing
Graduate Student
Lead Presenter's Name
Kanchon Gharami
Faculty Mentor Name
Shafika Showkat Moni
Abstract
Unmanned Aerial Vehicles (UAVs) are gaining wide acceptance from different sectors, including public services, military, emergency response, and commercial applications. While the potential benefits of UAVs are growing significantly, they can exhibit unexpected behavior due to sensor malfunctions, unforeseen environmental circumstances, or power outages. These anomalies can severely affect UAV missions, especially in swarm-based UAV operations, by compromising decision-making and trajectory planning processes. Traditional intrusion detection systems (IDS) typically rely on binary classification for individual UAVs using computationally expensive neural networks, limiting their ability to predict multiple types of attacks while operating within the resource-constrained environments of real-world UAV scenarios. To address these challenges, we propose a federated continuous learning approach to facilitate decentralized training across diverse UAV swarms using heterogeneous datasets while preserving data privacy. By leveraging a lightweight CNN(Convolutional Neural Network) – LSTM (Long Short Term Memory) model that captures both spatial and temporal features, our method significantly improves multi-class classification accuracy while using a model that has five times fewer parameters and computationally faster than traditional approaches. Experimental results demonstrate significant improvements, with detection accuracies of 96.85% on the TLM-UAV dataset, 99.45% on UKM-IDS, 99.99% on UAV-IDS, and 98.05% on Cyber-Physical datasets. This research showcases the potential of federated learning to enhance the security of UAV swarm networks through robust multi-class classification.
Did this research project receive funding support from the Office of Undergraduate Research.
No
A Lightweight and Efficient Multi-Class Intrusion Detection Scheme Based on Federated Continuous Learning for UAV Swarm Networks
Unmanned Aerial Vehicles (UAVs) are gaining wide acceptance from different sectors, including public services, military, emergency response, and commercial applications. While the potential benefits of UAVs are growing significantly, they can exhibit unexpected behavior due to sensor malfunctions, unforeseen environmental circumstances, or power outages. These anomalies can severely affect UAV missions, especially in swarm-based UAV operations, by compromising decision-making and trajectory planning processes. Traditional intrusion detection systems (IDS) typically rely on binary classification for individual UAVs using computationally expensive neural networks, limiting their ability to predict multiple types of attacks while operating within the resource-constrained environments of real-world UAV scenarios. To address these challenges, we propose a federated continuous learning approach to facilitate decentralized training across diverse UAV swarms using heterogeneous datasets while preserving data privacy. By leveraging a lightweight CNN(Convolutional Neural Network) – LSTM (Long Short Term Memory) model that captures both spatial and temporal features, our method significantly improves multi-class classification accuracy while using a model that has five times fewer parameters and computationally faster than traditional approaches. Experimental results demonstrate significant improvements, with detection accuracies of 96.85% on the TLM-UAV dataset, 99.45% on UKM-IDS, 99.99% on UAV-IDS, and 98.05% on Cyber-Physical datasets. This research showcases the potential of federated learning to enhance the security of UAV swarm networks through robust multi-class classification.