LUMEN: A Lightweight UAV Multi-Enhanced Network for PSD-Based RF Fingerprinting on Edge Devices.

Area of Science:

• Computer Science
• Electrical Engineering
• Artificial Intelligence

Background:

• Edge computing demands efficient Unmanned Aerial Vehicle (UAV) identification systems.
• Resource-constrained hardware limits the deployment of complex identification models.
• Real-time performance is crucial for effective UAV monitoring and security.

Purpose of the Study:

• To develop LUMEN, a Lightweight UAV Multi-Enhanced Network for accurate and efficient edge-based UAV identification.
• To enable real-time UAV detection on lightweight, single-board computer platforms.
• To balance high classification accuracy with low inference latency for practical deployment.

Main Methods:

• Utilized power spectral density (PSD)-based signal representation for UAV radio frequency (RF) data.
• Designed a lightweight neural network architecture with multi-channel PSD stacking.
• Incorporated multi-scale feature extraction to capture spectral variations and RF patterns.
• Deployed and evaluated the model on an RK3582 edge platform.

Main Results:

• LUMEN achieved a classification accuracy of 0.975, outperforming baseline models.
• Maintained an average inference latency of 1.73 ms and a throughput of 578.95 FPS.
• Demonstrated low CPU utilization, minimal memory usage, and stable thermal performance during continuous operation.

Conclusions:

• LUMEN offers a practical solution for real-time UAV identification on edge devices.
• The network effectively balances high identification accuracy with efficient runtime performance.
• The proposed method is suitable for resource-constrained environments requiring UAV detection.