Real-Time Detection Sensor for Unmanned Aerial Vehicle Using an Improved YOLOv8s Algorithm

Area of Science:

• Computer Vision
• Artificial Intelligence
• Aerospace Engineering

Background:

• Unmanned Aerial Vehicle (UAV) localization is critical for low-altitude economy applications.
• Conventional YOLOv8s algorithms struggle with missed detections due to single-frame limitations.
• Unauthorized drone activity, or "black-flying," poses significant airspace management challenges.

Purpose of the Study:

• To improve the accuracy and real-time detection of unauthorized UAVs.
• To overcome the limitations of single-frame feature reliance in existing algorithms.
• To enhance UAV trajectory prediction and tracking capabilities.

Main Methods:

• Integration of a Long-Short-Term Memory (LSTM) network with the YOLOv8s framework.
• Utilizing time-series modeling for historical feature retention and dynamic trajectory prediction.
• Employing a combined loss function (bounding box regression and binary cross-entropy) optimized with the Adam algorithm.
• Validating training data distribution using Monte Carlo random sampling for improved generalization.

Main Results:

• Significant enhancement in UAV detection performance compared to conventional methods.
• Demonstrated robustness in detecting unauthorized drones, even in complex scenarios.
• Achieved a low false negative rate when deployed on an RK3588 embedded system.

Conclusions:

• The proposed LSTM-enhanced YOLOv8s algorithm offers superior UAV detection and tracking.
• The method shows strong potential for practical applications in airspace management and counter-UAV operations.
• Real-time performance and accuracy make it suitable for critical security applications.