End-To-End Framework for Real-Time Drone Detection and Alerting Using Lightweight Deep Learning

Abstract

Real-time drone detection is a significant issue in airspace security and privacy because of the increasing use of UAVs; this issue affects airport security directly as it results in various inclusions such as airport incursions and unauthorized surveillance. We proposed end-to-end framework for real time drone detection and alerting integrating multiple YOLO architectures, our system trained on a largescale composite dataset of over 77,000 labeled images for drones, expanded to approximately 90,000 through targeted data augmentation. The Yolo models first trained and evaluated independently, then combined using ensemble learning strategies to reduce false detections while preserving real-time performance. The final ensemble of (YOLOv8, YOLOv11, and YOLOv12) achieves 96.1% accuracy and 93.9% mAP@0.5 Index Terms—Drone Detection, YOLO, Ensemble Learning, Real-Time Alert Systems, Unmanned Aerial Vehicles (UAVs), Lightweight DL Architecture.