Journal of Contemporary Politics

Abstract

Unmanned Aerial Vehicles (UAVs), often referred to as drones, are creating growing difficulties in security‑critical zones, demanding dependable automated detection solutions. In this work we introduce a bespoke real‑time drone detection platform built on the YOLOv8 object‑detection framework, which was trained on a small collection of 15 drone photographs that were broadened via extensive data‑augmentation methods. The resulting augmented set comprised more than 150 enriched training instances, enhancing the model’s resilience to changes in illumination, viewpoint, size, and background intricacy. The YOLOv8‑S architecture was trained on a CPU with a 50‑epoch workflow employing default hyperparameters and transferred pre‑trained weights. The resulting model reached a precision of 82–88 %, recall of 75–83 %, and a mAP@50 of 0.86, indicating robust detection performance even with the limited data. For rollout, a Streamlit‑driven web interface was built, allowing users to carry out drone identification via uploaded images, webcam feeds, or live video streams. Real‑world trials revealed that the solution correctly identified drones in 200 of 175 test images and kept an average inference time of 60–70 ms per frame, rendering it fit for near‑real‑time use. Overall, this project demonstrates that accurate drone detection is achievable even with minimal training data when supported by augmentation and a modern YOLO detector. The combined system provides a lightweight, practical solution for surveillance and monitoring environments, with potential extensions into multi-drone tracking, long-range detection, and thermal-camera integration.

Keywords: Drone Detection, YOLOv8, Object Detection, Deep Learning, Real-Time Surveillance