YOLOv8‑Based Drone Detection: Building a Robust Model with Extensive Data

Abstract

Drone (UAV) applications are rapidly expanding in surveillance, photography, and delivery services, raising safety and privacy concerns. Effective detection systems that can identify and track drones in real time are essential. In this study, we present a comprehensive dataset and propose an advanced drone detection model based on the YOLOv8 architecture.

Introduction

Widespread drone adoption creates an urgent need for reliable detection systems to ensure public safety. Traditional object detection methods struggle with drones due to their small size, fast motion, and diverse appearances. Consequently, advanced models capable of accurately recognizing drones in complex environments are required.

Dataset

The dataset is sourced from the publicly available Kaggle YOLO Drone Detection Dataset. It comprises diverse annotated images captured under various environmental conditions and camera viewpoints, featuring drone instances alongside other common objects to facilitate robust detection and classification.

Methodology

We employ the YOLOv8 framework—an efficient, single‑stage detector that simultaneously predicts bounding boxes and class probabilities. YOLOv8 delivers real‑time performance, making it well‑suited for drone detection applications.

Experimental Setup

Training and evaluation are conducted on the Colab platform, leveraging GPU acceleration. The curated dataset is used to train the YOLOv8 model, with hyper‑parameter tuning to maximize detection accuracy and efficiency.

YOLO

• Single‑Shot Detection: Unlike region‑proposal methods, YOLO performs detection in a single pass by dividing the input image into a grid and predicting boxes and class probabilities for each cell.
• Grid‑Based Prediction: The image is split into a fixed‑size grid (e.g., 13×13). Each cell predicts multiple bounding boxes (each with a confidence score) and class probabilities.
• Anchor Boxes: Pre‑defined boxes of various shapes and sizes allow YOLO to handle objects of different scales and aspect ratios. The network predicts offsets relative to these anchors.
• Training: Optimization minimizes a combination of localization loss (bounding‑box accuracy) and classification loss (class prediction accuracy).
• Speed‑Accuracy Trade‑off: YOLO sacrifices some localization precision for real‑time speed, yet achieves competitive accuracy compared with slower methods such as Faster R‑CNN.

Keywords

• Drone Detection
• YOLOv8
• Object Detection
• Deep Learning
• Surveillance
• Safety

Results and Discussion

We present comprehensive performance metrics on training and test sets, including precision, recall, and F1 score. The model’s behavior under varying environmental conditions is analyzed, and strengths and limitations are discussed.

Conclusion

By providing a thorough dataset and an advanced YOLOv8‑based model, this work addresses the critical need for reliable drone detection in autonomous driving and related domains. The dataset and model contribute valuable resources for enhancing safety and privacy measures where drones pose potential risks.