Unmanned Aerial Vehicles (UAVs) have become an essential tool for various applications such as surveillance, environmental monitoring, and disaster management. Real-time object detection from UAVs, however, is hindered by challenges such as varying object scales, changing perspectives, limited computational resources, and unpredictable environmental conditions. YOLOv8 (You Only Look Once) has been recognized for its speed and efficiency in object detection, yet it faces significant limitations when applied to UAV imagery due to these challenges. This paper proposes several advancements to YOLOv8, specifically designed to optimize its performance in UAV-based object detection. Our approach incorporates multi-scale feature fusion, dynamic anchor box optimization, lightweight architecture modifications, and enhanced post-processing strategies. These improvements address the unique challenges posed by UAV imagery, resulting in a robust and efficient model for real-time object detection. Extensive experiments on publicly available UAV datasets demonstrate the superior performance of the enhanced YOLOv8, achieving higher accuracy, faster inference times, and improved robustness under dynamic flight conditions.
Enhancing real-time object detection for UAVs using optimized YOLOv8 architecture
International Conference on Computer Graphics, Artificial Intelligence, and Data Processing (ICCAID 2024) ; 2024 ; Nanchang, China
Proc. SPIE ; 13560
2025-04-10
Conference paper
Electronic Resource
English
Real-Time Road Damage Detection Using YOLOv8
ASCE | 2025
|Drone design for object detection using YOLOv8
IEEE | 2025
|