计算机技术与发展2026，Vol.36Issue(1)：133-139,7.DOI:10.20165/j.cnki.ISSN1673-629X.2025.0202
无人机视角下YOLOv8火焰烟雾检测算法改进
Improved Fire and Smoke Detection Algorithm Based on YOLOv8 from Perspective of Drones
摘要
Abstract
To address the issues of insufficient accuracy and low detection coverage for small targets in fire and smoke detection using deep learning-based visual recognition algorithms from a drone's perspective,an improved YOLOv8 algorithm for real-time detection of fire and smoke images is proposed.Firstly,the CBAM and SENet attention mechanism modules are introduced and integrated with the backbone network,enabling the model to extract features from fire images more effectively.Secondly,a new small object detection layer,SMOH,is added to the original three detection layers in the neck network,enhancing the model's attention to local features of small target fire images.Lastly,the original YOLOv8 loss function,CIoU,is replaced with WIoU to reduce the interference of low-quality anchor boxes on the overall training effect.Experiments show that the improved YOLOv8 algorithm has increased precision,recall,and mAP50 by3.9,3,and2.8 percentage points respectively compared to the original YOLOv8 algorithm.Compared to YOLOv5,YOLOv3,and SSD algorithms,the mAP has improved by 3.2,5.6,and 6.9 percentage points respectively.Verified through practical detection and drone-based recognition experiments,the improved YOLOv8 algorithm performs excellently in detecting small target fire and smoke images from a drone's perspective,demonstrating strong practical application significance.关键词
无人机/YOLOv8/火焰/目标检测/CBAM/SENetKey words
drones/YOLOv8/fire/object detection/CBAM/SENet分类
信息技术与安全科学引用本文复制引用
周凌云,徐元铭..无人机视角下YOLOv8火焰烟雾检测算法改进[J].计算机技术与发展,2026,36(1):133-139,7.基金项目
基础加强计划技术领域基金项目(2019-JCJQ-JJ-247) （2019-JCJQ-JJ-247）
