软件导刊2025，Vol.24Issue(6)：185-191,7.DOI:10.11907/rjdk.241396
YOLOv8n-GCB:轻量化的烟火检测网络
YOLOv8n-GCB:Lightweight Fire Detection Network
摘要
Abstract
A lightweight multi-scale fireworks recognition algorithm YOLOv8n GCB is proposed to address the low accuracy of flame and smoke detection in YOLOv8n.Firstly,adding an additional small target detection head to enhance the fusion effect of contextual information and features,and improve the recognition rate of the model for small fireworks targets;Secondly,the integration of GAM attention mechanism in the network solves the problems of background interference and insufficient regional attention in complex fire scenes;Again,modify the fea-ture fusion area in the network to a cross layer weighted bidirectional feature pyramid network,making the fusion of smoke and fire features more complete and uniform,in order to improve the accuracy of the model's predicted boxes;Finally,EIOU is used as the loss function of the model to improve the impact of target boxes of different sizes on model accuracy.The experiment shows that the proposed algorithm has an aver-age recognition accuracy of 81.2%for flames and smoke in embedded devices,which is 6.1%higher than the original YOLOv8n algorithm;The detection speed for video streams is 33.4 FBS,which has significant advantages compared to current mainstream object detection networks and can achieve real-time detection results.关键词
烟火识别/YOLOv8n/小目标检测/跨层加权双向特征金字塔/嵌入式设备Key words
fire detection/YOLOv8n/small object detection/cross layer weighted bidirectional feature pyramid/embedded devices分类
计算机与自动化引用本文复制引用
韩逸,李敏..YOLOv8n-GCB:轻量化的烟火检测网络[J].软件导刊,2025,24(6):185-191,7.基金项目
湖北省教育厅科学技术研究计划重点项目(D20211701) （D20211701）
中国高校产学研创新基金项目(2020HYA02015) （2020HYA02015）
