燕山大学学报2026，Vol.50Issue(2)：112-120,9.DOI:10.3969/j.issn.1007-791X.2026.02.002
基于改进YOLOv8的城市火灾检测算法
City fire detection algorithm based on improved YOLOv8
摘要
Abstract
In view of the fact that the traditional fire detection algorithm has low detection accuracy and high false detection rate in complex urban backgrounds,a city fire detection algorithm based on improved YOLOv8 is proposed.Firstly,based on the YOLOv8 object detection model,within the neck network,the Bi-directional Feature Pyramid Network structure is introduced to replace the Path Aggregation Network-Feature Pyramid Network feature fusion layer,fusing multi-scale feature information and enhancing the model's feature learning ability.Secondly,the Efficient Multi Scale Attention mechanism is integrated into the BiFPN to improve the network's feature extraction capability and further enhance the accuracy of smoke and fire detection.Finally,the partial convolution module is introduced into the backbone network to replace the C2f module with the C2f-Faster module,improving the detection efficiency of the model and reducing redundant calculations.The improved algorithm is applied to a self-compiled dataset of smoke and fire for experimentation.The result demonstrates that the improved model achieved a mAP@50 of 73.6%compared to the original model,reduced the number of parameters by 8.99%,and reduced the computational complexity to 7.7 GFLOPs.While enhancing the detection accuracy,the model has been lightweight.The improved model can meet the requirements of smoke and fire detection in complex urban backgrounds.关键词
烟火检测/YOLOv8/多尺度融合/EMA/轻量化Key words
fire detection/YOLOv8/multi-scale fusion/EMA/lightweight分类
信息技术与安全科学引用本文复制引用
苏连成,贾潇彬,丁伟利..基于改进YOLOv8的城市火灾检测算法[J].燕山大学学报,2026,50(2):112-120,9.基金项目
河北省自然科学基金资助项目(F2024203051) （F2024203051）
广西科技重大专项项目(桂科AA22067064) （桂科AA22067064）
