机械与电子2025，Vol.43Issue(11)：33-39,7.
基于改进YOLOv11的复杂环境下火焰目标检测
Flame Target Detection in Complex Environments Based on Improved YOLOv11
摘要
Abstract
To address the real-time and accuracy demands of flame target detection in complex envi-ronments,this study proposes an enhanced YOLOv11 flame detection model.In terms of model optimiza-tion,the C3k2_Ghost module is introduced to reduce computational complexity,and a Feature Refinement Module(FRM)is designed to enhance multi-scale feature representation.Furthermore,the SEAM atten-tion mechanism is employed to optimize detection performance in occlusion scenarios,and an additional de-tection layer at 160×160 is incorporated to improve the ability to capture fine details.Experimental results based on a dataset of 4 653 flame images demonstrate that the improved model achieves a mean Average Precision(mAP)of 86.1%,a recall rate of 84.3%,and a detection speed of 64.1 FPS,significantly outper-forming mainstream models such as YOLOv11 and YOLOv8 in both accuracy and efficiency.The results indicate that the proposed model effectively balances lightweight design and feature representation capabili-ties,providing an efficient solution for real-time fire hazard warning in complex industrial scenarios.关键词
深度学习/目标检测/YOLO模型/多尺度特征/注意力机制Key words
deep learning/target detection/YOLO model/multi-scale features/attention mechanism分类
信息技术与安全科学引用本文复制引用
郭庆,陈川,季强东,权晓柯,易灿灿..基于改进YOLOv11的复杂环境下火焰目标检测[J].机械与电子,2025,43(11):33-39,7.基金项目
国家自然科学基金资助项目(51805382) （51805382）
湖北省应急管理厅安全生产专项资金科技项目(GEM-CK-JS-2023033101) （GEM-CK-JS-2023033101）
