王妍玮, 王钰, 吕东翰,等.基于YOLOv11-MCF的煤矿井下作业头盔佩戴检测方法[J].河南理工大学学报（自然科学版）,2026,45(3):1-9.

WANG Y W, WANG Y, LYU D H, et al. Helmet wearing detection method for underground coal mine operations based on YOLOv11-MCF[J]. Journal of Henan Polytechnic University (Natural Science) ,2026,45(3):1-9.

基于YOLOv11-MCF的煤矿井下作业头盔佩戴检测方法

王妍玮1, 王钰1, 吕东翰2, 金戈2, 王浩3, 杨晨升1

1.黑龙江科技大学 机械工程学院，黑龙江 哈尔滨  150022；2.中国安全生产科学研究院 预警中心，北京  100012；3.煤炭科学研究总院有限公司 矿山人工智能研究院，北京 100013

摘要:目的 在煤矿井下光照不均、粉尘干扰等复杂环境下，现有头盔检测方法存在漏检及误检率高等问题。为此，提出一种基于YOLOv11-MCF（YOLOv11 with multi-scale convolutional block，convolutional block attention module，and more focused intersection over union）的煤矿井下作业头盔佩戴检测方法。 方法 首先，将MSCB（multi-scale convolutional block）模块与C3k2相结合，深度可分卷积与多尺度卷积相结合，捕捉多尺度的特征，通过残差连接和通道混洗增强特征的传播和交互能力；其次，在检测头部前引入卷积注意力机制模块（convolutional block attention module，CBAM），提升模型对头盔特征的提取能力，抑制背景干扰；最后，针对样本不平衡问题，采用Focaler-IoU（more focused intersection over union）损失函数，增强模型在昏暗环境下对难以区分样本的关注度，以更好地处理正负样本的不平衡，同时显著提高边界框的回归精度，进一步提升检测精度。基于CUMT-HelmeT数据集进行实验，通过对比实验确定最佳改进策略，并通过消融实验验证组合策略的优越性，在相同训练参数与硬件环境下，迭代100轮次。  结果 实验结果表明，YOLOv11-MCF算法检测头盔精度达到86.7%，分别比YOLOv11和YOLOv13提高了1.8%和4.2%，同时满足实时性的要求。  结论 通过改进之后得到的YOLOv11-MCF模型能更好地提高对安全头盔的检测能力，为煤矿作业人员安全头盔检测提供了一种新的解决思路。

关键词:煤矿安全;目标检测;卷积注意力机制;多尺度卷积块;头盔

doi:10.16186/j.cnki.1673-9787.2025080010

基金项目:国家自然科学基金资助项目（52304169）；黑龙江省科研基本业务费项目（2023-KYYWF-0540）；新疆维吾尔自治区重点研发课题资助项目（2022B03004-2）

收稿日期:2025/08/08

修回日期:2025/10/28

出版日期:2026/04/28

Helmet wearing detection method for underground coal mine operations based on YOLOv11-MCF

Wang Yanwei1, Wang Yu1, Lyu Donghan2, Jin Ge2, Wang Hao3, Yang Chensheng1

1.School of Mechanical Engineering， Heilongjiang University of Science and Technology， Harbin  150022， Heilongjiang， China；2.Early Warning Center， China Academy of Safety Science and Technology， Beijing  100012， China；3.China Coal Research Institute Co.， Ltd.， Research Institute of Mine Artificial Intelligence， Beijing  100013， China

Abstract: Objectives In complex underground coal mine environments characterized by uneven illumination and dust interference， existing helmet detection methods suffer from high miss detection and false alarm rates. For this， A helmet wearing detection method for underground coal mine operations based on YOLOv11 MCF （YOLOv11 with multi scale convolutional block， convolutional block attention module， and more focused intersection over union） is proposed. Methods First， the MSCB （multi scale convolutional block） module is combined with C3k2， integrating depthwise separable convolution with multi scale convolution to capture multi scale features， while residual connections and channel shuffling enhance feature propagation and interaction. Second， the convolutional block attention module （CBAM） is introduced before the detection head to improve the model’s ability to extract helmet features and suppress background interference. Finally， to address the class imbalance problem， the Focaler IoU （more focused intersection over union） loss function is adopted， which enhances the model’s attention to hard to distinguish samples under dim lighting， better handles the imbalance between positive and negative samples， and significantly improves bounding box regression accuracy， further boosting detection precision. Experiments are conducted on the CUMT HelmeT dataset. Comparative experiments determine the optimal improvement strategy， and ablation studies verify the superiority of the combined strategy. All models are trained for 100 epochs under identical training parameters and hardware conditions.  Results Experimental results show that the YOLOv11 MCF algorithm achieves a helmet detection accuracy of 86.7%， which is 1.8% and 4.2% higher than that of YOLOv11 and YOLOv13， respectively， while also meeting real time requirements.  Conclusions The improved YOLOv11 MCF model can significantly enhance the detection capability for safety helmets， providing a new solution for safety helmet detection of coal mine workers.

Key words:coal mine safety;object detection;convolutional block attention module;multi-scale convolutional block;helmet