基于YOLOv7的矿工吸烟识别方法研究  

Research on Miner Smoking Recognition Method Based on YOLOv7

作　　者：王彬[1,2] 赵作鹏 WANG Bin;ZHAO Zuopeng(School of Computer Science&Technology,China University of Mining and Technology,Xuzhou 221116,China;Department of Information Technology,Jiangsu Union Technical Institute,Xuzhou 221008,China)

机构地区：[1]中国矿业大学计算机科学与技术学院,江苏徐州221116 [2]江苏联合职业技术学院信息技术系,江苏徐州221008

出　　处：《现代信息科技》2024年第6期66-69,73,共5页Modern Information Technology

摘　　要：井下矿工的吸烟行为严重影响煤矿生产安全,对井下矿工吸烟行为的有效识别迫在眉睫。针对煤矿井下的特殊环境和传统识别方法准确率低的问题,提出一种基于YOLOv7的矿工吸烟行为识别算法YOLO-SFN。将SimAM嵌入到YOLOv7的网络结构中,用Focus模块替换MPConv下分支中的3×3卷积核,提高模型在复杂背景下的特征提取能力。在后处理阶段采用Soft-NMS作为网络模型的后处理算法,解决了传统NMS算法在复杂密集环境中的漏检问题。实验结果表明,该方法的准确率为96.45%,召回率为92%,精确率为97.05%。研究成果已经在陈四楼煤矿得以推广应用,实现了对煤矿井下矿工吸烟行为的有效监管。Smoking behavior of underground miners seriously affects the production safety of coal mines,and effective recognition of underground miners'smoking behavior is imminent.Aiming at the special environment of underground coal mines and the problem of low accuracy of traditional recognition methods,it proposes a YOLOv7-based miners'smoking behavior recognition algorithm YOLO-SFN.SimAM is embedded into the network structure of YOLOv7,and the Focus module is used to replace the 3×3 convolution kernel in the lower branch of MPConv,so as to improve the model's feature extraction ability in the complex background.Soft-NMS is used as the post-processing algorithm for the network model in the post-processing stage,which solves the leakage detection problem of the traditional NMS algorithm in the complex and dense environment.The experimental results show that the accuracy rate of the method is 96.45%,the recall rate is 92%,and the precision rate is 97.05%.The research results have been popularized and applied in Chensilou coal mine,realizing the effective supervision of the smoking behavior of miners in underground coal mines.

关 键 词：目标检测 注意力机制 YOLOv7 NMS算法 吸烟识别 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]