基于改进YOLOv8n算法的井下人员行为检测研究  

Research on Underground Personnel Behavior Detection Based on Improved YOLOv8n Algorithm

作　　者：王庆磊 杨宇航 陈立业 于浩然 杜昊杰 于梓晗 史健婷 Wang Qinglei;Yang Yuhang;Chen Liye;Yu Haoran;Du Haojie;Yu Zihan;Shi Jianting(School of Computer and Information Engineering,Heilongjiang University of Science and Technology,Harbin 150022,China;Graduate College,Heilongjiang University of Science and Technology,Harbin150022,China)

机构地区：[1]黑龙江科技大学计算机与信息工程学院,哈尔滨150022 [2]黑龙江科技大学研究生学院,哈尔滨150022

出　　处：《煤矿机械》2025年第5期195-200,共6页Coal Mine Machinery

基　　金：黑龙江大学生创新创业国家级重点项目(202310219130S);黑龙江省省属高校基本科研业务费项目(2023-KYYWF-0547)。

摘　　要：煤矿井下复杂环境中的干扰信息、低照明度及机械设备遮挡导致现有目标检测算法在人员异常动作检测中面临精度低、漏检和误检的挑战。提出了一种改进的YOLOv8n算法,旨在提升井下人员行为检测的速度和精度。首先,把iRMB和EMA融合创新了iREMA注意力机制,得到C2f_iREMA,增强了复杂背景中对特征目标的定位能力;其次,在C2f中引入iAFF,以有效地整合来自不同层次的特征;最后,引入DySample模块,避免了传统下采样方法可能带来的信息损失,提高了检测精度。实验表明,该算法在井下工人行为检测任务中,精确率提升了0.8%,召回率提升了3.1%,mAP@0.5提升了3.2%。这些改进显著增强了模型的检测精度、召回能力及整体性能,适用于井下环境中的实时行为检测。The interference information,low illumination and mechanical equipment obstruction in the complex environment of underground coal mine lead to pose challenges to the low accuracy,missed detection and false detections by existing object detection algorithms in detecting abnormal human movements.Proposed an improved YOLOv8n algorithm aimed at improving the speed and accuracy of underground personnel behavior detection.Firstly,the fusion of iRMB and EMA innovated the iREMA attention mechanism,resulting in C2f_iREMA,which enhances the localization ability of feature targets in complex backgrounds.Secondly,iAFF was introduced in C2f to effectively integrate features from different levels.Finally,the DySample module was introduced to avoid the information loss that traditional downsampling methods may cause and improve detection accuracy.Experiments show that this algorithm improves accuracy by 0.8%,recall rate by 3.1%,and mAP@0.5 by 3.2%in underground worker behavior detection tasks.These improvements significantly enhance the detection accuracy,recall capability and overall performance of the model,and are suitable for real-time behavior detection in underground environments.

关 键 词：YOLOv8n 行为检测 注意力机制 DySample 特征融合 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TD76[自动化与计算机技术—控制科学与工程]