基于YOLOv8-ECW的井下人员行为实时检测算法  

Real-time detection algorithm of underground personnel behavior based on YOLOv8-ECW

作　　者：骆津津 陈伟[1,3,4] 田子建 张帆[1,3] 刘毅[1,3] LUO Jinjin;CHEN Wei;TIAN Zijian;ZHANG Fan;LIU Yi(School of Artificial Intelligence,China University of Mining and Technology-Beijing,Beijing 100083,China;Datang Binzhou Power Generation Co.,Ltd.,Binzhou Shandong 256651,China;Key Laboratory of Intelligent Mining and Robotics,Ministry of Emergency Management,Beijing 100083,China;School of Computer Science and Technology,China University of Mining and Technology,Xuzhou Jiangsu 221116,China)

机构地区：[1]中国矿业大学(北京)人工智能学院,北京100083 [2]大唐滨州发电有限公司,山东滨州256651 [3]煤矿智能化与机器人创新应用应急管理部重点实验室,北京100083 [4]中国矿业大学计算机科学与技术学院,江苏徐州221116

出　　处：《矿业科学学报》2025年第2期316-327,共12页Journal of Mining Science and Technology

基　　金：国家自然科学基金(52274160,51874300,52074305,52374165,52121003)。

摘　　要：针对现有煤矿井人员行为检测模型存在精度低、计算量大等问题,提出一种基于YOLOv8-ECW的井下人员行为实时检测算法。算法在YOLOv8n的基础上对骨干网络进行改进,提出多尺度卷积模块EMSC,再与C2f卷积相结合设计出C2f_EMSC模块,有效捕获目标的多尺度特征,减少模型的计算量、参数量;在网络中引入CGBlock下采样模块融合全局的上下文信息,引入WIoU损失函数提升检测框的定位精度和模型收敛速度。在矿井人员行为检测数据集上进行实验,结果表明:①相比于基线YOLOv8n模型,YOLOv8-ECW模型对各类目标平均精度均值mAP50为92.4%,上升了2.1%;mAP50-95为75.4%,上升了4.0%。②YOLOv8-ECW的检测速度为238 F/s,较YOLOv8n模型提高了5 F/s。③与YOLOv6、YOLOv7等主流网络模型相比,YOLOv8-ECW模型的检测性能最佳且具有较好的鲁棒性。The existing models for detecting the behaviors of personnel in coal mine wells suffer from issues such as low accuracy and significant computational load.Therefore,a real-time detection algorithm for the behaviors of underground personnel based on YOLOv8-ECW is proposed.Based on YOLOv8n,the backbone network is enhanced by presenting the multi-scale convolution module EMSC.It is combined with the C2f convolution to design the C2f_EMSC module,effectively capturing the multi-scale features of the target and reducing the computational volume and parameter quantity of the model.The CGBlock downsampling module is introduced into the network to fuse the global context in-formation.The WIoU loss function is incorporated to enhance the positioning accuracy of the detection box and the convergence speed of the model.Experiments conducted on the self-established dataset for detecting the behaviors of personnel in coal mines reveal the following results:①Compared with the baseline YOLOv8n model,the average precision mean(mAP50)of the YOLOv8-ECW model for vari-ous targets is 92.4%,an increase of 2.1%;and the mAP50-95 is 75.4%,an increase of 4.0%.②The detection speed of the YOLOv8-ECW is 238 frames per second,which is 5 frames per second high-er than that of the YOLOv8n model.③Compared with the mainstream network models such as YOLOv6 and YOLOv7,the detection performance of the YOLOv8-ECW model is the best and it exhib-its better robustness.

关 键 词：煤矿井下 YOLOv8 行为检测 C2f_EMSC WIoU 特征融合 

分 类 号：TD76[矿业工程—矿井通风与安全] TP181[自动化与计算机技术—控制理论与控制工程]