融合注意力机制的YOLOv5轻量化煤矿井下人员检测算法  

YOLOv5 Lightweight Coal Mine Underground Personnel Detection Algorithm Base on Attention Mechanism

作　　者：芦碧波[1] 周允 李小军[2] 谷亚楠 王文轲 LU Bibo;ZHOU Yun;LI Xiaojun;GU Yanan;WANG Wenke(School of Computer Science and Technology,Henan University of Technology,Jiaozuo 454000,China;School of Energy Science and Engineering,Henan University of Technology,Jiaozuo 454000,China)

机构地区：[1]河南理工大学计算机科学与技术学院,河南焦作454000 [2]河南理工大学能源科学与工程学院,河南焦作454000

出　　处：《煤炭技术》2023年第10期200-203,共4页Coal Technology

摘　　要：针对现有煤矿井下环境复杂、监控视频模糊、人员安全状况检测困难和检测算法参数量大、运行速度慢问题,提出了一种轻量化的YOLOv5煤矿井下检测算法。首先,将轻量化ShuffleNetv2作为主干网络,减少了模型计算参数量,降低了网络的复杂度;接着引入一种改进后的注意力机制F-CBAM模块,使通道注意力和空间注意力直接学习输入的特征图,增加对目标物体的关注度。模型最终的检测精度为98.6%,高于经典网络性能,降低了对硬件的需求,提高了井下人员识别的实时性。Aiming at the complex underground environment of existing coal mines, blurred surveillance video, difficulty in detecting personnel safety conditions, detection algorithm with a large number of parameters and slow running speed, a lightweight YOLOv5 coal mine underground detection algorithm is proposed. First, the lightweight ShuffleNetv2 is used as the backbone network, which reduces the amount of model calculation parameters and reduces the complexity of the network;then an improved attention mechanism F-CBAM module is introduced, which enables channel attention and spatial attention to directly learn the input feature map and increase the attention to the target object. The final detection accuracy of the model is 98.6%, which is higher than the performance of the classical network,which reduces the demand for hardware and improves the real-time performance of underground personnel identification.

关 键 词：煤矿井下 轻量化 YOLOv5 ShuffleNetv2 注意力机制 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] TD76[自动化与计算机技术—计算机科学与技术]