摘要
近年来,由于油田作业现场作业人员安全意识薄弱,导致火灾时有发生。传统的人工监控方式存在诸多问题,为实现对油田作业现场作业人员吸烟行为的智能化检测,设计了一种采用YOLOv5深度学习算法的油田作业现场吸烟行为检测方法。通过监控摄像头及单反相机采集真实场景下中距离的吸烟数据集,使用LabelImg平台手动标注,经过YOLOv5的四种预训练框架训练,通过对比训练过程中产生的结果得到最佳训练模型。实验结果表明,该模型能够准确检测与追踪中距离吸烟行为,其检测准确率达89%,平均检测速度为17.2 ms,满足实时性要求。目前,该算法已经部署到油田作业现场的监控中,可以对油田作业现场吸烟行为进行有效检测。
In recent years,due to the weak safety awareness of oil field operators,fire often occurs.At present,there are many problems in the way of manual monitoring.In order to realize the intelligent detection of smoking behavior of oil field workers,a method of oil field workers’smoking behavior detection based on YOLOv5 deep learning algorithm is designed.Through the monitoring camera and SLR camera to collect the real scene of the middle distance smoking data set,using labelimg platform manual annotation,through YOLOv5 four kinds of pre training framework training,by comparing the results of the training process to get the best training model.The experimental results show that the model can accurately detect and track middle distance smoking behavior by using self-made data set and sufficient parameter optimization process.The detection accuracy is 89%,and the average detection speed is 17.2 ms,which meets the real-time requirements.At present,the algorithm has been deployed to the monitoring of oil field operation site,which can effectively detect the smoking behavior of oil field operation personnel.
作者
孙召龙
徐昕
朱云龙
田枫
SUN Zhaolong;XU Xin;ZHU Yunlong;TIAN Feng(Xianhe Oil Production Plant of Sinopec Shengli Oilfield,Sinopec Chemical Co.,Ltd,Dongying 257000,China;School of Computer&Information Technology of Northeast Petroleum University,Daqing 163318,China;Qingxin Oilfield Development Co.,Ltd,Daqing 163000,China)
出处
《系统仿真技术》
2021年第2期89-93,共5页
System Simulation Technology
基金
黑龙江省省属本科高校基本科研业务费项目(KYCXTD201903)
东北石油大学研究生教育创新工程项目(JYCX_11_2020)
黑龙江省省属本科高校基本科研业务费项目(2020YDL-11)
作者简介
孙召龙,男(1970-),山东寿光人,信息工程首席专家,主要研究方向为油气开发单位信息化建设应用。
