基于二维图像识别油田违章行为监测技术研究  

Research on Oil Field Violation Monitoring Technology Based on Two-dimensional Image Recognition

作　　者：王美[1,2] 杨欣欣[1] 李慧颖 刘泱[1] 张俊华[3] WANG Mei;YANG Xinxin;LI Huiying;LIU Yang;ZHANG Junhua(Shengli Oilfield Technology Inspection Center,Dongying Shandong 257000,China;College of Computer and Technology,China University of Petroleum,Qingdao Shandong 266580,China;Shengli Oil Production Plant,Dongying Shandong 257000,China)

机构地区：[1]胜利油田分公司技术检测中心,山东东营257000 [2]中国石油大学(华东)计算机科学与技术学院,山东青岛266580 [3]胜利采油厂,山东东营257000

出　　处：《安全》2023年第5期55-60,共6页Safety & Security

基　　金：中国石油化工股份有限公司科研项目(320044)。

摘　　要：为了在油田施工作业过程中及时准确地发现施工人员的违章行为,保证油田的安全生产,提出一种基于二维图像识别油田违章行为的监测技术。该技术采用一种目标检测与关键点定位的多任务联合学习算法,提高传统算法的准确率;采用目标属性二次分析的方法识别目标精细情况;采用目标追踪与时序人体行为分析方法实时跟踪人的行为动作。并通过不同场景的2367例样本的应用显示,识别准确率80%以上,召回率75%以上,表明该技术对一般违章行为识别的准确率较高。In order to timely and accurately detect the violations of construction personnel during oilfield construction operations and ensure the safe production of oilfields,a monitoring technology based on two-dimensional image recognition of oilfield violations is proposed.The technology adopts a multi-task joint learning algorithm of target detection and key point localization to improve the accuracy of the traditional algorithm;A secondary analysis of target properties was used to identify target fine conditions;Target tracking and temporal human behavior analysis are used to track human behavior in real time.Through the application of 2367 samples in different scenarios,the identification accuracy is more than 80%and the recall rate is more than 75%,indicating that the technology has a high accuracy rate for general violation recognition.

关 键 词：违章监测 图像识别 目标检测 时序分析 

分 类 号：X924[环境科学与工程—安全科学]