基于Python语言与图像识别算法的油气田仪表识别研究  

Research on Oil and Gas Field Instrument Identification Based on Python Language and Image Recognition Algorithm

作　　者：闫新 缑雅洁 郝学磊 苗润金 康军虎 刘蕊祎 Yan Xin;Gou Yajie;Hao Xuelei;Miao Runjin;Kang Junhu;Liu Ruiyi(No.1 Gas Production Plant of PetroChina Changqing Oilfield Company,Yulin,Shaanxi 718500,China;Technical Monitoring Center of PetroChina Changqing Oilfield Company,Xi’an 710018,China)

机构地区：[1]长庆油田分公司第一采气厂,陕西榆林718500 [2]长庆油田分公司技术检测中心,西安710018

出　　处：《机电工程技术》2024年第8期201-204,共4页Mechanical & Electrical Engineering Technology

摘　　要：指针仪表识别并读数是油气生产领域日常巡检的重要任务,但基于深度学习的指针式仪表读数识别需要针对每个项目专门进行模型训练,不具有普遍适应性,工程实施成本高。基于图像处理和深度学习算法,采用Python语言设计了一种适用于油气生产领域的指针式仪表识别系统。首先,通过巡检机器人采集生产现场巡检的数据集;其次,对巡检机器人原搭载的模型采用Python语言进行改进,并使用均值滤波除去图像噪点、Hough检测对图像特征的捕捉与分析来实现仪表的自动检测与读数;最后,通过Python代码调用tkinter实现GUI人机交互界面,达到批量识别现场仪表数据的目的。现场试验结果表明:改进后的模型准确率较高,解决了原模型识别“死点”问题,平均精度达到95%,该模型现场实测运行的误差变异系数控制在0.05以内,可以满足现场巡检需求,具有较好的油气田领域仪表识别应用价值。The identification and reading of pointer instrument are important tasks in daily inspection in oil and gas production field.The reading recognition of pointer instrument based on deep learning requires special model training for each project,which is not universally adaptable and has high engineering implementation cost.Based on image processing and deep learning algorithm,a pointer instrument identification system suitable for oil and gas production field is designed by using Python language.Firstly,the data set of production site inspection is collected by inspection robot.Secondly,Python language is used to improve the original model of the inspection robot,and mean filtering is used to remove image noise,and Hough detection is used to capture and analyze image features to realize automatic detection and reading of the instrument.Finally,the GUI man-machine interface is realized by calling tkinter through Python code,and the purpose of batch identification of field instrument data is achieved.The field test results show that the improved model has high accuracy,solves the problem of"dead point"identification of the original model,and the average accuracy reaches 95%.The coefficient of error variation of the model is controlled within 0.05,which can meet the requirements of field inspection and has good application value of instrument identification in the field of oil and gas field.

关 键 词：仪表识别 均值滤波 Hough检测 图像处理 

分 类 号：TE967[石油与天然气工程—石油机械设备] TP391.4[自动化与计算机技术—计算机应用技术]