一种用于石油化工厂环境下的仪表自动检测方法  

An Instrument Automatic Detection Method for Petrochemical Works

作　　者：李伟[1] 王飒 丁健刚 陈昊 肖力炀 LI Wei;WANG Sa;DING Jiangang;CHEN Hao;XIAO Liyang(School of Information Engineering,Chang’ an University,Xi’ an,Shaanxi 710064,China;Xi’ an Thermal Power Research Institute,Xi’ an,Shaanxi 710032,China)

机构地区：[1]长安大学信息工程学院,陕西西安710064 [2]西安热工研究院有限公司,陕西西安710032

出　　处：《西安石油大学学报（自然科学版）》2022年第2期102-109,共8页Journal of Xi’an Shiyou University（Natural Science Edition）

基　　金：国家自然科学基金资助项目(51978071);中央高校基本科研业务费专项资金资助项目(300102249301,300102249306)。

摘　　要：针对石油化工厂中人工抄表导致的低效、高误差和成本高等弊端,以及仪表图像拍摄条件场景复杂等问题,提出了一种基于改进Faster RCNN模型的工业数字表检测方法。首先,在特征提取网络阶段对卷积层低层和高层的网络特征进行融合,提高模型对细粒度细节和小目标的敏感度;其次,结合SENet网络结构,使模型关注不同通道的重要程度,通过分配不同的学习权重来强化对目标的关注度;最后,利用RPN网络进行最后处理,提取出数字表图像的边界框位置信息。结果表明,本文提出的模型检测精度为97.3%,相对于传统目标检测算法来说能更精准地识别出数字表。In order to solve the problems of low efficiency,high error and high cost caused by manual meter reading in petrochemical plants,as well as the complex conditions and scenes of instrument image shooting,an industrial digital instrument identification method based on the improved Faster RCNN network model is proposed.Firstly,in the feature extraction network stage,the network features of the lower and higher layers of the convolutional layer are fused to improve the sensitivity of the model to fine-grained details and small targets.Secondly,combined with the SENet network structure,the model pays attention to the importance of different channels,and strengthens the attention to the target by assigning different learning weights.Finally,the RPN network is used for final processing to extract the position information of the bounding of the digital instrument image.The result shows that the detection accuracy of the proposed model is 97.3%on the data set of digital instruments collected in this paper,which is better than that of traditional target detection algorithm.

关 键 词：Faster RCNN 特征融合 SENet 数字表检测 

分 类 号：TH851[机械工程—仪器科学与技术]