基于深度学习的指针仪表识别系统  

Deep Learning-based Pointer Instrument Recognition System

作　　者：黄杨乐天 刘宜胜[1] 陈锬 刘丹丹[1] HUANG Yangletian;LIU Yisheng;CHEN Tan;LIU Dandan(School of Mechanical Engineering,Zhejiang Sci-Tech University,Hangzhou 310018,China)

机构地区：[1]浙江理工大学机械工程学院,杭州310018

出　　处：《自动化与仪表》2023年第11期100-104,108,共6页Automation & Instrumentation

摘　　要：指针仪表识别并读数是机房巡检的重要任务,该文基于图像处理和深度学习算法,设计了一种指针仪表识别系统。首先,通过巡检机器人采集机房巡检的数据集;其次,对Faster R-CNN模型进行池化策略改进,并使用更深的ResNet152残差网络,实现仪表区域的识别和定位;最后,使用边缘检测结合霍夫变换对仪表进行矫正和读数。实验结果表明,改进后的Faster R-CNN模型准确率较高,平均精确度提高了6.2%,也优于当前主流的网络模型。该模型测试的读数误差在2%以内,企业测试人员实际运行后的误差在4%内,可以满足企业巡检需求,具有较好的应用价值。Pointer meter recognition and reading is an important task for machine room inspection.In this paper,a pointer meter recognition system is designed based on image processing and deep learning algorithms.Firstly,the machine room inspection data set is collected by an inspection robot.Secondly,the Faster R-CNN model is improved with pooling strategy and the ResNet152 deeper residual network is used to realize the recognition and localization of the meter region.Finally,the meters are corrected and read using edge detection combined with Hough transform.The experimental results show that the improved Faster R-CNN model is more accurate,the mAP is improved by 6.2%,which is also better than the current mainstream network models.The reading error of the model test is within 2%,and the error after the actual operation of enterprise testers is within 4%,which can meet the needs of enterprise inspection,and has a good application value.

关 键 词：指针仪表 霍夫变换 边缘检测 深度学习 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]