基于深度学习和图像识别的数字式仪表读值研究

Research on Reading Value of Digital Instrument Based on Depth Learning and Image Recognition

随着自动化程度的不断提升,变电所在巡视中使用的机器人越来越多,方便了数码仪器的图像获取。但是在采集到的大量变电所巡视的图像中,指针式仪表的读数和变压器零件的辨识精度仍较低,已成了一个研究的难点。近年来,基于海量信息的深度神经网络技术在图像辨识中的应用得到了极大的发展。文章结合电力设备图像的特点,基于深度学习算法重点研究了变电站数字式仪表读数的自动识别和变压器小部件的自动识别,提出了使用卷积神经网络YOLOv5模型识别图表的方法。该方法适用于对实时性、准确性要求高的场景,使用可扩展置标语言(EXtensible Markup Language,XML)解析手段获取数字式仪表位置及大小信息,可自动化标注数字式仪表。

With the continuous improvement of automation,a large number of robots are used in the inspection of substations,which facilitates the image acquisition of digital instruments.However,in the collected images of a large number of substation patrols,the reading of pointer instruments and the identification accuracy of transformer parts have not been well solved,which has become a research difficulty.In recent years,the application of depth neural network technology based on massive information in image recognition has been greatly developed.Based on the characteristics of power equipment images,this paper focuses on the automatic recognition of digital instrument readings and transformer small parts based on deep learning algorithm.A graph recognition method using the convolution neural network YOLOv5 model is proposed.This method is applicable to scenes with high requirements for real-time and accuracy.Use Extensible Markup Language(XML)parsing to obtain the position and size information of the digital instrument,and automatically label the digital instrument.