基于Mask-RCNN算法的复合绝缘子串红外图像分割与温度读取  

Infrared Image Segmentation and Temperature Reading of Composite Insulator Strings Based on Mask-RCNN Algorithm

作　　者：黄新宇 张洋 王黎明[1] 梅红伟[1] 张中浩[1] 文路 HUANG Xinyu;ZHANG Yang;WANG Liming;MEI Hongwei;ZHANG Zhonghao;WEN Lu(Graduate School at Shenzhen,Tsinghua University,Guangdong Shenzhen 518055,China;Shandong Electric Power Research Institute,Jin'an 250000,China)

机构地区：[1]清华大学深圳研究生院,广东深圳518055 [2]山东电力科学研究院,济南250000

出　　处：《高压电器》2021年第9期87-94,共8页High Voltage Apparatus

基　　金：国家重点研发计划资助项目(2017YFB0902702)。

摘　　要：红外成像可以快速发现复合绝缘子异常温升,从而实现缺陷检测。为了解决复合绝缘子红外检测中人工诊断时间成本高、效率低、背景干扰多的问题,文中提出了基于Mask-RCNN的温升绝缘子自动检测方法。首先使用Mask-RCNN对红外图像数据进行学习训练,进而对其中的绝缘子目标进行识别检测和图像分割。在智能分割之后,通过识别图中的温度范围数据并将绝缘子区域图像灰度与之比对,得到了绝缘子的温度信息,用于绝缘子异常发热故障的诊断。实验研究表明:在对南方某市绝缘子红外测温图像的自动识别中,Mask-RCNN有着优异的表现,能够运用于在巡线工作中,可以有效减轻人工处理数据的压力,对于提升推进智能巡线的智能化水平、保障输电线路安全运行有着重要的意义。Infrared images can find the abnormal temperature rise of the composite insulator quickly,thus achieving defect detection.For solving such problems in the infrared detection of the composite insulator as high cost,low efficiency and many backgrounds interference in the artificial diagnosis,an automatic detection method for temperature rise of insulator based on mask regional convolutional neural network(Mask-RCNN)is proposed in this paper.Firstly,the infrared image data is learnt and trained by the use of Mask-RCNN model.And then,identification detection and image segmentation for the insulator target are performed.After the intelligent segmentation,the temperature information of the insulator is obtained by way of identifying temperature range data in the image and comparing it with the area image gray level,which is used to diagnose the abnormal heating of the insulator.The experimental study shows that in the automatic identification of the infrared temperature image for insulator in one city in the south part of China,the Mask-RCNN shows superior performance.It can be used in line patrol work,can reduce effectively the pressure of artificial processing data and has important significance in improving the intelligent level of intelligent lien patrol and assuring safe operation of transmission line.

关 键 词：复合绝缘子 异常发热 深度学习 红外成像 图像识别 

分 类 号：TM216[一般工业技术—材料科学与工程] TP391.41[电气工程—电工理论与新技术] TN219[自动化与计算机技术—计算机应用技术]