基于改进Canny算法与深度残差网络的车顶绝缘子憎水性识别方法  

Hydrophobicity Identification Method of Roof Insulator Based on Improved Canny Operator and Deep Residual Network

作　　者：周利军[1] 李杰 权圣威 张祥宇 张海彬 俞剑飞[2] ZHOU Lijun;LI Jie;QUAN Shengwei;ZHANG Xiangyu;ZHANG Haibin;YU Jianfei(School of Electrical Engineering,Southwest Jiaotong University,Chengdu 6l1756,China;Hangzhou Qiantang River Electric Group Co.,Ltd.,Hangzhou 311243,China)

机构地区：[1]西南交通大学电气工程学院,成都611756 [2]杭州钱江电气集团股份有限公司,杭州311243

出　　处：《高电压技术》2023年第10期4297-4306,共10页High Voltage Engineering

基　　金：四川省科技计划项目(2020JDTD0009)。

摘　　要：定期检测绝缘子憎水性等级(hydrophobicity class,HC)能及时避免污闪事故的发生,为了解决目前车顶绝缘子憎水性检测方法识别效率低、人工判断主观性强的问题,提出一种基于改进Canny算法与深度残差网络的车顶绝缘子憎水性识别方法。首先,通过喷水试验获取HC1—HC7的车顶绝缘子伞裙表面喷水图像;然后,采用改进Canny算法提取绝缘子伞裙表面的水珠边缘轮廓图,从而消除绝缘子伞裙色彩、光照阴影等因素对憎水性图像识别的影响;最后,采用ResNet101神经网络进行迁移学习,针对憎水性图像水珠形态多变,引入可形变卷积网络(deformable convolutional networks,DCN)加强了模型的鲁棒性。试验结果表明:利用改进Canny算法对绝缘子憎水性图像作预处理,结合DCN-ResNet101模型进行判断,测试准确率达到92.9%。Regular detection of insulator hydrophobicity class(HC)can timely avoid the occurrence of pollution flashover accidents.In order to solve the problems of low identification efficiency and strong subjectivity of artificial judgment of current roof insulator hydrophobicity detection methods,a hydrophobicity identification method for roof insulator based on improved Canny algorithm and deep residual network was proposed.Firstly,water spraying images of HC1 to HC7 roof insulator umbrella skirt surface were obtained by water spraying tests.Then,the improved Canny algorithm was used to extract the water drop edge contour diagram on the surface of insulator umbrella skirt,so as to eliminate the influence of color,light and shadow on hydrophobic image recognition.Finally,transfer learning was applied after using the ResNet101 neural network.Meanwhile,in accordance with variability in water droplet shapes in hydrophobicity images,deformable convolutional networks(DCN)were introduced to enhance the model’s robustness.The experimental results show that applying the improved Canny algorithm for preprocessing hydrophobicity images of insulators and combining it with the DCN-ResNet101 model for evaluation can realize the test accuracy of 92.9%.

关 键 词：车顶绝缘子 憎水性等级 改进Canny算法 残差网络 图像识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TM216[自动化与计算机技术—计算机科学与技术] U225.43[一般工业技术—材料科学与工程]