深度学习在绝缘子红外图像异常诊断的应用  

Application of Deep Learning in Abnormal Insulator Infrared Image Diagnosis

作　　者：范鹏 冯万兴[1,2] 周自强 赵淳[1,2] 周盛 姚翔宇[1,2] FAN Peng;FENG Wanxing;ZHOU Ziqiang;ZHAO Chun;ZHOU Sheng;YAO Xiangyu(NARI Group(State Grid Electric Power Research Institute)Co.,Ltd.,Nanjing 211106,China;Wuhan NARI Limited Liability Company,State Grid Electric Power Research Institute,Wuhan 430074,China)

机构地区：[1]南瑞集团(国网电力科学研究院)有限公司,江苏南京211106 [2]国网电力科学研究院武汉南瑞有限责任公司,湖北武汉430074

出　　处：《红外技术》2021年第1期51-55,共5页Infrared Technology

基　　金：国网电力科学研究院有限公司科技项目(524625190054)。

摘　　要：绝缘子的红外图像分析一般采用图像处理的方法,易受背景环境和数据量的影响,准确率和效率均较低,本文提出一种深度学习的异常诊断方法,基于改进的Faster R-CNN方法搭建检测网络,开展不同类型的绝缘子测试。研究结果表明:相对于神经网络(Back Propagation,BP)、Faster R-CNN方法,本文方法可高效地诊断出绝缘子的异常缺陷,平均检测精度达到90.2%;单I型和V型绝缘子的异常诊断准确率高于双I型绝缘子。研究结果可为输电线路绝缘子异常诊断提供一定的参考。Because of the effects of the background environment and data volume,the accuracy and efficiency of abnormal defects in traditional infrared images of insulators are generally low.In this study,a deep-learning anomaly diagnosis method is proposed.Based on the improved faster region-based convolutional neural network(R-CNN)method,a detection network is built to test different types of insulators.Results show that compared with the back propagation neural network and faster R-CNN methods,the proposed method can diagnose abnormal defects of insulators efficiently with a mean average precision of 90.2%.In addition,the diagnostic accuracy of single type I and type V insulators is higher than that of double type I insulators.The results can provide a reference for insulator defect identification in transmission lines.

关 键 词：绝缘子 异常诊断 深度学习 Faster R-CNN 平均检测精度 红外图谱 

分 类 号：TN219[电子电信—物理电子学]