基于双视卷积神经网络的输电线路自动巡检  

Research on automatic inspection of transmission line based on cross-view convolution neural network

作　　者：戴永东 王茂飞 唐达獒[1] 毛锋 仲坚 倪莎 DAI Yongdong;WANG Maofei;TANG Daao;MAO Feng;ZHONG Jian;NI Sha(Taizhou Electric Power Company of State Grid,Taizhou 225300,China;Allcorehatress(Beijing)Technology Co.,Ltd.,Beijing 100089,China)

机构地区：[1]国网泰州供电公司,江苏泰州225300 [2]众芯汉创(北京)科技有限公司,北京100089

出　　处：《电力科学与技术学报》2021年第5期201-210,共10页Journal of Electric Power Science And Technology

基　　金：国网江苏省电力有限公司科技项目(5500-202018082A-0-0-00)。

摘　　要：卷积神经网络算法被广泛应用于输电线路自动巡检领域,但传统卷积神经网络电力缺陷识别模型的泛化能力较差。为此,提出一种融合双角度图像信息的卷积神经网络检测算法(CVR-RCNN),其利用2个视角可见光图像识别输电线路的常见缺陷。经实验测试,CVR-RCNN模型具有良好的鲁棒性,受试者工作特性(ROC)曲线下面积(AUC)值高达0.927,缺陷检测准确度较传统算法有显著提高。因此,CVR-RCNN能明显改善电力缺陷检测效果,可为无人机自动巡检输电线路提高准确稳定的算法架构。The convolutional neural network algorithm is widely applied in the automatic inspection of transmission lines.However,the generalization ability of traditional convolutional neural network power defect-recognition model is not ideal.Under the background,this paper proposes a cross-view relation region convolutional neural network(CVR-RCNN)detection algorithm that integrates dual-angle image information,which utilizes two-view visible light images to identify typical defects in transmission lines.The testing shows that the CVR-RCNN model has good robustness.The area under curve(AUC)value of the receiver operating characteristic(ROC)curve is as high as 0.927,and the defect detection accuracy is significantly improved compared with traditional algorithms.Therefore,CVR-RCNN can significantly improve power defect detection and improve the accuracy and stability of the algorithm architecture for the automatic inspection of transmission lines by UAVs.

关 键 词：无人机巡检 输电线路缺陷 深度学习 卷积神经网络 

分 类 号：TM712[电气工程—电力系统及自动化]