基于CascadeR-CNN算法的输电线路小目标缺陷检测方法  

Transmission Line Small Target Defect Detection Method Based on Cascade R-CNN Algorithm

作　　者：吴军 白梁军 董晓虎 潘尚智 金哲 范亮 程绳 WU Jun;BAI Liangjun;DONG Xiaohu;PAN Shangzhi;JIN Zhe;FAN Liang;CHENG Sheng(Maintenance Company of State Grid Hubei Electric Power Co.,Ltd.,Wuhan 430050,Hubei,China;School of Electrical Engineering and Automation,Wuhan University,Wuhan 430072,Hubei,China;Guangzhou China Sciences Intelligent Inspection Technology Co.,Ltd.,Guangzhou 510000,Guangdong,China)

机构地区：[1]国网湖北省电力有限公司检修公司,湖北武汉430050 [2]武汉大学电气与自动化学院,湖北武汉430072 [3]广州中科智巡科技有限公司,广东广州510000

出　　处：《电网与清洁能源》2022年第4期19-27,36,共10页Power System and Clean Energy

基　　金：国家电网有限公司科技项目(521520200018)。

摘　　要：输电线路无人机航拍图像缺陷识别是维护线路安全运行的重要巡检手段,但目前的识别算法对于销钉、螺母等小目标缺陷存在识别精确度低、易漏判等问题。将Cascade RCNN算法应用于输电线路缺陷检测中,利用ResNet101网络进行特性提取,增强的网络的特征提取能力,并利用多层级联检测器对输电线路小目标进行判别和分类。基于无人机航拍图像数据集进行实验,实验结果表明,相比于Yolov3检测器和Lighthead R-CNN检测器,Cascade R-CNN算法提高了小目标缺陷检测中的召回率和精确度。The defect identification of UAV aerial images of transmission lines is an important means of inspection to maintain the safe operation of transmission lines.However,the current identification algorithm has problems of missed detection and false detection in small parts such as pins,nuts and other parts.This paper applies the Cascade R-CNN algorithm to transmission line defect detection,uses the ResNet101 network for feature extraction,and enhances the feature extraction capability of the network,and uses a multi-layer cascade detector to distinguish and classify pin targets.Experiments are conducted based on the UAV aerial image data set,and the experimental results show that compared with the Yolov3 detector and the Lighthead R-CNN detector,the Cascade R-CNN algorithm has improved the recall rate and accuracy of the defect detection of small parts.

关 键 词：Cascade R-CNN网络 输电线路 缺陷检测 卷积神经网络 

分 类 号：TM752.5[电气工程—电力系统及自动化] TP18[自动化与计算机技术—控制理论与控制工程]