基于DeepLab V3+的无人机航拍绝缘子图像自动化分割方法  

Automatic Segmentation Method for UAV Aerial Images of Insulators Based on DeepLab V3+

作　　者：徐越 段鑫 刘子祎 袁晨 刘君[2] XU Yue;DUAN Xin;LIU Ziyi;YUAN Chen;LIU Jun(State Grid Jiangxi Electric Power Company,Nanchang Power Supply Branch,Nanchang 330000,China;Nanchang Hangkong University,College of Information Engineering,Nanchang 330063,China)

机构地区：[1]国家电网江西省电力有限公司南昌供电分公司,南昌330000 [2]南昌航空大学信息工程学院,南昌330063

出　　处：《电瓷避雷器》2023年第2期180-188,共9页Insulators and Surge Arresters

基　　金：国家自然科学基金(编号:61963030);江西省电力有限公司科技项目(编号:5218A020003A)。

摘　　要：绝缘子是输电线路的重要组成部分。为了实现输电线路巡检的智能化,对绝缘子区域的自动化识别是基础性工作之一。笔者提出了一种基于DeepLab V3+的无人机航拍绝缘子图像自动化分割方法,同时将DeepLab V3+的分割性能与现有的主流卷积神经网络U-Net、Residual U-Net、Dilated U-Net进行了比较。结果显示DeepLab V3+方法不但在DICE指标上得到了93.28%的分割精度,在IOU指标、敏感性和Jaccard定量分析中也都表现出最好的性能。实验证明该方法为自动化分割绝缘子区域,提高输电线路巡检的效率和准确性提供了一种潜在的工具。Insulators are the important part of the grid.In order to realize the intelligent inspection of the grid,the automatic segmentation of the insulator area is one of the basic requirements.By using the Dee-pLab V3+network,theauthor developed an automatic segmentation method for UAV aerial insulator ima-ges,and the segmentation performance of the proposed method was compared with the three existing pop-ular convolutional neural networks U-Net,Residual U-Net and Dilated U-Net.The results showed that the DeepLab V3+gained 93.28%segmentation accuracy in DICE evaluation.Meanwhile,it also a-chieved the best performance in IOU,sensitivity,and Jaccard quantitative analysis than the other three methods.It can be concluded that the DeepLab V3+is a potential tool for automatically segmenting the insulator area and improving the efficiency and accuracy of grid inspection.

关 键 词：电网巡检 绝缘子 卷积神经网络 图像分割 

分 类 号：TM75[电气工程—电力系统及自动化] TP391.41[自动化与计算机技术—计算机应用技术]