陕西科技大学学报2018，Vol.36Issue(4)：153-157,5.
基于U-net网络的航拍绝缘子检测
Aerial insulator detection based on U-net network
摘要
Abstract
To make sure that the transmission circuits work safely and stably,it is necessary to inspect the circuits on a regular basis.As the core component of transmission line,the in-sulator detection plays an important role in the inspection of transmission line fault in power grid.For unmanned aerial vehicles with a large number of complex background and low reso-lution images of aerial insulators,the traditional manual feature extraction method and classi-cal convolutional neural network identification method are easy to lose location and details,and the location accuracy is relatively low.It is difficult to accurately identify the insulator images in complex background.Therefore,an aerial photo insulator detection method based on deep learning U-net network is proposed,automatic extraction of hierarchical feature,through the overlapping method combining the characteristics of shallow and high dimen-sional features,the high resolution feature map of the shallow layer is used for pixel location,and the deep high dimensional feature graph is used to classify the pixels.It avoids the loss of detailed information such as the location of the target,improves the positioning accuracy,and effectively detects the insulator under complicated background.The accuracy rate reaches 88.9％,which is an effective insulator detection method.关键词
U-net网络/航拍图像/绝缘子/检测Key words
U-net network/aerial images/insulator/detect分类
信息技术与安全科学引用本文复制引用
陈景文,周鑫,张蓉,张东..基于U-net网络的航拍绝缘子检测[J].陕西科技大学学报,2018,36(4):153-157,5.基金项目
西安市科技局技计划项目(2017068CG/RC031(SXKD001)) （2017068CG/RC031(SXKD001）
