电网机巡图像分析框架与深度学习方法

doi:10.11959/j.issn.1000-0801.2020056

摘要/Abstract

摘要：

随着智能制造与物联网技术的发展，无人机被电网企业广泛地应用于输电线路巡视检查，同时也产生了大量的巡检图像数据亟需分析与处理。针对机巡图像分析中面临的多类多尺度、光照变化及遮挡等挑战，设计了一套从用户数据归集、分析与自动标注到用户评价反馈的U2U图像分析框架，在此基础上研究了Faster R-CNN和SSD两种深度学习方法在绝缘子、防震锤、均压环、屏蔽环等电力部件检测中的应用，提出了基于K-means++聚类分析的兴趣对象锚点信息框设定方法。实验结果表明，本文提出的方法有效地提高了深度学习方法对多尺度兴趣部件的适应能力与检测精度，为后续更大规模的机巡图像缺陷检测及机巡深化应用提供了有益的借鉴。

关键词: 智能电力系统, 深度学习, 多部件检测, 无人机

Abstract:

With the development of intelligent manufacturing and IoT,UAV has been widely utilized by power grid enterprise in patrolling transmission lines.At the same time,massive patrol image data need to be analyzed urgently.A U2U image analysis framework was designed from user data collection,automatic annotation and to user feedback.Two deep learning methods,which were faster R-CNN and SSD,were explored and applied in U2U to detect five types of power components,including insulators,dampers,grading rings and shielding rings.A refining method based on K-means++ was proposed for the parameters of anchor box.Experimental results demonstrate that the proposed methods can effectively improve the adaptability and detection accuracy of deep learning method for multiple-scale power components and provide a useful reference for subsequent defect detection and deep application of UAV patrolling.

Key words: intelligent power system, deep learning, multiple-component detection, UAV

中图分类号:

TP183

李远宁,宁柏锋,董召杰. 电网机巡图像分析框架与深度学习方法[J]. 电信科学, 2020, 36(8): 167-174.

Yuanning LI,Baifeng NING,Zhaojie DONG. Patrol image analysis framework and deep learning method for power grid[J]. Telecommunications Science, 2020, 36(8): 167-174.

图/表 11

图1

图2

图3

图4

表1

图5

图6

表2

表3

图7

表4

参考文献 23

[1]	王志宏, 杨震 . 人工智能技术研究及未来智能化信息服务体系的思考[J]. 电信科学, 2017,33(5): 1-11.
	WANG Z H , YANG Z . Research on artificial intelligence technology and the future intelligent information service architecture[J]. Telecommunications Science, 2017,33(5): 1-11.
[2]	窦兰秋 . 浅谈高压输电线路设计工作中应注意的要点[J]. 通讯世界, 2016(10): 203-204.
	DOU L Q . Talking about the key points in the design of high voltage transmission line[J]. Communication World, 2016(10): 203-204.
[3]	赵振兵, 孔英会, 戚银城 ,等. 面向智能输变电的图像处理技术[M]. 北京: 中国电力出版社, 2014.
	ZHAO Z B , KONG Y H , QI Y C ,et al. Image processing technology for intelligent power transmission and transformation[M]. Beijing: China Electric Power PressPress, 2014.
[4]	DALAL N , TRIGGS B . Histograms of oriented gradients for human detection[C]// International Conference on computer vision ＆ Pattern Recognition (CVPR’05). Piscataway:IEEE Press, 2005: 886-893.
[5]	LOWE D G . Distinctive image features from scale-invariant keypoints[J]. International Journal of Computer Vision, 2004,60(2): 91-110.
[6]	BAY H , ESS A , TUYTELAARS T ,et al. Speeded-up robust features (SURF)[J]. Computer Vision and Image Understanding, 2008,110(3): 346-359.
[7]	LIENHART R , MAYDT J . An extended set of haar-like features for rapid object detection[C]// International Conference on Image Processing. Piscataway:IEEE Press, 2002.
[8]	翟永杰, 王迪, 伍洋 ,等. 基于骨架提取的航拍绝缘子图像分步识别方法[J]. 华北电力大学学报（自然科学版）, 2015,42(3): 105-110.
	ZHAI Y J , WANG D , WU Y ,et al. Multistep recognition method of aerial insulator image based on skeleton extraction[J]. Journal of North China Electric Power University (Natural Science), 2015,42(3): 105-110.
[9]	班孝坤, 韩军, 陆冬明 ,等. 自然场景中基于局部轮廓特征的类圆对象识别方法[J]. 计算机应用, 2016,36(5): 1399-1403.
	BAN X K , HAN J , LU D M ,et al. Circle like object recognition method based on local contour feature in natural scene[J]. Computer Application, 2016,36(5): 1399-1403.
[10]	KRIZHEVSKY A , SUTSKEVER I , HINTON G E . Imagenet classification with deep convolutional neural networks[J]. Advances in Neural Information Processing Systems, 2012: 1097-1105.
[11]	GIRSHICK R , DONAHUE J , DARRELL T ,et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway:IEEE Press, 2014: 580-587.
[12]	REN S , HE K , GIRSHICK R ,et al. Faster R-CNN:towards real-time object detection with region proposal networks[C]// Neural Information Processing Systems. Piscataway:IEEE Press, 2015.
[13]	REDMON J , DIVVALA S , GIRSHICK R ,et al. You only look once:Unified,real-time object detection[C]// IEEE Conference on Computer Vision and Pattern Recognition.[S.l.:s.n. ], 2016: 779-788.
[14]	LIU W , ANGUELOV D , ERHAN D ,et al. SSD:single shot multibox detector[C]// European Conference on Computer Vision. Piscataway:IEEE Press, 2016: 21-37.
[15]	王万国, 田兵, 刘越 ,等. 基于 RCNN 的无人机巡检图像电力小部件识别研究[J]. 地球信息科学学报, 2017,19(2): 256-263.
	WANG W G , TIAN B , LIU Y ,et al. Research on recognition of power widget in UAV Inspection Image based on RCNN[J]. Journal of Earth Information Science, 2017,19(2): 256-263.
[16]	FELZENSZWALB P F , GIRSHICK R B , MCALLESTER D ,et al. Object detection with discriminatively trained part-based models[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2010,32(9): 1627-1645
[17]	HE K , ZHANG X , REN S ,et al. Spatial pyramid pooling in deep convolutional networks for visual recognition[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015,37(9): 1904-1916.
[18]	汤踊, 韩军, 魏文力 ,等. 深度学习在输电线路中部件识别与缺陷检测的研究[J]. 电子测量技术, 2018,41(6): 60-65.
	TANG Y , HAN J , WEI W L ,et al. Research on component identification and defect detection in transmission line based on deep learning[J]. Electronic Measurement Technology, 2018,41(6): 60-65.
[19]	ARTHUR D , VASSILVITSKII S . K-means++:the advantages of careful seeding[C]// The Eighteenth Annual ACM-SIAM Symposium on Discrete Algorithms. New York:ACM Press, 2007.
[20]	GIRSHICK R , . Fast R-CNN[C]// The IEEE International Conference on Computer Vision (ICCV).[S.l.:s.n]. 2015.
[21]	SIMONYAN K , ZISSERMAN A . Very deep convolutional networks for large-scale image recognition[J]. arXiv:1409.1556, 2014
[22]	HE K , ZHANG X , REN S ,et al. Deep residual learning for image recognition[J]. arXiv:1512.03385,
[23]	HOWARD A G , ZHU M , CHEN B ,et al. MobileNets:efficient convolutional neural networks for mobile vision applications[Z]. 2017

目标类别		玻璃绝缘子	复合绝缘子	均压环	防震锤	屏蔽环
目标数量/个	train	2 257	1 316	1 109	1 937	166
	test	1 050	508	454	762	77

对比项	玻璃绝缘子	复合绝缘子	均压环	防震锤	屏蔽环	mAP	f/s
Faster R-CNN	90.2%	90.2%	90.1%	89..4%	83.4%	88.7%	9
SSD-300	87.7%	81.9%	82.8%	81.4%	85.7%	83.9%	23
SSD-512	89.2%	81.5%	83.7%	82.9%	87.9%	85.0%	13

对比项	锚点1	锚点2	锚点3
玻璃绝缘子	90.2%	90.3%	90.8%
复合绝缘子	90.2%	89.6%	91.1%
均压环	89.2%	89.7%	90.3%
防震锤	89.4%	89.6%	90.4%
屏蔽环	73.2%	79.1%	88.7%
mAP	86.4%	87.6%	90.3%
f/s	9.9	7.8	8.1

对比项	vgg16	res50	res101	res152	mobile
玻璃绝缘子	90.2%	89.8%	90.1%	90.8%	86.1%
复合绝缘子	90.2%	90.0%	90.4%	91.1%	85.5%
均压环	90.1%	89.5%	89.4%	90.3%	88.2%
防震锤	89.4%	89.7%	90.1%	90.4%	87.7%
屏蔽环	83.8%	85.6%	86.1%	88.7%	59.4%
mAP	88.7%	88.9%	89.2%	90.3%	81.4%
f/s	9.4	10.0	8.9	8.1	10.2