Please wait a minute...
浙江大学学报(工学版)
浙江大学学报(工学版)  2021, Vol. 55 Issue (12): 2382-2389    DOI: 10.3785/j.issn.1008-973X.2021.12.019
电子、通信与自动控制技术     
基于多线型特征增强网络的架空输电线检测
陈雪云1(),夏瑾1,杜珂1,2
1. 广西大学 电气工程学院， 广西 南宁 530004
2. 广西电网有限责任公司南宁供电局， 广西 南宁 530000
Overhead transmission line detection based on multiple linear-feature enhanced detector
Xue-yun CHEN1(),Jin XIA1,Ke DU1,2
1. School of Electrical Engineering, Guangxi University, Nanning 530004, China
2. Nanning Power Supply Station ofGuangxi Power Grid Corporation, Nanning 530000, China
 全文: PDF(1094 KB)   HTML
摘要：

针对架空输电线可见光图像中环境背景复杂、电力线像素占比小，导致电力线检测精度低、断点率高的问题，提出具有强化线型特征提取和减少断点能力的多线型特征增强网络(MLED). 利用双路残差框架提取线型电力线目标的主干和边缘特征，通过多特征融合模块，在不同尺度的层次上实现主干、边缘和高层特征的深度整合，输出检测结果. 在多特征融合模块中嵌入残差、反卷积、多尺度结合等多路运算. 实验结果表明，MLED的检测能力较PSPNet、FCRN、UNet 有明显提高，多特征融合模块优于传统的残差连接块，可视化结果的 F 检验(F-Measure)、IoU 平均值(Mean IoU)分别为78.4%、77.8%，断点率为30.8%.
关键词： 架空输电线检测;  复杂背景;  多线型特征融合;  多尺度特征损失;  多线型特征增强网络(MLED)    
Abstract:

A multiple linear-feature enhanced detector (MLED) was proposed with the enhanced linear feature extraction from complicated background and the reduction of break-point rate capabilities, aiming at the problems of low power line detection accuracy and high break-point rate, which were caused by complex environmental background and low power line pixel ratio in the visible light image of overhead transmission lines. Firstly, several pairs of the trunk features and edge features were extracted from image by the dual-path residual framework. And then, through the multi-feature fusion module, the in-depth integration of backbone, edge and high-level features at different scale levels were realized. Eventually, the prediction result was obtained. Especially, multiple operations were embedded in the multi-feature fusion module including residual connection, up convolution and multi-scale coalescence. Experimental results showed that the ability of MLED was significantly improved while comparing with PSPNet, FCRN and UNet. The multi-feature fusion module was superior to traditional residual module. The F Measure and Mean IoU scores of the visualization results was 78.4% and 77.8% respectively, and the break-point rate was 30.8%.
Key words: overhead transmission lines detection    complicated background    multi-linear-feature fusion    multi-scale-feature loss    multiple linear-feature enhanced detector (MLED)
收稿日期: 2021-01-27 出版日期: 2021-12-31
CLC:  TP 391.4  
基金资助: 国家自然科学基金资助项目(62061002)
作者简介: 陈雪云(1969—)，男，副教授，博士，从事机器学习与模式识别研究. orcid.org/0000-0001-5276-1707. E-mail: cxy163167@163.com
服务　　
把本文推荐给朋友
加入引用管理器
E-mail Alert
作者相关文章　　
陈雪云
夏瑾
杜珂

引用本文:

陈雪云,夏瑾,杜珂. 基于多线型特征增强网络的架空输电线检测[J]. 浙江大学学报(工学版), 2021, 55(12): 2382-2389.

Xue-yun CHEN,Jin XIA,Ke DU. Overhead transmission line detection based on multiple linear-feature enhanced detector. Journal of ZheJiang University (Engineering Science), 2021, 55(12): 2382-2389.

链接本文:

https://www.zjujournals.com/eng/CN/10.3785/j.issn.1008-973X.2021.12.019    或    https://www.zjujournals.com/eng/CN/Y2021/V55/I12/2382

图 1  多线型特征增强网络结构图
图 2  3种多特征融合模块
图 3  MLED与主流网络的对比实验结果
图 4  对比实验精度−召回率曲线
网络 $ \mathrm{F}\mathrm{M} $ $ {M}_{\mathrm{I}\mathrm{o}\mathrm{U}} $ $ \mathrm{B}\mathrm{P}\mathrm{R} $ $ \mathrm{F}\mathrm{P}\mathrm{S} $
PSPNet 0.134 0.386 0.972 8.8
FCRN 0.703 0.651 0.442 17.2
UNet 0.734 0.679 0.399 11.9
MLED
0.784 0.778 0.308 13.55
表 1  各网络预测结果的评价指标与FPS
图 5  MLED的消融实验结果
图 6  消融实验精度−召回率曲线
网络 $ \mathrm{F}\mathrm{M} $ $ {M}_{\mathrm{I}\mathrm{o}\mathrm{U}} $ $ \mathrm{B}\mathrm{P}\mathrm{R} $
UNet 0.734 0.679 0.399
FU_Lf 0.754 0.728 0.343
MLED_Lf 0.751 0.742 0.317
MLED_A 0.762 0.758 0.313
MLED_B 0.776 0.771 0.311
MLED_C 0.784 0.778 0.308
表 2  消融实验各网络的评价指标
1 中国南方电网有限责任公司. 架空输电线路机巡技术[M]. 北京: 中国电力出版社, 2019: 1–3.
2 吕鹏程. 电力线视频监控方法研究与软件设计[D]. 杭州: 浙江大学, 2018: 1-4.
LV Peng-cheng. Video surveillance methods research and software design for power line [D]. Hangzhou: Zhejiang University, 2018: 1–4.
3 孙实超, 肖潇, 李栋, 等 利用参数关联性的输电线检测方法[J]. 广西大学学报:自然科学版, 2015, 40 (6): 1461- 1468
SUN Shi-chao, XIAO Xiao, LI Dong, et al Power transmission line detection utilizing the relationship between the parame-ters[J]. Journal of Guangxi University: Natural Science Edition, 2015, 40 (6): 1461- 1468
4 YANG G, ZHAO D, CHANG W, et al, Detection for power transmission line in low-resolution aerial image [C]// 2017 IEEE International Conference on Robotics and Biomimetics. Macau: IEEE, 2017: 1406–1411.
5 周封, 任贵新. 基于颜色空间变量的输电线图像分类及特征提取[J]. 电力系统保护与控制, 2018, 46(5): 89-98.
ZHOU Feng, REN Gui-xin. Image classification and feature extraction of transmission line based on color space variable [J]. Power System Protection and Control, 2018, 46(5): 89-98.
6 张晓. 无人机输电线巡检图像识别方法研究[D]. 哈尔滨: 哈尔滨理工大学, 2019: 1-27.
ZHANG Xiao. Research on image recognition method of UAV transmission line inspection [D]. Harbin: Harbin University of Science and Technology, 2019: 1-27.
7 ZHAO L, WANG X, YAO H, et al Power line extraction from aerial images using object-based Markov random field with anisotropic weighted penalty[J]. IEEE Access, 2019, 7: 125333- 125356
doi: 10.1109/ACCESS.2019.2939025
8 PAN C, CAO X B, WU D. Power line detection via back-ground noise removal [C]// 2016 IEEE Global Conference on Signal and Information Processing. Washington DC: IEEE, 2016: 871-875.
9 LEE S J, YUN J P, CHOI H, et al. Weakly supervised learning with convolutional neural networks for power line loca-lization [C]// 2017 IEEE Symposium Series on Computational Intelligence. Honolulu: IEEE, 2017: 1–8.
10 YETGIN Ö E, BENLIGIRAY B, GEREK Ö N Power line recognition from aerial images with deep learning[J]. IEEE Transactions on Aerospace and Electronic Systems, 2019, 55 (5): 2241- 2252
doi: 10.1109/TAES.2018.2883879
11 LI Y, PAN C, CAO X, et al Power line detection by pyramidal patch classification[J]. IEEE Transactions on Emerging Topics in Computational Intelligence, 2018, 3 (6): 416- 426
12 CHOI H, KOO G, KIM B J, et al Weakly supervised power line detection algorithm using a recursive noisy label update with refined broken line segments[J]. Expert Systems with Applications, 2020, 165: 113895
13 SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition [DB/OL]. (2014-12-04) [2020-01-10]. https://arxiv.org/pdf/1409.1556.pdf.
14 ZHAO H, SHI J, QI X, et al. Pyramid scene parsing network [C]// 2017 IEEE Conference on Computer Vision and Pattern Recognition. Honolulu: IEEE, 2017: 6230–6239.
15 CHEN L, PAPANDREOU G, KOKKINOS I, et al DeepLab: semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected CRFs[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2018, 40 (4): 834- 848
doi: 10.1109/TPAMI.2017.2699184
16 LAINA I, RUPPRECHT C, BELAGIANNIS V, et al. Deeper depth prediction with fully convolutional residual networks [C]// 2016 Fourth International Conference on 3D Vision. Palo Alto: IEEE, 2016 : 239–248.
17 HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition [C]// 2016 IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas: IEEE, 2016: 770–778.
18 RONNEBERGER O, FISCHER P, BROX T. U-Net: con-volutional networks for biomedical image segmentation [DB/OL]. (2015-05-18) [2020-01-10]. https://arxiv.org/pdf/1505.04597.pdf.
19 CHEN S, WANG B, TAN X, et al Embedding attention and residual network for accurate salient object detection[J]. IEEE Transactions on Cybernetics, 2020, 50 (5): 2050- 2062
doi: 10.1109/TCYB.2018.2879859
20 PANG Y, ZHAO X, ZHANG L, et al. Multi-scale interactive network for salient object detection [C]// 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle: IEEE, 2020: 9410–9419.
21 BOJARSKI M, CHOROMANSKA A, CHOROMANSKI K, et al. VisualBackProp: efficient visualization of CNNs for autonomous driving [C]// 2018 IEEE International Conference on Robotics and Automation. Brisbane: IEEE, 2018: 4701–4708.
[1] 赖小波,朱世强,方纯洁. 一种复杂背景图像三维重建算法及其医学应用[J]. J4, 2012, 46(11): 2061-2067.