基于深度学习的配电网线路设备缺陷智能检测Intelligent Defect Detection of Distribution Line Equipment Based on Deep Learning
李晨曦,邵蒙悦,冯杰
LI Chenxi,SHAO Mengyue,FENG Jie
摘要(Abstract)：
无人机配电线路巡检已经广泛开展应用,线路设备缺陷人工识别过程复杂、工作量较大等问题仍然存在。为了提高作业效率,根据无人机发展现状,对现有数据进行深入挖掘,利用标记系统对数据进行处理,再利用深度学习算法实现配电网无人机自动巡检、缺陷自动研判。该算法是基于残差双尺度检测器的巡检目标智能检测,优势在于可识别两种规格尺寸的目标对象,相比于传统的双阶段目标检测方案,其运行速度更快,更适合在终端资源受限设备中运行。目前该算法已达到95%的准确率。
Unmanned aerial vehicle(UAV) inspection on power distribution line has been widely applied.However, some problems remain unresolved-complex process and heavy workload in manual identification of line equipment defects. To improve the operation efficiency, this paper carries out in-depth mining on the existing data and processes the data with the marking system; then it implements automatic inspection of the UAV on the distribution network and automatic defect judgment in distribution network with the help of deep learning. As an intelligent inspection target detection based on residual dual scale detector, the method is advantageous in recognizing target of two sizes with higher operation speed and is more suitable for operation in the terminal resource-constrained devices. At present, this accuracy of the algorithm reaches up to 95%.
关键词(KeyWords)：
配电网;无人机;智能巡检;边缘计算
distribution network;UAV;intelligent inspection;edge computing
基金项目(Foundation):
作者(Author):
李晨曦,邵蒙悦,冯杰
LI Chenxi,SHAO Mengyue,FENG Jie
DOI: 10.19585/j.zjdl.202103015
参考文献(References)：
- [1]汤明文,戴礼豪,林朝辉,等.无人机在电力线路巡视中的应用[J].中国电力,2013(3):35-38.
- [2]张征凯,齐浪.基于Faster R-CNN的工件螺纹孔目标检测[J/OL].Journalof Measurement Science and Instrumen tation,2020(11):1-8[2020-12-02].http://kns.cnki.net/kcms/detail/14.1357.TH.20201118.1618.007.html.
- [3]陈为,梁晨红.基于改进SSD的航空发动机目标缺陷检测[J/OL].控制工程,2020(11):1-6[2020-12-02].https://doi.org/10.14107/j.cnki.kzgc.CPCC2019-063.
- [4]张勤,王家辉,李彬.基于YOLOv3目标检测的秧苗列中心线提取方法[J].农业机械学报,2020,51(8):34-43.
- [5]YEONGMIN Y,JIN J U,HA H Y,et al.Data augmentation-based prediction of system level performance under model and parameter uncertainties:role of designable generative adversarial networks(DGAN)[J/OL].Reliability Engineering and System Safety,2021,2(206):107316[2020-12-20].http://www.sciercediraect.com/science/article/abs/poo/S095/832020308/03.DOI:10.1016/j.ress.2020.107316.
- [6]高瑞,郝乐,刘宝,等.基于改进ResNet网络的井下钻杆计数方法[J].工矿自动化,2020,46(10):32-37.
- [7]占浩,朱振才,张永合,等.基于残差网络的图像序列闭环检测方法[J/OL].激光与光电子学进展,2020(9):1-13.[2020-11-27].http://kns.cnki.net/kcms/detail/31.1690.TN.20200916.1254.008.html.
- [8]孟琭,徐磊,郭嘉阳.一种基于改进的MobileNetV2网络语义分割算法[J].电子学报,2020,48(9):1769-1776.
- [9]周博文.基于MobileNetV2的目标检测算法研究与应用[D].武汉：湖北工业大学,2020.
- [10]李敏,李红娇,陈杰.差分隐私保护下的Adam优化算法研究[J].计算机应用和软件,2020,37(6):253-258.
- [11]顾超越,李喆,史晋涛,等,基于改进Faster-RCNN的无人机巡检架空线路销钉缺陷检测[J].高电压技术,2020,46(9):3089-3096.
- [12]吴锡,王梓屹,宋柯,等.基于Faster RCNN检测器的输电线路无人机自主巡检系统[J].电力信息与通信技术,2020,18(9):8-15.
- [13]STEVENS E,ANTIGA L.Deep learning with PyTorch[M].Greanwich:Manning Publications,2020.
- [14]马原东,罗子江,倪照风,等.改进SSD算法的多目标检测[J/OL].计算机工程与应用,2020(8):1-19[2020-11-27].http://kns.cnki.net/kcms/detail/11.2127.TP.20200831.1929.024.html.
- [15]孙帅,陈辉.基于增强YOLOv3算法的建筑物窗户检测[C]//中国仿真学会.2020中国仿真大会论文集,2020:7.