电工技术学报  2021, Vol. 36 Issue (7): 1407-1415    DOI: 10.19595/j.cnki.1000-6753.tces.L90049
“电力装备智能感知与智能终端”专题(特约主编:成永红教授) |
基于层次多任务深度学习的绝缘子自爆缺陷检测
徐建军, 黄立达, 闫丽梅, 伊娜
东北石油大学电气信息工程学院 大庆 163318
Insulator Self-Explosion Defect Detection Based on Hierarchical Multi-Task Deep Learning
Xu Jianjun, Huang Lida, Yan Limei, Yi Na
School of Electrical Engineering and Information Northeast Petroleum University Daqing 163318 China
全文: PDF (1946 KB)   HTML
输出: BibTeX | EndNote (RIS)      
摘要 绝缘子是电力线路中重要且使用广泛的器件,随着近年来无人机巡线的迅速普及,从航拍图像中检测绝缘子自爆缺陷成为热点问题。在航拍图像中,自爆绝缘子与正常绝缘子的区分难度相对更大,该文提出一种基于层次多任务深度学习的绝缘子自爆缺陷检测模型,使用专用的卷积神经网络区分自爆绝缘子和正常绝缘子,并结合多任务学习和特征融合方法提高分类准确率。同时,针对缺乏自爆类数据的问题,提出制作合成图像的数据增强方法。实验结果表明,添加合成图像能有效提高自爆类召回率;层次多任务学习模型与平面分类模型及普通层次模型相比具有更强的分类能力。
服务
把本文推荐给朋友
加入我的书架
加入引用管理器
E-mail Alert
RSS
作者相关文章
徐建军
黄立达
闫丽梅
伊娜
关键词 绝缘子多任务学习缺陷检测深度学习层次分类    
Abstract:Insulators are important and widely used devices in power lines. With the rapid popularization of unmanned aerial vehicles in recent years, detecting insulator self-explosion defect from aerial images has become a hot issue. In aerial images, distinguishing self-explosion insulators from normal insulators is more difficult than from other contents. The paper proposed an insulator self-explosion defect detection model based on hierarchical multi-task deep learning, using a dedicated convolutional neural network to distinguish self-explosion insulators from normal insulators. To improve the classification accuracy, multi-task learning and feature fusion method were used in the hierarchical model. Furthermore, in view of the lack of self-explosion insulator data, the paper proposed to use synthetic pictures for data augmentation. The experimental results show that adding synthetic pictures can effectively improve the recall rate of self-explosion class, and the hierarchical multi-task learning model achieves stronger classification performance than the flat classification model and the ordinary hierarchical model.
Key wordsInsulator    multi-task learning    defect detection    deep learning    hierarchical classification   
收稿日期: 2020-06-10     
PACS: TP216  
基金资助:国家自然科学基金(51774088)和黑龙江省自然科学基金(LH2019E016)资助项目
通讯作者: 闫丽梅 女,1971年生,教授,博士,研究方向为电力系统安全稳定。E-mail:yanlimeidaqing@163.com   
作者简介: 徐建军 男,1971年生,教授,博士生导师,研究方向为电力系统安全稳定。E-mail:xujj@nepu. edu. cn
引用本文:   
徐建军, 黄立达, 闫丽梅, 伊娜. 基于层次多任务深度学习的绝缘子自爆缺陷检测[J]. 电工技术学报, 2021, 36(7): 1407-1415. Xu Jianjun, Huang Lida, Yan Limei, Yi Na. Insulator Self-Explosion Defect Detection Based on Hierarchical Multi-Task Deep Learning. Transactions of China Electrotechnical Society, 2021, 36(7): 1407-1415.
链接本文:  
https://dgjsxb.ces-transaction.com/CN/10.19595/j.cnki.1000-6753.tces.L90049          https://dgjsxb.ces-transaction.com/CN/Y2021/V36/I7/1407