基于卷积神经网络的主变压器外观缺陷检测方法An Exterior Defects Detecting Method of Main Transformer Based on Convolutional Neural Networks
位一鸣,童力,罗麟,杨珊
WEI Yiming,TONG Li,LUO Ling,YANG Shan
摘要(Abstract)：
主变压器是变电站中最重要的电气设备之一,其运行状况直接影响所连区域电网的安全可靠运行。为全面提升主变压器外观缺陷检出效率,提出了一种基于CNN(卷积神经网络)算法的主变压器外观缺陷机器视觉识别检测方法。针对海量设备图像,该方法中的SSD算法模块能够准确提取目标设备(主变压器), CNN算法模块可对图像中所含缺陷信息进行解析。为了提升检测方法的准确性,针对检测算法负样本不足的问题,利用VGG-Net的图像迁移算法对主变压器缺陷样本进行扩充,以提升整个算法模型的泛化能力。最后,利用实际运维检修工作中采集整理的主变压器图像样本集进行算法验证,结果表明该方法能较准确地识别出变压器外观缺陷,具有较高的有效性和可行性。
Main transformer is one of the most important equipment in substation, and its operation condition is concerned with operation safety and reliability of the power grid. In order to improve the efficiency of exterior defect detection of the main transformer, the paper proposes computer vision technology for exterior defect detection based on CNN(convolutional neural networks). The SSD algorithm module can accurately extract object equipment(main transformer), and the CNN algorithm module can process defect information in the massive images. For accuracy improvement of the detection method, the image immigrating algorithm of VGG-Net is used to expand image defect samples to improve the generalization of the whole algorithm model.Finally, the main transformer image sample sets collected from operation and maintenance are used to verify the algorithm, and the result shows that the method can accurately detect exterior defects of transformer with its effectiveness and feasibility.
关键词(KeyWords)：
主变压器;外观缺陷;机器视觉;目标检测;风格迁移
main transformer;exterior defect;computer vision;object detection;style transformation
基金项目(Foundation): 国网浙江省电力有限公司群创项目(5211ZS180010)
作者(Author):
位一鸣,童力,罗麟,杨珊
WEI Yiming,TONG Li,LUO Ling,YANG Shan
DOI: 10.19585/j.zjdl.201904011
参考文献(References)：
- [1]江秀臣,盛戈皞.电力设备状态大数据分析的研究和应用[J].高电压技术,2018,44(4):1041-1050.
- [2]刘梓权,王慧芳,曹靖,等.基于卷积神经网络的电力设备缺陷文本分类模型研究[J].电网技术,2018,42(2):644-651.
- [3]刘梓权,王慧芳.基于知识图谱技术的电力设备缺陷记录检索方法[J].电力系统自动化,2018,42(14):158-164.
- [4]金立军,张达,段绍辉,等.基于红外与紫外图像信息融合的绝缘子污秽状态识别[J].电工技术学报,2014,29(8):309-318.
- [5]姚建刚,关石磊,陆佳政,等.相对温度分布特征与人工神经网络相结合的零值绝缘子识别方法[J].电网技术,2012,36(2):170-175.
- [6]姚建刚,付鹏,李唐兵,等.基于红外图像的绝缘子串自动提取和状态识别[J].湖南大学学报(自然科学版),2015,42(2):74-80.
- [7]李唐兵,付鹏,朱向前,等.绝缘子串红外图像中绝缘子盘面和钢帽区域的自动提取方法[J].红外技术,2014,36(8):644-648.
- [8]郭晓旭,刘志刚,张桂南,等.角点配准与图像差分的接触网绝缘子故障检测[J].电力系统及其自动化学报,2015,27(2):8-14.
- [9]张桂南,刘志刚.基于角点匹配与谱聚类的接触网绝缘子破损/夹杂异物故障检测[J].仪器仪表学报,2014,35(6):1370-1377.
- [10]李军锋,何双伯,冯伟夏,等.基于改进CNN的增强现实变压器图像识别技术[J].现代电子技术,2018,41(7):29-32.
- [11]辛明勇,肖小兵,刘斌,等.基于图像处理的变压器呼吸器自动检测[J].数学的实践与认识,2018,48(2):127-133.
- [12]GIRSHICK R,DONAHUE J,DARRELL T,et al.Rich feature hierarchies for accurate object detection and semantic segmentation[C]//ACM,2013.[S.l.]:ACM,2013:580-587.
- [13]GIRSHICK R.Fast RCNN[J].arXiv:1504.08083.
- [14]REN S Q,HE K M,GIRSHICK R,et al.Faster R-CNN:towards real-time object detection with region proposal networks[J].arXiv:1506.01497.
- [15]REDMON J,DIVVALA S,GIRSHICK R,et al.You only look once:unified,real-time object detection[J].arXiv:1506.02640.
- [16]LIU W,ANGUELOV D,ERHAN D,et al.SSD:Single shot MultiBox detector[J].arXiv:1512.02325.
- [17]L魪CUN Y,BOTTOU L,BENGIO Y,et al.Gradient-based learning applied to document recognition[J].Proceedings of the IEEE,1998,86(11):2278-2324.
- [18]IAN J G,JEAN P-A,MEHDI M,et al.Generative adversarial networks[J].arXiv:1406.2661.
- [19]DIEDERIK P K,MAX W.Auto-encoding variational bayes[J].arXiv:1312.6114.
- [20]LEON A G,ALEXANDER S E,MATTHIAS B.A neural algorithm of artistic Style[J].arXiv:1508.06576.
- [21]李航.统计学习方法[M].北京:清华大学出版社,2012.
- [22]欧家祥,史文彬,张俊玮,等.基于深度学习的高效电力部件识别[J].电力大数据,2018,21(9):1-8.
- [23]杨涛,黄军凯,许逵,等.基于深度学习的变压器故障诊断方法研究[J].电力大数据,2018,21(6):23-30.
- [24]朱沛恒.基于果蝇算法优化的概率神经网络在变压器故障诊断中的应用[J].电力大数据,2018,21(6):37-43.
- [25]黄春梅,马宏忠,张艳,等.基于小波包和模糊自适应共振神经网络的变压器绕组状态识别[J].广东电力,2017,30(7):89-95.