基于生成对抗数据增殖的输电设备可视缺陷检测技术研究  

Research on Visual Defect Detection Technology for Transmission Equipment Based on Generative Adversarial Data Proliferation

作　　者：蔡常雨 莫文昊 王万国 张英强 郭鹏天 CAI Changyu;MO Wenhao;WANG Wanguo;ZHANG Yingqiang;GUO Pengtian(Artificial Intelligence Application Department,China Electric Power Research Institute,Haidian District,Beijing 100192,China;State Grid Shandong Electric Power Company,Jinan 250001,Shandong Province,China)

机构地区：[1]中国电力科学研究院有限公司人工智能应用研究所,北京市海淀区100192 [2]国网山东省电力公司,山东省济南市250001

出　　处：《电力信息与通信技术》2023年第9期38-43,共6页Electric Power Information and Communication Technology

基　　金：国家电网有限公司总部科技项目资助“电力云边端协同人工智能模型共享关键技术研究”(5700-202116265A-0-0-00)。

摘　　要：输电设备巡检影像的小样本特征愈发凸显,为输电线路智能巡检带来了新的挑战。样本分布不平衡导致大量输电图像样本资源无法充分利用,为此提出基于生成对抗数据增殖的输电设备可视缺陷检测技术。通过引入循环生成对抗网络对输电设备巡检影像样本进行增殖,而后利用扩增后的样本完成输电设备可视缺陷检测模型训练,实现可视缺陷检测效果的优化提升。通过仿真实验验证了循环生成对抗数据增殖对输电设备可视缺陷检测有效性与稳定性的提升效果,为输电设备可视缺陷检测提供了新的视角与思路。The small sample features of transmission equipment inspection images are becoming more and more prominent,which brings new challenges to the intelligent inspection of transmission lines.Unbalanced sample distribution leads to the inability to fully utilize a large number of power transmission image sample resources.For this reason,a visual defect detection technology for transmission equipment based on generative adversarial data proliferation is proposed.Through the introduction of the cyclic generative adversarial network,the image samples of transmission equipment inspection are multiplied,and then the amplified samples are used to complete the training of the visual defect detection model of transmission equipment,to realize the optimization and improvement of the visual defect detection effect.Through the simulation experiment,it is verified that the cyclic generative adversarial data proliferation improves the effectiveness and stability of the visual defect detection of transmission equipment,which provides a new perspective and idea for the visual defect detection of transmission equipment.

关 键 词：生成对抗网络 数据增殖 缺陷检测 目标检测 输电设备 

分 类 号：TN915.853[电子电信—通信与信息系统]