基于金字塔多层特征自适应的变电设备缺陷检测
投稿时间:2020-06-02  修订日期:2020-09-21  点此下载全文
引用本文:
摘要点击次数: 37
全文下载次数: 0
作者单位E-mail
王丽蓉 国网湖南省电力有限公司检修公司 329986607@qq.com 
华雄 安徽南瑞继远电网技术有限公司 136012368@qq.com 
伍艺佳 国网湖南省电力有限公司检修公司  
陈红波 中国科学院合肥物质科学研究院 智能机械研究所 136012368@qq.com 
中文摘要:针对现有的巡检机器人缺陷图像识别检测算法不能充分融合特征金字塔网络多层特征的问题,提出了一种基于金字塔多层特征自适应的变电设备缺陷图像识别检测算法。首先,构建出一种新的RoI特征金字塔结构用以提取各个层级的变电设备缺陷图像特征图;其次,设计出一种多层特征自适应学习函数来自动学习不同层级特征权重;最后,结合目标大小尺度检测框架实现变电站缺陷图像区域检测。所设计的方法增强了变电设备缺陷图像检测的鲁棒性。实验结果显示,所提方法的mAP达到了71.9%。
中文关键词:变电设备  缺陷图像  金字塔特征  自适应
 
Substation Equipment Defect Detection based on Pyramid Multi level feature adaptation
Abstract:In order to solve the problem that the existing defect image recognition and detection algorithm can not fully integrate the multi-layer features of the pyramid network, an adaptive pyramid multi-layer feature based defect image recognition and detection algorithm is proposed for substation equipment. Firstly, a new ROI feature pyramid structure is constructed to extract the defect image feature map of each level of substation equipment; secondly, a multi-layer feature adaptive learning function is designed to automatically learn the feature weights of different levels; Finally, the substation defect image region detection framework is designed. Our proposed method improve the robustness of defect detection. Extensive experiments show that our method achieve recognition and detection mAP accuracy of 71.9%.
keywords:substation  equipment, defect  image, pyramid  features, self-adaption
查看全文   查看/发表评论   下载pdf阅读器