利用层次模型进行训练学习的线路设备缺陷检测方法  

Defect Detection of Line Facility Using Hierarchical Model with Learning Algorithm

作　　者：付晶[1] 邵瑰玮[1] 吴亮[2] 刘磊[2] 季铮[2] 

机构地区：[1]中国电力科学研究院,武汉430074 [2]武汉大学遥感信息工程学院,武汉430079

出　　处：《高电压技术》2017年第1期266-275,共10页High Voltage Engineering

摘　　要：无人机输电线路巡检中的线路部件缺陷自动化检测一直是一个难题,为了解决输电线路中螺母-销钉、防振锤部件缺陷的判别问题,提出了利用层次模型"与或图"对目标进行分解表达,建立部件之间的约束关系,构建多向的判别路径的方法。利用基于类Haar特征和级联Ada Boost分类器对目标基元进行识别,通过数据合成扩充数据样本用于训练,以提高分类器的训练性能。实验结果表明,该方法有效综合了识别检测技术,在少量样本的条件下,能有效地实现若干输电线路部件的缺陷判别,为输电线路缺陷检测任务提供了一种参考方法。In order to solve the problem of detecting the defects of nuts, pins, and anti-vibration hammer parts in the transmission line, we propose a method of using the hierarchical model ＂And-Or＂ graph to detect the defects of the circuit components in UAV power transmission line inspection. The decomposition of the expression as well as the establishment of the relationship between the constraints of components is used to build multi-way of determining the path of the meth- od. Moreover, we classify the target primitives by hierarchical AdaBoost classifier using Hart feature, and extend the data samples for training through data synthesis to improve the training performance of the classifier. The experimental results show that the proposed method can effectively detect the defects of several transmission line components under the condi- tion of insufficiency of samples, and provide a reference method for transmission line defect detection tasks.

关 键 词：线路设备 缺陷检测 层次模型 训练 合成样本 

分 类 号：TM75[电气工程—电力系统及自动化]