基于HSV特征变换与目标检测的变压器呼吸器缺陷智能识别方法  

Intelligent Defect Identification Method of Transformer Dehumidifier Based on HSV Transformation and Object Detection

作　　者：李瑞生[1] 许丹[1] 翟登辉[1] 陈晓民 张旭 张彦龙 LI Ruisheng;XU Dan;ZHAI Denghui;CHEN Xiaomin;ZHAGN Xu;ZHANG Yanlong(XJ Group Corporation Ltd.,Xuchang 461000,China;XJ Electric Co.,Ltd.,Xuchang 461000,China)

机构地区：[1]许继集团有限公司,许昌461000 [2]许继电气股份有限公司,许昌461000

出　　处：《高电压技术》2020年第9期3027-3034,共8页High Voltage Engineering

基　　金：国家电网公司科技项目(面向智能电网运维场景的视觉主动感知与协同认知技术研究与应用)(5600-202046347A-0-0-00)。

摘　　要：为了利用深度学习算法提高变压器运维的智能化水平,提出了一种基于色相饱和度值(hue-saturation-value,HSV)特征变换与目标检测的变压器呼吸器缺陷智能识别方法。该方法利用单发多盒探测器(single shot multibox detector,SSD)网络框架进行呼吸器目标提取,采用HSV颜色转换完成空间映射,通过设定HSV特征阈值进行呼吸器正常颜色和异常颜色的跟踪和提取,进而通过各颜色分量比例与分布情况进行呼吸器状态的智能判断。研究结果表明:所提识别方法能够利用图像特征对变压器呼吸器进行准确定位与状态识别。论文研究可为电力设备锈蚀识别等其他类似场景提供参考。In order to use the deep learning method to improve the intelligent level of transformer operation and maintenance,we proposed an intelligent recognition method of transformer dehumidifier defects based on hue-saturation-value(HSV)feature transformation and object detection.Firstly,the respirators are extracted by adopting single shot multibox detector(SSD)network,and the features are achieved through spatial mapping by HSV color transformation.Then,the normal and abnormal color components of dehumidifier are tracked and extracted by setting the HSV characteristic thresholds.Finally,the dehumidifier state is estimated by proportion and distribution of each color components.The research results show that the proposed recognition method can be adopted to accurately locate and identify the status of the transformer dehumidifier by using image features.The research can provide a reference for other similar scenes such as rust recognition of power equipment.

关 键 词：变压器呼吸器 目标检测 深度学习 HSV变换 颜色跟踪 状态分类 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TM407[自动化与计算机技术—计算机科学与技术]