基于多光源图像信息融合的绝缘子污秽状态识别

Recognition of Insulator Contamination Grades Based on Multi-Light Images Information Fusion

绝缘子污秽状态非接触检测是智能变电巡检的重要组成部分,为有效提高绝缘子污秽状态识别率,提出了一种基于多光源图像决策级融合的污秽状态诊断方法.以沿海地区多所变电站中不同污秽状态的绝缘子为研究对象,采用种子区域生长法进行图像分割后,分别提取其可见光颜色空间特征、红外图像的灰度化特征以及环境特征,再依据Fisher判据筛选得到最优表征量,并设计支持向量机多值分类器进行污秽状态初判.基于各自识别结果,引入D-S理论进行决策级融合,实现绝缘子污秽状态的有效识别.试验结果表明:本文融合算法的正确率明显高于单种图像源的识别,达到95%左右,为带电检测作业中绝缘子的非接触检测及故障智能诊断提供了新思路.

Non-contact detection of insulator contamination grades is an important part of intelligent substation inspection,in order to effectively improve the recognition rate of insulator contamination state,a method for the diagnosis of the contamination state based on multi-light source images'decision-level fusion is proposed.Contaminated insulators in some transformer substations in the coastal areas are chosen as the data source of images.Firstly,adopt the seed region growing method to realize image segmentation and extract its visible light color space characteristics,grayscale characteristics of the infrared image and environmental characteristics respectively.Then,get the optimal vector characters based on Fisher criterion and design the multi-valued support vector machine to obtain the initial contamination classify,Finally,based on their separate recognition result,adopt the D-S theory to achieve the decision-level fusion to realize the effective and accurate recognition.The experiment results show that the accuracy of the fusion algorithm is higher than the recognition of single image source,rising to about 95%,which provides a new idea for the non-contact detection of insulators and intelligent fault diagnosis in the process of electriferous monitoring.