基于机器视觉的电触头瑕疵检测与分类  

defect detection and classification of electrical contacts based on machine vision

作　　者：刘威 熊经先 刘思尧 闫坤 LIU Wei;XIONG Jingxian;LIU Siyao;YAN Kun(School of Information and Communication,Guilin University of Electronic Technology,Guilin 541004,China;Guilin Jinge Electrical and Electronic Materials Technology Co.,Ltd.,Guilin 541004,China)

机构地区：[1]桂林电子科技大学信息与通信学院,广西桂林541004 [2]桂林金格电工电子材料科技有限公司,广西桂林541004

出　　处：《桂林电子科技大学学报》2023年第6期523-529,共7页Journal of Guilin University of Electronic Technology

基　　金：国家自然科学基金(62101147);广西自然科学基金(2020GXNSFAA159146);广西创新驱动发展专项(桂科AA21077008);认知无线电与信息处理教育部重点实验室基金(CRKL190108)。

摘　　要：在电力系统中,电触头材料在开关电器中不可缺少,其表面出现瑕疵时会导致接触面的电阻增大,继而使接触面发热,严重时可能会使开关电器失灵,影响高压电器的质量及使用寿命,且目前很多企业对零件表面质量检测还在用人工方法。针对上述问题,提出一种基于机器视觉与机器学习的电触头表面瑕疵检测与分类方法。该方法基于电触头的瑕疵种类和瑕疵特点搭建检测平台,并对采集的基于机器视觉与机器学习的电触头零件图像进行预处理,采用模板匹配的方法对图像中的电触头零件进行定位及表面优劣判断,对预筛选后的电触头图像进行表面降噪,并采用Otsu阈值分割算法对瑕疵区域进行分割。为了能有效地对瑕疵进行分类,对瑕疵区域进行多特征提取,通过设计的特征选择算法对特征进行最优选择,用决策树分类器进行瑕疵分类,分类准确率达92.6%。与支持向量机(SVM)算法相比,决策树分类器在分类时间和效率上优于SVM算法。实验结果表明,对分类数据集进行特征降维可提高分类的准确率。In the power system,the electrical contact material plays an indispensable role in the switchgear.When the surface is de-fective,the resistance of the contact surface will increase,which will cause the contact surface to heat up.In severe cases,the switchgear may fail and affect the quality and service life of high-voltage electrical appliances,and many companies are still using manual inspection for surface quality inspection of parts.Aiming at the above problems,a method for detection and classification of electrical contact surface defects based on machine vision and machine learning was proposed.This method builded a detection plat-form based on the defect types and characteristics of the electrical contacts,preprocessed the collected images of the electrical con-tact parts,and used the template matching method to locate the electrical contact parts in the image and judge the quality of the sur-face,the pre-screened electrical contact image was denoised on the surface,and the Otsu threshold segmentation algorithm was used to segment the defect area.In order to effectively classify the defects,multiple features were extracted for the defect area,and a fea-ture selection algorithm was designed to select the features optimally.The decision tree classifier was used to classify the defects,and the classification accuracy can reach 92.6%.Compared with the Support Vector Machine(SVM)algorithm,the decision tree model is better than the SVM model in terms of classification time and efficiency.The experimental results show that the feature di-mension reduction of the classification data set can improve the classification accuracy.

关 键 词：电触头 机器视觉 模板匹配 特征优选 决策树 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]