现代制造工程 ›› 2017, Vol. 445 ›› Issue (10): 106-109.doi: 10.16731/j.cnki.1671-3133.2017.10.020

• 制造技术/工艺装备 • 上一篇    下一篇

基于支持向量机的焊缝缺陷类型识别研究

李宁1, 卢子广2   

  1. 1 广西机电职业技术学院电气工程系,南宁 530007;
    2 广西大学电气工程学院,南宁 530004
  • 收稿日期:2016-08-26 出版日期:2017-10-20 发布日期:2018-01-08
  • 作者简介:李宁,副教授,硕士,研究方向为焊缝跟踪系统、工业机器人控制。 卢子广,教授,博士,研究方向为智能控制,无线传感器网络。 E-mail:li_ning_gxjd@sina.com;lu_ziguang@21cn.com
  • 基金资助:
    广西高等学校优秀中青年骨干教师培养工程资助项目

Type recognition of weld defects based on support vector machines

Li Ning1, Lu Ziguang2   

  1. 1 Department of Electrical Engineering,Guangxi Technological College of Machinery and Electricity,Nanning 530007,China;
    2 College of Electrical Engineering,Guangxi University,Nanning 530004,China
  • Received:2016-08-26 Online:2017-10-20 Published:2018-01-08

摘要: 针对焊缝的线形和圆形两种主要缺陷,提出了一种基于支持向量机的焊缝缺陷种类识别算法。首先,对焊缝X射线图像运用模糊C均值聚类、区域填充、均值滤波、边缘检测、大津阈值及谷发现图像预处理算法,获取焊缝缺陷的位置,然后通过逆表面阈值算法将缺陷从焊缝中分割出来;利用基于分段分形纹理分析算法提取焊缝缺陷的特征值;最后将特征值输入到基于支持向量机的焊缝缺陷分类器中,识别出焊缝缺陷种类。试验结果表明,通过对150张焊缝X射线图像进行训练,对80张焊缝X射线图像进行测试,平均正确识别焊缝缺陷种类的准确率达97.5%,满足工业要求。

关键词: 焊缝, 缺陷, 图像处理, 支持向量机

Abstract: For linear and circular two kinds of weld defects,proposed a method of weld defect type recognition algorithms based on Support Vector Machine (SVM).First of all,some image pre-processing algorithms such as fuzzy C means clustering,region filling algorithm,average filtering,edge detection,Otsu thresholding and inverse thresholding,to get the approximate location of weld defects.The information of the particular region will be extracted using segmentation based fractal texture analysis,SVM is used to classify the segmented defect as line or circular defects based on the extracted features lastly.The results showed that,the average accuracy rate is 97.5% for correcting identification of the type of weld defects,by 150 weld X-ray image is trained and 80 X-ray image weld test,which can meet industrial requirements.

Key words: weld, defect, image processing, Support Vector Machine(SVM)

中图分类号: 


版权所有 © 《现代制造工程》编辑部 
地址:北京市东城区东四块玉南街28号 邮编:100061 电话:010-67126028 电子信箱:2645173083@qq.com
本系统由北京玛格泰克科技发展有限公司设计开发 技术支持:support@magtech.com.cn
访问总数:,当日访问:,当前在线: