基于分离程度的二叉树的多类SVM在焊接缺陷分类和识别中的应用  

Multi-class SVM with binary decision tree based on degree of separation applied in welding defect classification and recognition

作　　者：李坤[1] 任清安[2] 文斌[3] 罗爱民[4] 

机构地区：[1]四川大学电子信息学院,成都610064 [2]四川大学数学学院,成都610065 [3]成都信息工程学院,成都610225 [4]四川大学轻纺与食品学院,成都610065

出　　处：《四川大学学报（自然科学版）》2010年第3期520-524,共5页Journal of Sichuan University(Natural Science Edition)

基　　金：成都信息工程学院自然科学与技术发展基金(csrf200805)

摘　　要：提出了基于分离程度的SVM决策树的焊缝缺陷分类识别方法.首先对X射线焊缝图像进行缺陷特征提取,然后结合聚类的思想,定义了分离程度,每次将分离程度最大的缺陷类分离出来,成功解决了传统欧氏距离不能处理的类交叉分类情况,得到了累积误差更小的决策树.将基于分离程度的二叉树的多类SVM算法运用于X射线焊接缺陷图像的分类识别,通过计算机仿真,表明该方法比其它SVM多分类算法在分类精度和识别效果方面有明显的提高.This paper proposed a welding defect identification and classification method based on the degree of separation of SVM decision tree. Firstly, the welding defect features of X-ray images were extracted. Then combined with the clustering thoughts, the degree of separation was defined, and the type of defect with the greatest degree of separation was separated at each time, which successfully solved the cross-category classification problem that traditional Euclidean distance could not deal with. Thus a decision tree with smaller accumulated error was established. Multi-class SVM with decision tree algorithm based on the degree of separation was researched in welding defect classification and identification of X- ray images, through computer simulation, it showed that the accuracy of results had been greatly improved in the identification and classification than any other SVM classification algorithms.

关 键 词：决策二叉树 支持向量机 分离程度 缺陷识别 

分 类 号：TN99[电子电信—信号与信息处理]