基于改进二叉树多分类SVM的焊缝缺陷分类方法  

Method of multi-classification by improved binary tree based on SVM for welding defects recongnition

作　　者：罗爱民[1,2] 沈才洪[2] 易彬[2] 李坤[3] 

机构地区：[1]四川大学轻纺与食品学院,成都610064 [2]泸州老窖股份有限公司,四川泸州646000 [3]四川大学电子信息学院,成都610065

出　　处：《焊接学报》2010年第7期51-54,共4页Transactions of The China Welding Institution

基　　金：国家自然科学基金资助项目(50275100);四川大学青年基金资助项目(JS校200323)

摘　　要：为了进一步提高焊缝缺陷识别精度,定义了一种类分离度,提出了改进二叉树多分类SVM的焊缝缺陷分类方法.在焊缝缺陷分类时,计算每个类的类分离度,将类分离度最小的两个类进行训练得到SVM子分类器SVM_1,并将这两个类合并成一个新簇G;同理对新簇G和剩下的k-2类进行类分离度的评估,将类分离度最小的两类训练得到SVM子分类器SVM_2,并合并成新簇H,直至得到k-1个SVM分类器,训练结束得到良好的二叉树的分类结构.利用聚类生成好的优化二叉树SVM进行判别焊接缺陷.结果表明,新算法具有高的分类精度和推广能力.To further recognition accuracy,the multi-classification by improved binary tree based on SVM is raised for welding defects recognition. In the welding defects classification,each class separation is computed,the classes of the two smallest class separation are trained to generate the sub-classification SVM_1 and then are combined into a new cluster G. The new cluster G and the remaining classes are computed similarly,and the second sub-classification SVM_2 and new merged cluster H are produced. This work would be repeated until the (k-1)-th sub-classification SVM is obtained and finally a balanced binary tree is come into being. Then,the optimized binary tree based on SVM by clustering is applied in welding defects recognition. The experiments show high recognition accuracy and strong generalization ability by our new algorithm.

关 键 词：支持向量机 类分离度 二叉树 焊缝缺陷识别 

分 类 号：TG115.28[金属学及工艺—物理冶金]