基于自适应SVM决策树的焊缝缺陷类型识别  

Welding Defects Classification Based on Adaptive SVM Decision Tree

作　　者：李坤[1] 文斌[2] 任清安[3] 罗爱民[4] 

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

出　　处：《无损检测》2010年第3期171-174,178,共5页Nondestructive Testing

基　　金：成都信息工程学院自然科学与技术发展基金资助(csrf200805)

摘　　要：针对传统X射线焊缝缺陷检测方法普遍存在分类识别精度不高的问题,提出了一种基于分离程度的自适应SVM决策树算法。首先对滤波后的X-Ray焊缝缺陷图像进行数学形态学重建,然后根据分离程度,每次将分离程度最大的缺陷类别首先分离出来,构造自适应二叉树的SVM分类器,从而达到了减小二叉树的累积误差,得到了分类性能优良的的SVM决策树,并用其对X-Ray焊缝缺陷图像进行分类识别。实验结果表明,该算法取得了好的分类精度和识别效果。An adaptive SVM（Support Vector Machines） based on binary tree using the degree of separation is proposed in this paper, aiming at the problem that it＇ s difficult for traditional detection methods to accurately identify the welding defects of X-Ray images. Firstly, mathematical morphological reconstruction is applied to the filtered X-Ray images of welding defects. It is proposed to separate category of defects with the largest degree of separation as a priority, and to construct adaptive SVM classifiers based on binary tree, thus decreasing the accumulated error. Finally, a SVM decision tree of good classification performance can be obtained, which is used to classify and identify the X-Ray images of welding defects, and it shows that the algorithm has made a good classification and recognition accuracy results.

关 键 词：决策二叉树 支持向量机 分离程度 数学形态学 缺陷识别 

分 类 号：TG115.28[金属学及工艺—物理冶金]