Abstract: It is a hotspot of detection without damage to automatically analyze and recognize the defects classification of weld image.After the noise reduction and image enhancement，the defects of weld seam image are processed by threshold extraction and the weld defects features are selected.Then the classification rules are established by decision tree in order to recognize the threshold extracted defects of weld seam image.The results of experiments show that the defects classification is very accurate and the knowledge expression is very easy to understand.

Key words: defects of weld seam, recognition, decision tree, image process

摘要： 自动对射线底片图像进行分析和缺陷类型识别是无损探伤研究领域的一个热点。在对焊缝射线底片进行图像去噪和图像增强的基础上，对焊缝底片图像进行二值化处理，进而提取焊缝缺陷图像及其特征，再采用决策树方法建立焊缝缺陷特征的分类规则，并用这些规则对二值化后的焊缝缺陷图像进行分类识别。实验结果表明，基于决策树方法对焊缝缺陷图像识别的准确率高，而且所表达的知识易于理解。

关键词: 焊缝缺陷, 识别, 决策树, 图像处理