Abstract

An automated defect recognition algorithm is presented for detecting and classifying weld defects by photographic images. The proposed recognition algorithm selects a defective domain in a segmented image, extracts geometric features from the image, and relates the defect to one of six classes: no defect, cavity, longitudinal crack, transverse crack, burn-through, or multiple defect. The algorithm is implemented in the Matlab 2018b MathWorks environment and has been tested on 60 photographs of defects of various classes; the accuracy of recognition was 85%.

中文翻译：

---

基于几何属性的视觉检查过程中的计算机辅助焊接接头缺陷识别

摘要

提出了一种自动缺陷识别算法，用于通过摄影图像对焊接缺陷进行检测和分类。所提出的识别算法在分割的图像中选择缺陷域，从图像中提取几何特征，并将缺陷与六类之一相关：无缺陷，腔，纵向裂纹，横向裂纹，烧穿或多重缺陷。该算法在Matlab 2018b MathWorks环境中实现，并已在60张各种类别的缺陷照片上进行了测试；识别的准确性为85％。