Abstract: In order to detect hole defects on cold rolled steel strip edge, a novel detection method combining on-line inspection and off-line extraction of defect details is proposed. An adaptive region growing method is proposed for image segmentation in on-line module. Morphological processing and optimal entropy method are combined to find out seed points, which can overcome the shortage of selecting seed points by manual in traditional method, and the area of the largest component in the result of region growing is employed to judge whether hole defects exist. For making sure the accuracy of detection, block projection histogram matching features and Tamura texture are individually used to extract defect information in off-line module. Defect details can be obtained by integrating two detection results. Multiple images are processed by the algorithm proposed, and experimental results show that the on-line part has high processing speed and discrimination rate. From the off-line part, the number, area and location of holes can be obtained. The speed and precision meet the practical requirements.

Key words: strip edge, hole defect, image processing, quality control

摘要： 针对冷轧带钢边部孔洞缺陷信息的提取问题，综合在线检测和离线精确提取缺陷信息两方面进行了方案设计。首先在在线检测模块提出了一种自适应区域生长算法对图像进行分割，该法通过最佳熵法与形态学处理相结合实现种子点的自动选择，以替代传统算法中种子点的人工选取，以区域生长结果中最大连通体的面积判断是否存在缺陷；然后在离线模块，分别用块投影直方图匹配特征与Tamura纹理特征对缺陷图像进行缺陷信息提取，综合两种缺陷提取结果，得到缺陷详细信息。针对多幅图像采用提出的算法进行检测，在线部分具有较高的处理速度和缺陷判别率，离线部分实现了对缺陷图像中孔洞个数、每个孔洞面积及位置的定量估计。

关键词: 带钢边部, 孔洞缺陷, 图像处理, 质量控制