关键词: 钢卷端面, 表面缺陷检测, 灰度共生矩阵, 仿射变换, CUSUM统计算法

Abstract: To address the challenges in detecting edge damage and loose coil surface defects in steel coils end-face,a steel coil end-face defect detection method based on pixel texture features was proposed. First,Gray-Level Co-occurrence Matrix (GLCM) and Gaussian filtering were combined to preprocess the images. This facilitated the extraction of contour information using gradient algorithms. Next,rotation centers were calculated from the extracted contour matrices.Rotation matrices were generated to iteratively align the steel coil strips to a vertical orientation using affine transformation. Then,the CUSUM statistical algorithm was applied to distinguish high-brightness and high-darkness regions in longitudinal pixel values obtained from PCA-based statistics.This allowed the differentiation between edge damage defects and loose coil defects. Finally,the peak and valley values in the image data were calculated. This enabled the automated detection and precise localization of the two types of defects in arbitrary texture orientations.The experimental results showed that the precision,recall,and mean Average Precision (mAP) for edge damage defect detection reached 96.2 %,95.6 %,and 96.4 %,respectively.For loose coil defect detection,these metrics were 98.0 %,96.9 %,and 99.2 %,respectively.The overall precision,recall,and mAP achieved 97.7 %,96.8 %,and 97.3 %,respectively. The results demonstrated that the proposed method achieved high accuracy and reliability in detecting edge damage and loose coil defects. This method provided a potential solution for automated detection of steel coil end-face defects in industrial production.

Key words: steel coil end-face, surface defect detection, Gray-Level Co-occurrence Matrix (GLCM), affine transformation, CUSUM statistical algorithm