不均匀光照下带钢表面缺陷检测方法的研究
Study on Detection Method of Surface Defects of Strip Steel Under Uneven Illumination
黄新波 1孙苏珍 1张烨 1李博涛 1任玉成 2赵隆1
作者信息
- 1. 西安工程大学电子信息学院,西安 710048
- 2. 中国重型机械研究院股份公司,西安 710032
- 折叠
摘要
针对带钢表面存在光照不均及缺陷类型繁杂的问题导致缺陷检测精度不高的情况,提出一种基于梯度相似性引导模板的带钢缺陷检测方法.在利用二维高斯函数估计背景模板的基础上,结合梯度相似性引导其最优参数的选取;将原图像与优化后的背景模板进行图像差分操作,消除部分不均匀光照,同时构造分段函数对差分图像进行光照补偿,并结合Meanshift算法去除噪点,进一步增强差分图像;利用改进的K-means算法进行聚类分割并结合区域标记实现带钢表面不同缺陷的准确检测.实验结果表明,该方法能够在不均匀光照下准确检测出夹杂、麻点、酸洗、划痕等多种带钢表面缺陷,召回率和精确度分别达到 93.2%和 96.8%.
Abstract
Aiming at the problems of uneven illumination and complex defect types on the strip surface,which lead to low defect detection accuracy,a strip defect detection method based on gradient similarity-guided template is proposed.First,on the basis of estimating the background template with a two-dimensional Gaussian function,the selection of the optimal parameters is guided by the gradient similarity;secondly,the image difference operation is performed on the original image and the optimized background template to eliminate some uneven illumination,and at the same time a piecewise function is constructed to compensate the difference image for illumination,and the Meanshift algorithm is used to remove noise to further enhance the difference image.Finally,the improved K-means algorithm is used to perform cluster segmentation and combined with regional markers to achieve accurate detection of different defects on the strip surface.The experimental results show that the method can accurately detect various surface defects of strip steel such as inclusions,pits,pickling,and scratches under uneven illumination,and the recall rate and accuracy reach 93.2%and 96.8%,respectively.
关键词
不均匀光照/图像差分/K-means聚类分割/表面缺陷检测Key words
uneven illumination/image difference/K-means clustering segmentation/surface defect detection引用本文复制引用
基金项目
陕西省自然科学基础研究计划(2022JQ-568)
陕西省教育厅科研项目(21JK0661)
出版年
2024
国家科技期刊平台
NSTL
万方数据