基于彩涂钢板表面质量缺陷的检测算法  

Detection algorithm of surface quality defects based on color coated steel plate

作　　者：林永春 夏鹏飞 董森 高东旭 宋久远 Lin Yongchun;Xia Pengfei;Dong Sen;Gao Dongxu;Song Jiuyuan(Qingdao Branch of HBIS Digital Technology Co.,Ltd.,Qingdao 266101,Shandong;Qingdao HBIS New Material Technology Co.,Ltd.,Qingdao 266599,Shandong)

机构地区：[1]河钢数字技术股份有限公司青岛分公司,山东青岛266101 [2]青岛河钢新材料科技股份有限公司,山东青岛266599

出　　处：《河北冶金》2025年第1期47-51,共5页Hebei Metallurgy

摘　　要：针对高速生产过程中彩涂钢板表面结构性及非结构性缺陷存在尺寸极小、特征边界模糊、人工检测准确率低等问题,提出了一种彩涂钢板表面质量缺陷检测算法。首先收集彩涂钢板缺陷图像并建立多类型缺陷图像数据集;其次利用传统图像处理算法提取缺陷感兴趣区域(Region of Interest,ROI);最后将提取的缺陷感兴趣区域送入卷积神经网络输出缺陷分类结果。实验结果表明:缺陷检测算法对于彩涂钢板表面杂质及搁伤等缺陷的检出率和准确率分别达到了95%和91%,由此证明该算法具有较高的缺陷识别能力,满足了彩涂钢板表面质量检测需求;算法处理速度10帧/s,达到了彩涂钢板生产速度要求。该算法实现了彩涂钢板表面缺陷自动检测功能,达到了提高彩涂钢板缺陷检测准确率、降低人工负荷、节约彩涂钢板生产成本的目标,对彩涂钢板生产行业具有重要的意义。Aiming at the structural and non-structural defects appearing on the surface of the color coated steel plate in the production process,there are problems such as extremely small size,fuzzy feature boundaries,not easy to capture and low accuracy of manual detection,etc.In this paper,we propose an algorithm for detecting the defects on the surface quality of the color coated steel plate.Firstly,collecting defective images of the steel plate and establishing a multi-typed defective dataset.Secondly,extracting the defective ROIs by using the traditional image processing.Finally,sending the extracted ROIs into the convolutional neural network to output the defect classification results.Application practice shows that the defect detection algorithm of this paper for the color coated steel surface impurities and stranded injuries and other defects in the detection rate and accuracy of 95%and 85%respectively,which proves that the algorithm has a high defect recognition ability to meet the needs of the color coated steel surface quality.Make the algorithm processing performance of 8 frames/s to meet the requirements of the production speed of the color coated steel plate.The algorithm in this paper achieves the function of automatic detection of defects on the surface of color coated steel plate and achieves the purpose of improving the accuracy of defect detection of color coated steel plate,reducing the manpower load and saving the production cost of color coated steel plate.This is of great significance to the color coated steel plate production industry.

关 键 词：图像处理 卷积神经网络 彩涂钢板 缺陷检测 表面质量 

分 类 号：TG115[金属学及工艺—物理冶金]