硬质合金辊环表面缺陷检测算法  

Surface Defect Detection Algorithm for Hard Alloy Roller Collars

作　　者：郝鑫山 柴郡[1] 高艳菲[1] 姚立学 王云香 Hao Xinshan;Chai Jun;Gao Yanfei;Yao Lixue;Wang Yunxiang(Jinchuan Group Co.,Ltd.,Jinchang Gansu 737100,China)

机构地区：[1]金川集团股份有限公司,甘肃金昌737100

出　　处：《硬质合金》2025年第1期48-54,共7页Cemented Carbides

摘　　要：为可靠、精准检测硬质合金辊环微小以及相似等不同类别的表面缺陷,提出硬质合金辊环表面缺陷检测算法。该算法以采集到的硬质合金辊环表面图像为基础,对图像进行颜色空间处理,生成硬质合金辊环图像颜色空间分布方差显著图;将该显著图输入多尺度注意力机制网络模型中,通过模型编码器部分中金字塔切分注意力模块提取显著图中的边缘轮廓缺陷,将其与全局平均池化后输出的内部区域缺陷相结合,再经由上采样处理,输出表面缺陷检测结果。测试结果显示:该算法具备较好的应用效果,生成显著图中可呈现辊环的整体轮廓结构以及辊环上的缺陷位置,为后续辊环表面缺陷检测提供可靠依据;精准完成微小以及相似等不同类别的表面缺陷的检测,定位误差均值低于0.018 mm,检测精度较高。A surface defect detection algorithm for hard alloy roller collars was proposed to reliably and accurately detect small and similar surface defects of different categories on hard alloy roller collars.This algorithm was based on the collected surface image of roller collars,and color space processing was performed on the hard alloy roller collar image to generate a variance saliency map of the color space distribution of the hard alloy roller collar image.The saliency map was input into a multi-scale attention mechanism network model,and edge contour defects were extracted from the saliency map through the pyramid segmentation attention module in the model encoder section.They were combined with internal region defects output by global average pooling,and then surface defect detection results were output through upsampling processing.The test results show that the algorithm has good application effects,generating a saliency map that can present the overall contour structure of the roller collar and the location of defects on the roller collar,providing a reliable basis for subsequent surface defect detection of roller collars.It accurately detects small and similar surface defects of different categories,with an average positioning error of less than 0.018 mm and high detection accuracy.

关 键 词：硬质合金 辊环 表面缺陷检测 颜色空间处理 显著图 特征结合 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TG115[自动化与计算机技术—计算机科学与技术]