提高钢板表面缺陷检出率的分割新方法  

A new segmentation method to improve the detection rate of steel plate surface defects

作　　者：赵章焰[1] 栗子玉 Zhao Zhangyan;Li Ziyu

机构地区：[1]武汉理工大学交通与物流工程学院,武汉430070

出　　处：《起重运输机械》2024年第16期27-35,共9页Hoisting and Conveying Machinery

摘　　要：文中针对传统图像分割方法难以准确分割钢板表面小目标缺陷的问题,提出了一种基于二维灰度直方图反向投影的缺陷分割新方法。该方法无需手工标注大量缺陷样本,利用图像的二维灰度直方图反向投影,提取小目标缺陷轮廓。当图像不具备明显双峰效应时,排除背景像素的干扰,将缺陷轮廓图像进行形态学处理和二值化,实现目标缺陷与背景像素的准确分割。实验结果表明,所提方法在钢板表面缺陷数据集上分割精确度综合评价指标达到92.45%,优于其他先进检测方法。同时搭建实验平台,结果显示在模拟工程实验中所提方法能够准确分割多种缺陷。Considering that it is difficult to accurately segment small target defects on steel plate surface by traditional image segmentation methods,a new defect segmentation method based on two-dimensional gray histogram back projection is proposed in this paper,which does not need to tag a large number of defect samples manually,and can extract the small target contour defects through the back projection of the two-dimensional gray histogram of the image.When there was no obvious double-hump effect in the image,the interference of background pixels was eliminated,and the defect contour image was morphologically processed and binarized to realize the accurate segmentation of the target defect and background pixels.The experimental results show that the segmentation accuracy of the proposed method on the data set of steel plate surface defects reached 92.45%,which was superior to other advanced detection methods.In addition,an experimental platform was built,and the results show that the proposed method can accurately segment various defects in simulation engineering experiments.

关 键 词：缺陷检测 图像分割 钢板表面缺陷 新方法 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]