基于X射线图像的厚钢管焊缝中气孔缺陷的自动检测

发布时间：2018-01-29 08:34

本文关键词： X射线图像独立分量分析缺陷检测机器学习厚钢管　出处：《计算机应用》2017年03期 　论文类型：期刊论文

【摘要】：由于厚钢管X射线图像强度分布不均匀,对比度低、噪声大,且气孔缺陷的大小、形状、位置、对比度各异,使得自动检测各种类型的气孔较为困难。针对传统缺陷检测算法中手工标记缺陷数据工作量大,焊缝边缘难以准确提取等问题,提出一种新的无监督学习的各种气孔缺陷检测算法。首先,采用快速独立分量分析从钢管X射线图像集合中学习一组独立基底,并用该基底的线性组合来选择性重构带气孔缺陷的测试图像;随后,测试图像与其重构图像相减获得差异图像,通过全局阈值从差异图像中将各种气孔分割出来。实验的训练集有320幅,测试集有60幅图像,所提算法检测结果的平均敏感性和准确率为90.5%和99.7%。实验结果表明,该算法无需手工标记数据或提取焊缝边缘,可准确检测各种气孔缺陷。
[Abstract]:Because of the uneven distribution of X-ray image intensity, low contrast, large noise, and the size, shape, position and contrast of the hole defects. It is difficult to detect various types of holes automatically. Aiming at the problems of manual marking defect data and the difficulty of accurately extracting weld edge in the traditional defect detection algorithm. A new unsupervised learning algorithm for detecting stomatal defects is proposed. Firstly, a set of independent substrates is studied by using fast independent component analysis (FICA) from the set of X-ray images of steel tubes. The linear combination of the substrate is used to selectively reconstruct the test image with stomatal defects. Then, the test image is subtracted from the reconstructed image to obtain the differential image, and a variety of pores are segmented from the differential image by the global threshold. The experimental training set has 320 pieces, and the test set has 60 images. The average sensitivity and accuracy of the proposed algorithm are 90.5% and 99.7.The experimental results show that the algorithm can accurately detect various pore defects without manually marking data or extracting weld edges.
【作者单位】：上海交通大学电子信息与电气工程学院;宝山钢铁股份有限公司研究院;宝山钢铁股份有限公司钢管条钢事业部;
【基金】：国家自然科学基金资助项目(61375020)~~
【分类号】：TG441.7;TP391.41
【正文快照】： 0引言焊接缺陷会影响焊接钢管的使用寿命,因此自动检测焊接缺陷可提高焊接钢管出厂的合格率[1-3]。在基于钢管X射线图像进行无损探测中,自动检测焊缝缺陷是一个重要的研究主题[4-10]。由于X射线图像强度分布不均匀,缺陷的形状、尺寸大小、亮度、分布位置各异,使得从钢管X射线

【相似文献】