基于特征识别的焊缝表面质量检测方法  

Surface Quality Detection Method of Weld Based on Feature Recognition

作　　者：牛玉玺 杨守鹏 凌鹤[3] NIU Yu-xi;YANG Shou-peng;LING He(School of Civil Engineering and Architecture,Wuhan University of Technology,Wuhan 430070,China;Wuhan Tobacco Monopoly Administration,Wuhan 430040,China;School of Mechanical and Electrical Engineering,Wuhan University of Technology,Wuhan 430070,China)

机构地区：[1]武汉理工大学土木工程与建筑学院,武汉430070 [2]武汉市烟草专卖局(公司),武汉430040 [3]武汉理工大学机电工程学院,武汉430070

出　　处：《武汉理工大学学报》2022年第3期79-86,共8页Journal of Wuhan University of Technology

基　　金：武汉理工大学自主创新研究基金(2019Ⅲ146CG)。

摘　　要：针对现有焊缝质量检测效率低且无法满足实时性需求等问题,提出了一种基于灰度图像特征识别的焊缝质量检测方法。采用高亮度激光发射器作为主动光源投射到被检测物体表面,避免了环境光对焊缝图像的干扰;通过一种高度映射灰度值的方法对三维点云进行降维处理,大幅降低焊缝模型数据量;使用主成分分析法对图像进行灰度和几何形状特征提取;通过一种在有限数据集下具有良好分类效果的机器学习方法——支持向量机(SVM),来实现灰度图像中缺陷特征的有效识别和分类。实验中可以有效提取出焊缝表面缺陷特征并实现识别和分类,在检测精度和效率方面有良好的表现。The existing weld quality inspection efficiency is low and can not meet the real-time requirements,so a method of weld quality inspection based on gray image feature recognition is proposed.The high-brightness laser transmitter is used as the active light source to project to the surface of the object to be inspected,avoiding the interference of ambient light on the weld image;through a method of highly mapping gray values,the three-dimensional point cloud is reduced in dimension,which greatly reduces the weld Model data volume;use principal component analysis to extract features of grayscale and geometric shapes;achieve grayscale images through a machine learning method with a good classification effect under limited data sets-Support Vector Machine(SVM)Fast identification and classification of defects in medium.In the experiment,the surface defect characteristics of the weld can be effectively extracted and recognized and classified,and it has a good performance in detection accuracy and efficiency.

关 键 词：点云降维 主成分分析 特征提取 多分类支持向量机 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]