基于多尺度LBP特征的带钢表面缺陷图像SVM分类  

SVM Classification of Surface Defect Images of Strip Based on Multi-scale LBP Features

作　　者：刘启浪 汤勃[1] 孔建益[1] 王兴东[1] LIU Qi-lang;TANG Bo;KONG Jian-yi;WANG Xing-dong(School of Machinery and Automation Engineering,Wuhan University of Science and Technology,Wuhan 430081,China)

机构地区：[1]武汉科技大学机械自动化学院,武汉430081

出　　处：《组合机床与自动化加工技术》2020年第12期27-30,共4页Modular Machine Tool & Automatic Manufacturing Technique

摘　　要：为提高带钢表面缺陷图像的分类准确率,文章研究了带钢表面缺陷图像多尺度局部二值模式(Local Binary Pattern,LBP)特征。通过提取多种类型的多尺度LBP特征以及不同尺度的LBP联合特征,并与灰度共生矩特征进行对比;利用支持向量机(Support Vector Machine,SVM)进行分类实验。实验结果表明,对于带钢表面缺陷图像的LBP特征,(16,2)尺度LBP特征的分类准确率高于(8,1)尺度LBP特征;两种尺度联合特征分类准确率高于单一尺度特征;各类LBP特征与灰度共生矩特征中,LBP直方图傅里叶变换特征具有更高的分类准确率。In order to improve the classification accuracy of strip surface defect images,multi-scale local binary pattern(LBP)features of strip surface defect images were studied.By extracting multiple types of multi-scale LBP features and LBP joint features at different scales and compare it with the feature of gray level co-occurrence matrix;Use support vector machine(SVM)for classification experiments.The experimental results show that for LBP features of strip surface defect images,the classification accuracy rate of(16,2)scale LBP features is higher than(8,1)scale LBP features;the classification accuracy rate of combined features of two scales is higher than that of single scale features;Among the various types of LBP features and gay level co-occurrence matrix features,the LBP histogram Fourier transform feature has higher classification accuracy.

关 键 词：带钢缺陷 局部二值模式 多尺度 特征提取 SVM分类 

分 类 号：TH16[机械工程—机械制造及自动化] TG506[金属学及工艺—金属切削加工及机床]