全域优化孪生向量机的钢板表面缺陷图像分类  

Classification of defect images on steel plate surface by global optimized twin support vector machine

作　　者：胡鹰 侯政通 安宇 乔磊明 HU Ying;HOU Zhengtong;AN Yu;QIAO Leiming(College of Computer Science and Technology,Taiyuan University of Science and Technology,Taiyuan 030024,China)

机构地区：[1]太原科技大学计算机科学与技术学院,太原030024

出　　处：《重庆理工大学学报（自然科学）》2022年第10期140-150,共11页Journal of Chongqing University of Technology：Natural Science

基　　金：国家自然科学基金项目(52175354);山西省重大专项课题项目(20181102016);山西省专利推广项目(20210524)。

摘　　要：针对传统的钢板表面缺陷图像分类算法只是单纯地将异类样本分开,没有充分考虑到样本之间底层数据的关联性,在孪生支持向量机的基础上,提出了一种全域优化孪生支持向量机钢板表面缺陷图像分类算法。首先,嵌入线性判别分析,挖掘钢板表面图像数据全局结构信息,最小化同类样本间离散度;其次,采用K最近邻算法,最大化异类样本间离散度;最后,利用核函数,在高维空间解决非线性问题分类,得到表面缺陷图像分类结果。在2个公开数据集上的实验结果表明:所提方法对钢板表面缺陷图像分类的准确率可达94.90%和89.19%,比其他算法有进一步提升。Aiming at the traditional classification algorithm of steel plate surface defect image,it only separates dissimilar samples,and does not fully consider the correlation of underlying data between samples,a global optimization twin support vector machine classification algorithm for steel plate surface defect images is proposed in the paper based on Twin Support Vector Machine.Firstly,a Linear Discriminant Analysis is embedded to excavate the global structure information of steel plate surface image data and minimize the with-class scatter;secondly,a K-nearest neighbor algorithm is used to compensate the shortage of Linear Discriminant Analysis and maximize the between-class scatter;finally,a kernel function is used to solve the classification problem of nonlinear problems in high-dimensional space and obtain surface defect images classification results.The experimental results on two publicly available datasets show that the proposed method can achieve an accuracy of 94.90% and 89.19% for the classification of surface defects images of steel plates,which has a further improvement compared with other algorithms.

关 键 词：缺陷分类 图像处理 孪生支持向量机 全局信息 局部信息 K近邻 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]