• CN:11-2187/TH
  • ISSN:0577-6686

机械工程学报 ›› 2016, Vol. 52 ›› Issue (4): 13-19.doi: 10.3901/JME.2016.04.013

• 仪器科学与技术 • 上一篇    下一篇

基于Tetrolet变换的热轧钢板表面缺陷识别方法

徐科1, 2, 王磊2, 王璟瑜2   

  1. 1. 北京科技大学钢铁共性技术协同创新中心 北京 100083;
    2. 北京科技大学高效轧制国家工程研究中心 北京 100083
  • 出版日期:2016-02-15 发布日期:2016-02-15
  • 作者简介:徐科(通信作者),男,1972年出生,博士,研究员,博士研究生导师。主要研究方向为表面检测、图像识别、智能诊断等。E-mail:xuke@ustb.edu.cn;王磊,男,1987年出生,博士研究生。主要研究方向为表面检测、图像识别。E-mail:towanglei@163.com;王璟瑜,男,1987年出生,硕士研究生。主要研究方向为表面检测、图像识别、多尺度几何分析。E-mail:wanghehu3571@sina.com
  • 基金资助:
    “十二五”国家科技支撑计划(2012BAB19B06)和教育部博士点基金 (20120006110033)资助项目

Surface Defect Recognition of Hot-rolled Steel Plates Based on Tetrolet Transform

XU Ke1, 2, WANG Lei2, WANG Jingyu2   

  1. 1. Collaborative Innovation Center of Steel Technology, University of Science and Technology Beijing, Beijing 100083;
    2. National Engineering Research Center of Advanced Rolling Technology, University of Science and Technology Beijing, Beijing 100083
  • Online:2016-02-15 Published:2016-02-15

摘要: 通过Tetrolet变换将热轧钢板表面图像分解成不同尺度和方向的子带,提取子带的Tetrolet高通系数矩阵特征,得到一个高维的特征矢量。利用核保局投影算法对高维特征矢量进行降维,将降维后的低维特征矢量输入支持向量机,从而实现热轧钢板表面缺陷的分类识别。对现场采集到的热轧钢板表面图像样本进行试验,包括横向裂纹、纵向裂纹、横向划伤、纵向划伤、结疤、麻点、网纹、压痕等8类常见热轧钢板表面缺陷,以及氧化铁皮和无缺陷等样本。试验结果表明基于Tetrolet变换方法对样本图像的识别率可达97.38%,比基于Curvelet变换、Contourlet变换等方法得到的识别率提高1%左右。

关键词: Tetrolet变换, 表面检测, 多尺度几何分析, 热轧钢板, 特征提取

Abstract: Sample images of hot-rolled steel plates are decomposed into multiple subbands with different scales and directions by Tetrolet transform. The high-pass Tetrolet coefficients of subbands are combined into a high-dimensional feature vector. Kernel locality preserving projection(KLPP) is applied to the high-dimensional feature vector for dimension reduction, which results in a low-dimensional feature vector. The low-dimensional feature vector is fed into support vector machine(SVM) for surface defect recognition of hot-rolled steel plates. The method is tested with sample images from an industrial production line, including transversal cracks, longitudinal cracks, transversal scratches, longitudinal scratches, scars, pimples, net cracks, impressions, scales and no defect. The results show that the recognition rate with Tetrolet transform is 97.38%, which is about 1% higher than that with Curvelet transform and Contourlet transform.

Key words: feature extraction, hot-rolled steel plates, multiscale geometric analysis, surface inspection, Tetrolet transform

中图分类号: