上海大学学报(自然科学版) ›› 2019, Vol. 25 ›› Issue (4): 445-452.doi: 10.12066/j.issn.1007-2861.2018

• 研究论文 • 上一篇    下一篇

基于深度学习的连铸坯表面缺陷检测

胡嘉成1, 王向阳1(), 刘晗2   

  1. 1. 上海大学 通信与信息工程学院, 上海 200444
    2. 上海金艺检测技术有限公司, 上海 201900
  • 收稿日期:2017-12-18 出版日期:2019-08-30 发布日期:2019-09-04
  • 通讯作者: 王向阳 E-mail:wangxiangyang@shu.edu.cn

Defect detection of continuous casting slabs based on deep learning

Jiacheng HU1, Xiangyang WANG1(), Han LIU2   

  1. 1. School of Communication and Information Engineering, Shanghai University, Shanghai 200444, China
    2. Shanghai Jinyi Inspection Technology Co., Ltd., Shanghai 201900, China
  • Received:2017-12-18 Online:2019-08-30 Published:2019-09-04
  • Contact: Xiangyang WANG E-mail:wangxiangyang@shu.edu.cn

摘要:

采用提取图像的纹理、几何特征并利用支持向量机(support vector machine, SVM)进行检测和识别的方法, 对宝山钢铁现有的连铸坯表面裂纹、凹陷、夹杂物、气孔、划痕等缺陷进行分析, 缺陷检测准确率为 83${\%}$. 提出一种基于卷积神经网络(convolutional neural network, CNN)的方法进行缺陷检测. 该方法对裂纹缺陷的检测准确率为 93${\%}$, 对其他缺陷(由于凹陷、夹杂物、气孔、划痕等缺陷数据较少, 这些缺陷归为一类)的检测准确率为 88${\%}$. 实验结果表明, 采用深度学习的方法 能够有效检测、识别出具有缺陷的连铸坯, 检测准确率较高.

关键词: 缺陷检测, 支持向量机, 深度学习

Abstract:

For analyzing surface defect such as surface cracks, depressions, inclusions, blow hole and scratches of continuous casting slabs, the support vector machine (SVM) can be used, and this is realized through identifying with the image texture and geometric features. The accuracy with such method is about 83${\%}$. Convolutional neural network (CNN) is used to propose a method of defect detection. The crack defect detection accuracy is 93${\%}$, and the detection accuracy for other defects (like depression, inclusions, blow hole and scratches when they are grouped together) is 88${\%}$. Experimental results show that deep learning is an effective method for detecting and identifying the defective continuous casting slabs, and the accuracy is high.

Key words: defect detection, support vector machine (SVM), deep learning

中图分类号: