A deep-learning-based approach for fast and robust steel surface defects classification

Guizhong Fu, Peize Sun, Wenbin Zhu, Jiangxin Yang, Yanlong Cao, Michael Ying Yang, Yanpeng Cao

Research output: Contribution to journalArticleAcademicpeer-review

Abstract

Automatic visual recognition of steel surface defects provides critical functionality to facilitate quality control of steel strip production. In this paper, we present a compact yet effective convolutional neural network (CNN)model, which emphasizes the training of low-level features and incorporates multiple receptive fields, to achieve fast and accurate steel surface defect classification. Our proposed method adopts the pre-trained SqueezeNet as the backbone architecture. It only requires a small amount of defect-specific training samples to achieve high-accuracy recognition on a diversity-enhanced testing dataset of steel surface defects which contains severe non-uniform illumination, camera noise, and motion blur. Moreover, our proposed light-weight CNN model can meet the requirement of real-time online inspection, running over 100 fps on a computer equipped with a single NVIDIA TITAN X Graphics Processing Unit (12G memory). Codes and a diversity-enhanced testing dataset will be made publicly available.

Original languageEnglish
Pages (from-to)397-405
Number of pages9
JournalOptics and Lasers in Engineering
Volume121
Early online date15 May 2019
DOIs
Publication statusPublished - 1 Oct 2019

Fingerprint

Steel
Surface defects
surface defects
learning
steels
education
Neural networks
Testing
quality control
Quality control
inspection
strip
Lighting
Inspection
Cameras
illumination
cameras
Data storage equipment
requirements
Defects

Keywords

  • Convolutional neural network
  • Defect classification
  • Feature extraction
  • Multi-receptive field
  • Surface inspection
  • ITC-ISI-JOURNAL-ARTICLE

Cite this

Fu, Guizhong ; Sun, Peize ; Zhu, Wenbin ; Yang, Jiangxin ; Cao, Yanlong ; Yang, Michael Ying ; Cao, Yanpeng. / A deep-learning-based approach for fast and robust steel surface defects classification. In: Optics and Lasers in Engineering. 2019 ; Vol. 121. pp. 397-405.
@article{d248ffac0f3c4309b75e158a8b916ab7,
title = "A deep-learning-based approach for fast and robust steel surface defects classification",
abstract = "Automatic visual recognition of steel surface defects provides critical functionality to facilitate quality control of steel strip production. In this paper, we present a compact yet effective convolutional neural network (CNN)model, which emphasizes the training of low-level features and incorporates multiple receptive fields, to achieve fast and accurate steel surface defect classification. Our proposed method adopts the pre-trained SqueezeNet as the backbone architecture. It only requires a small amount of defect-specific training samples to achieve high-accuracy recognition on a diversity-enhanced testing dataset of steel surface defects which contains severe non-uniform illumination, camera noise, and motion blur. Moreover, our proposed light-weight CNN model can meet the requirement of real-time online inspection, running over 100 fps on a computer equipped with a single NVIDIA TITAN X Graphics Processing Unit (12G memory). Codes and a diversity-enhanced testing dataset will be made publicly available.",
keywords = "Convolutional neural network, Defect classification, Feature extraction, Multi-receptive field, Surface inspection, ITC-ISI-JOURNAL-ARTICLE",
author = "Guizhong Fu and Peize Sun and Wenbin Zhu and Jiangxin Yang and Yanlong Cao and Yang, {Michael Ying} and Yanpeng Cao",
year = "2019",
month = "10",
day = "1",
doi = "10.1016/j.optlaseng.2019.05.005",
language = "English",
volume = "121",
pages = "397--405",
journal = "Optics and Lasers in Engineering",
issn = "0143-8166",
publisher = "Elsevier",

}

A deep-learning-based approach for fast and robust steel surface defects classification. / Fu, Guizhong; Sun, Peize; Zhu, Wenbin; Yang, Jiangxin; Cao, Yanlong; Yang, Michael Ying; Cao, Yanpeng.

In: Optics and Lasers in Engineering, Vol. 121, 01.10.2019, p. 397-405.

Research output: Contribution to journalArticleAcademicpeer-review

TY - JOUR

T1 - A deep-learning-based approach for fast and robust steel surface defects classification

AU - Fu, Guizhong

AU - Sun, Peize

AU - Zhu, Wenbin

AU - Yang, Jiangxin

AU - Cao, Yanlong

AU - Yang, Michael Ying

AU - Cao, Yanpeng

PY - 2019/10/1

Y1 - 2019/10/1

N2 - Automatic visual recognition of steel surface defects provides critical functionality to facilitate quality control of steel strip production. In this paper, we present a compact yet effective convolutional neural network (CNN)model, which emphasizes the training of low-level features and incorporates multiple receptive fields, to achieve fast and accurate steel surface defect classification. Our proposed method adopts the pre-trained SqueezeNet as the backbone architecture. It only requires a small amount of defect-specific training samples to achieve high-accuracy recognition on a diversity-enhanced testing dataset of steel surface defects which contains severe non-uniform illumination, camera noise, and motion blur. Moreover, our proposed light-weight CNN model can meet the requirement of real-time online inspection, running over 100 fps on a computer equipped with a single NVIDIA TITAN X Graphics Processing Unit (12G memory). Codes and a diversity-enhanced testing dataset will be made publicly available.

AB - Automatic visual recognition of steel surface defects provides critical functionality to facilitate quality control of steel strip production. In this paper, we present a compact yet effective convolutional neural network (CNN)model, which emphasizes the training of low-level features and incorporates multiple receptive fields, to achieve fast and accurate steel surface defect classification. Our proposed method adopts the pre-trained SqueezeNet as the backbone architecture. It only requires a small amount of defect-specific training samples to achieve high-accuracy recognition on a diversity-enhanced testing dataset of steel surface defects which contains severe non-uniform illumination, camera noise, and motion blur. Moreover, our proposed light-weight CNN model can meet the requirement of real-time online inspection, running over 100 fps on a computer equipped with a single NVIDIA TITAN X Graphics Processing Unit (12G memory). Codes and a diversity-enhanced testing dataset will be made publicly available.

KW - Convolutional neural network

KW - Defect classification

KW - Feature extraction

KW - Multi-receptive field

KW - Surface inspection

KW - ITC-ISI-JOURNAL-ARTICLE

UR - https://ezproxy2.utwente.nl/login?url=https://library.itc.utwente.nl/login/2019/isi/yang_dee.pdf

UR - https://ezproxy2.utwente.nl/login?url=https://doi.org/10.1016/j.optlaseng.2019.05.005

U2 - 10.1016/j.optlaseng.2019.05.005

DO - 10.1016/j.optlaseng.2019.05.005

M3 - Article

VL - 121

SP - 397

EP - 405

JO - Optics and Lasers in Engineering

JF - Optics and Lasers in Engineering

SN - 0143-8166

ER -