A feature extraction and classification algorithm based on improved sparse auto-encoder for round steel surface defects
Traditional feature dimensionality reduction (FDR) algorithms can extract features by reducing feature dimensions. However, it may lose some useful information and affect the accuracy of classification. Normally, in traditional defect feature extraction, it first obtain the defect area of the defect...
Main Authors: | , |
---|---|
Format: | Article |
Language: | English |
Published: |
AIMS Press
2020-08-01
|
Series: | Mathematical Biosciences and Engineering |
Subjects: | |
Online Access: | https://aimspress.com/article/doi/10.3934/mbe.2020290?viewType=HTML |
id |
doaj-91c17e0f5117439492c02c7e2507a240 |
---|---|
record_format |
Article |
spelling |
doaj-91c17e0f5117439492c02c7e2507a2402021-09-14T01:35:28ZengAIMS PressMathematical Biosciences and Engineering1551-00182020-08-011755369539410.3934/mbe.2020290A feature extraction and classification algorithm based on improved sparse auto-encoder for round steel surface defectsXuguo Yan0Liang Gao1State Key Lab of Digital Manufacturing Equipment and Technology, School of Mechanical Science and Engineering, Huazhong University of Science and Technology, Wuhan 430074, ChinaState Key Lab of Digital Manufacturing Equipment and Technology, School of Mechanical Science and Engineering, Huazhong University of Science and Technology, Wuhan 430074, ChinaTraditional feature dimensionality reduction (FDR) algorithms can extract features by reducing feature dimensions. However, it may lose some useful information and affect the accuracy of classification. Normally, in traditional defect feature extraction, it first obtain the defect area of the defect image by image preprocessing and defect segmentation, select the original feature set of defects by prior knowledge, and extract the optimal features by traditional FDR algorithms to solve the problem of "curse of dimensionality". In this paper, a feature extraction and classification algorithm based on improved sparse auto-encoder (AE) is proposed. We adopt three traditional FDR algorithms at the same time, combine the defect features obtained in pairs, take the merged defect features as the input of sparse AE, then use the "bottleneck" of sparse AE to conduct the defects classification by Softmax classifier. The experimental results show that the proposed algorithm can extract the optimal features of round steel surface defects with less network training time than individual sparse AE, finally get higher classification accuracy than individual FDR algorithm in the actual production line.https://aimspress.com/article/doi/10.3934/mbe.2020290?viewType=HTMLfeature extractionfeature dimensionality reductionauto-encoderdefect detectionround steel |
collection |
DOAJ |
language |
English |
format |
Article |
sources |
DOAJ |
author |
Xuguo Yan Liang Gao |
spellingShingle |
Xuguo Yan Liang Gao A feature extraction and classification algorithm based on improved sparse auto-encoder for round steel surface defects Mathematical Biosciences and Engineering feature extraction feature dimensionality reduction auto-encoder defect detection round steel |
author_facet |
Xuguo Yan Liang Gao |
author_sort |
Xuguo Yan |
title |
A feature extraction and classification algorithm based on improved sparse auto-encoder for round steel surface defects |
title_short |
A feature extraction and classification algorithm based on improved sparse auto-encoder for round steel surface defects |
title_full |
A feature extraction and classification algorithm based on improved sparse auto-encoder for round steel surface defects |
title_fullStr |
A feature extraction and classification algorithm based on improved sparse auto-encoder for round steel surface defects |
title_full_unstemmed |
A feature extraction and classification algorithm based on improved sparse auto-encoder for round steel surface defects |
title_sort |
feature extraction and classification algorithm based on improved sparse auto-encoder for round steel surface defects |
publisher |
AIMS Press |
series |
Mathematical Biosciences and Engineering |
issn |
1551-0018 |
publishDate |
2020-08-01 |
description |
Traditional feature dimensionality reduction (FDR) algorithms can extract features by reducing feature dimensions. However, it may lose some useful information and affect the accuracy of classification. Normally, in traditional defect feature extraction, it first obtain the defect area of the defect image by image preprocessing and defect segmentation, select the original feature set of defects by prior knowledge, and extract the optimal features by traditional FDR algorithms to solve the problem of "curse of dimensionality". In this paper, a feature extraction and classification algorithm based on improved sparse auto-encoder (AE) is proposed. We adopt three traditional FDR algorithms at the same time, combine the defect features obtained in pairs, take the merged defect features as the input of sparse AE, then use the "bottleneck" of sparse AE to conduct the defects classification by Softmax classifier. The experimental results show that the proposed algorithm can extract the optimal features of round steel surface defects with less network training time than individual sparse AE, finally get higher classification accuracy than individual FDR algorithm in the actual production line. |
topic |
feature extraction feature dimensionality reduction auto-encoder defect detection round steel |
url |
https://aimspress.com/article/doi/10.3934/mbe.2020290?viewType=HTML |
work_keys_str_mv |
AT xuguoyan afeatureextractionandclassificationalgorithmbasedonimprovedsparseautoencoderforroundsteelsurfacedefects AT lianggao afeatureextractionandclassificationalgorithmbasedonimprovedsparseautoencoderforroundsteelsurfacedefects AT xuguoyan featureextractionandclassificationalgorithmbasedonimprovedsparseautoencoderforroundsteelsurfacedefects AT lianggao featureextractionandclassificationalgorithmbasedonimprovedsparseautoencoderforroundsteelsurfacedefects |
_version_ |
1717380148280426496 |