ODCL: An Object Disentanglement and Contrastive Learning Model for Few-Shot Industrial Defect Detection

Li, G; Peng, F; Wu, Z; Wang, S; Xu, RYD

ODCL: An Object Disentanglement and Contrastive Learning Model for Few-Shot Industrial Defect Detection

Li, G Peng, F Wu, Z Wang, S Xu, RYD

Permalink

Publisher:: Institute of Electrical and Electronics Engineers (IEEE)
Publication Type:: Journal Article
Citation:: IEEE Sensors Journal, 2024, 24, (11), pp. 18568-18577
Issue Date:: 2024-06-01

Recently Added

	Filename	Description	Size
	1732536.pdf	Published version	3.82 MB	Adobe PDF	View/Open

Copyright Clearance Process

Recently Added
In Progress
Open Access

This item is new to OPUS and is not currently available.

Full metadata record

Field	Value	Language
dc.contributor.author	Li, G
dc.contributor.author	Peng, F
dc.contributor.author	Wu, Z
dc.contributor.author	Wang, S
dc.contributor.author	Xu, RYD
dc.date.accessioned	2024-08-07T05:47:09Z
dc.date.available	2024-08-07T05:47:09Z
dc.date.issued	2024-06-01
dc.identifier.citation	IEEE Sensors Journal, 2024, 24, (11), pp. 18568-18577
dc.identifier.issn	1530-437X
dc.identifier.issn	1558-1748
dc.identifier.uri	http://hdl.handle.net/10453/180353
dc.description.abstract	Deep learning methods have shown promising achievements, yet require a substantial quantity of training data. In industrial manufacturing scenarios, the training samples for certain defect categories might be small. Such a few-shot learning problem severely obstacles the application of deep learning. Moreover, challenges such as small targets that are scarcely distinguishable from the background, coupled with defect category confusion, further complicate defect detection. To address these issues, this study proposes a novel approach called the object disentanglement and contrastive learning (ODCL) model. First, we introduce a significant region disentanglement module to decouple the foreground from the background. This is the pioneering application of disentanglement in few-shot industrial defect detection. Subsequently, we advance a supervised contrastive learning (SCL) model to alleviate defect category confusion. At last, we resolve the few-shot learning through a two-stage fine-tuning (TFA) method. Experimental results on three industrial datasets demonstrate that the ODCL achieves state-of-the-art results in various few-shot scenarios. Code and data are available at https://github.com/LiBiGo/ODCL.
dc.language	en
dc.publisher	Institute of Electrical and Electronics Engineers (IEEE)
dc.relation.ispartof	IEEE Sensors Journal
dc.relation.isbasedon	10.1109/JSEN.2024.3388714
dc.rights	info:eu-repo/semantics/restrictedAccess
dc.subject	0205 Optical Physics, 0906 Electrical and Electronic Engineering, 0913 Mechanical Engineering
dc.subject.classification	Analytical Chemistry
dc.subject.classification	40 Engineering
dc.title	ODCL: An Object Disentanglement and Contrastive Learning Model for Few-Shot Industrial Defect Detection
dc.type	Journal Article
utslib.citation.volume	24
utslib.for	0205 Optical Physics
utslib.for	0906 Electrical and Electronic Engineering
utslib.for	0913 Mechanical Engineering
pubs.organisational-group	University of Technology Sydney
pubs.organisational-group	University of Technology Sydney/Faculty of Engineering and Information Technology
pubs.organisational-group	University of Technology Sydney/Strength - INEXT - Innovation in IT Services and Applications
pubs.organisational-group	University of Technology Sydney/Strength - GBDTC - Global Big Data Technologies
pubs.organisational-group	University of Technology Sydney/Faculty of Engineering and Information Technology/School of Electrical and Data Engineering
pubs.organisational-group	University of Technology Sydney/All Manual Groups
pubs.organisational-group	University of Technology Sydney/All Manual Groups/Global Big Data Technologies Research Centre (GBDTC)
utslib.copyright.status	recently_added	*
dc.date.updated	2024-08-07T05:47:05Z
pubs.issue	11
pubs.publication-status	Published
pubs.volume	24
utslib.citation.issue	11

Abstract:

Deep learning methods have shown promising achievements, yet require a substantial quantity of training data. In industrial manufacturing scenarios, the training samples for certain defect categories might be small. Such a few-shot learning problem severely obstacles the application of deep learning. Moreover, challenges such as small targets that are scarcely distinguishable from the background, coupled with defect category confusion, further complicate defect detection. To address these issues, this study proposes a novel approach called the object disentanglement and contrastive learning (ODCL) model. First, we introduce a significant region disentanglement module to decouple the foreground from the background. This is the pioneering application of disentanglement in few-shot industrial defect detection. Subsequently, we advance a supervised contrastive learning (SCL) model to alleviate defect category confusion. At last, we resolve the few-shot learning through a two-stage fine-tuning (TFA) method. Experimental results on three industrial datasets demonstrate that the ODCL achieves state-of-the-art results in various few-shot scenarios. Code and data are available at https://github.com/LiBiGo/ODCL.

Please use this identifier to cite or link to this item:

http://hdl.handle.net/10453/180353