A novel micro-defect classification system based on attention enhancement

Song Lin, Zhiyong He () and Lining Sun
Additional contact information
Song Lin: Harbin Institute of Technology
Zhiyong He: Harbin Institute of Technology
Lining Sun: Harbin Institute of Technology

Journal of Intelligent Manufacturing, 2024, vol. 35, issue 2, No 14, 703-726

Abstract: Abstract A surface micro-defect is characterized by a small size and a susceptibility to noise. Micro-defect detection and classification is very challenging. This paper proposes a Micro-defect classification system based on attention enhancement (MDCS) for solving the detection and classification of micro-defects. We combine defect detection with defect classification in MDSC. Micro-defects classification can be better realized based on the auxiliary task of defect detection. In this system, the aim of attention formation in bionic vision is to guide the system to focus on the target by zooming in and out on micro-defects. To avoid noise interference, an attention module based on trilinear feature confluence has been incorporated. Last but not least, the enhancement process based on the attention map improves the classification ability of micro-defects. As part of comparative experiment, we analyzed data including 19,200 fabric images and 4,800 bamboo images. In the micro-defect classification experiment based on MDCS(ResNet-50), the accuracy of fabric data and bamboo data is 88.2% and 89.4% respectively. Compared with ResNet-50, the classification accuracy (64.8%, 67.7%) is improved by 23.4% and 21.7% respectively. In the object detection experiment of micro-defects based on MDCS (ResNet-50), the accuracy of fabric data and bamboo data is 65.1% mAPs and 63.3% mAPs respectively. Compared with HRDNet, the detection accuracy (59.6% mAPs, 52.2% mAPs) is improved by 5.5% mAPs and 11.1% mAPs respectively. Experimental results demonstrate that the proposed system can counteract the interference caused by noise in small object detection, localize micro-defects accurately, and improve micro-defect classification accuracy significantly.

Keywords: Convolutional neural network; Vision attention; Intelligent manufacturing; Defect classification; Deep learning (search for similar items in EconPapers)
Date: 2024
References: View references in EconPapers View complete reference list from CitEc
Citations:

Downloads: (external link)
http://link.springer.com/10.1007/s10845-022-02064-2 Abstract (text/html)
Access to the full text of the articles in this series is restricted.

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:spr:joinma:v:35:y:2024:i:2:d:10.1007_s10845-022-02064-2

Ordering information: This journal article can be ordered from
http://www.springer.com/journal/10845

DOI: 10.1007/s10845-022-02064-2

Access Statistics for this article

Journal of Intelligent Manufacturing is currently edited by Andrew Kusiak

More articles in Journal of Intelligent Manufacturing from Springer
Bibliographic data for series maintained by Sonal Shukla () and Springer Nature Abstracting and Indexing ().