Improving Fabric Defect Classification via Automatic Mining and Leveraging of Category Hierarchy
IEEE Transactions on Instrumentation and Measurement · 2026 · 75 · 1-15
DOI: 10.1109/tim.2026.3671939 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Instrumentation and Measurement, 2026. Verfasst von Xudong Duan und 5 weiteren Autor:innen. Die Arbeit nennt 97 Literaturangaben.
Abstract (Originalsprache)
Fabric defect classification is crucial in textile quality inspection, enabling category-specific defect handling. Although dozens of defect categories have been defined, annotation granularity in practice often ranges from binary (defect versus nondefect) to fine-grained multiclass classification. Typically, the annotated defect categories can be structured hierarchically or further subdivided based on domain knowledge or visual patterns. In this study, we propose a method to improve the deep classifier for the annotated categories by mining subcategories and leveraging constructed category hierarchies. Based on the annotated categories, we introduce two primary strategies: 1) mining subcategories to generate fine-grained pseudo-labels and 2) leveraging coarse-level hierarchies to create auxiliary coarse labels. By incorporating these two lightweight auxiliary tasks, the proposed method can improve defect classification and facilitate defect detection. Furthermore, we collect and release a fabric defect dataset from two factories, containing 24 annotated categories. Each sample is captured under three different illumination conditions: back, front, and side lighting. Extensive experiments on public datasets, including TILDA, ZJU-Leaper, Tianchi, MVTec-AD, and our dataset, validate the effectiveness of the proposed method, which consistently achieves improvements of over 1% in both the classification accuracy and mean average precision (mAP)@0.5.
Zitieren
APA
Duan, X., Lu, W., Cai, W., Tong, F., Mao, Y., & Xu, W. (2026). Improving Fabric Defect Classification via Automatic Mining and Leveraging of Category Hierarchy. IEEE Transactions on Instrumentation and Measurement, 75, 1-15. https://doi.org/10.1109/tim.2026.3671939
Vancouver
Duan X, Lu W, Cai W, Tong F, Mao Y, Xu W. Improving Fabric Defect Classification via Automatic Mining and Leveraging of Category Hierarchy. IEEE Transactions on Instrumentation and Measurement. 2026;75:1-15. doi:10.1109/tim.2026.3671939
BibTeX
@article{Duan2026,
author = {Duan, Xudong and Lu, Wenkai and Cai, Wenchao and Tong, Fei and Mao, Yu and Xu, Weishi},
title = {Improving Fabric Defect Classification via Automatic Mining and Leveraging of Category Hierarchy},
journal = {IEEE Transactions on Instrumentation and Measurement},
year = {2026},
volume = {75},
pages = {1--15},
doi = {10.1109/tim.2026.3671939}
}
EUVIMED