Full metadata record
| DC Field | Value | Language |
|---|---|---|
| dc.contributor | School of Fashion and Textiles | en_US |
| dc.contributor.advisor | Wong, Calvin (SFT) | en_US |
| dc.creator | Tan, Pengjie | - |
| dc.identifier.uri | https://theses.lib.polyu.edu.hk/handle/200/13833 | - |
| dc.language | English | en_US |
| dc.publisher | Hong Kong Polytechnic University | en_US |
| dc.rights | All rights reserved | en_US |
| dc.title | Computer vision based method for density measurement and defect detection for woven fabric | en_US |
| dcterms.abstract | In recent decades, the seamless integration of industry and computer image technology has achieved breakthrough outcomes. Among them, how to use computer vision (CV) technology to measure the density of woven fabric and detect the defects of woven fabric has always been a hot topic that needs to be addressed. In the textile and apparel industry, density measurement and defect detection play an important role; because they not only directly affect the quality of woven fabric, but are also closely related to the comfort and durability of clothing. Although related CV research aims to solve various problems in the woven fabric manufacturing process, research focused on developing CV models for density measurement and defect detection of different woven fabrics is still limited. | en_US |
| dcterms.abstract | There are still three limitations to current research on density measurement and defect detection of woven fabrics. 1)Current yarn detection methods based on neural networks cannot accurately locate bent yarns because they usually use quadrilaterals or rectangles to mark each yarn and generate labels with yarn regions for the training model. In addition, when the yarn in the woven fabric is bent or deformed, the current projection-based post-processing algorithm may produce multiple peak points on one yarn, thereby reducing the density measurement performance. 2) There is a lack of large-scale woven fabric density measurement data sets in existing research, which greatly limits the development of CV algorithms in the field of density measurement of woven fabrics. 3) Merely measuring the density of woven fabrics cannot guarantee their quality, defects must also be detected. Defect detection is an important and highly valuable area of research because it allows tracking of the causes of defects during fabric manufacturing. Although there are some supervised methods, the performance of these methods largely depends on the number and quality of defective samples in the training set, making it difficult to be widely applied in various industrial scenarios. To better utilize defect-free samples, many unsupervised defect detection methods have been proposed, but these methods require training a dedicated model for each category of fabric and require a large number of samples to train the model. | en_US |
| dcterms.abstract | This study develops an intelligent CV algorithm to solve density measurement and defect detection problems in the textile and apparel industry. This model proposes corresponding solutions to the limitations of existing methods. First, to solve the problems existing in the existing woven fabric density measurement methods, an arbitrary yarn detection network named AYDNet is proposed. Specifically, AYDNet uses polygonal regions to represent yarn of arbitrary shapes and generate labels within the polygonal region for the training model, which enables it to accurately detect yarn of various shapes such as deformation and bending. To further accurately measure the density of the woven fabric, AYDNet only intercepts the local yarn region and uses the approximate distance between the yarn center to remove the multiple peak points formed by the same yarn. Secondly, to verify the effectiveness of AYDNet and promote the development of CV algorithms in the field of woven fabric density measurement, this study constructs a data set including various types and colors of woven fabrics to evaluate the density measurement performance of different algorithms. The experimental results on the self-built dataset show that AYDNet outperforms the current state-of-the-art methods in terms of accuracy. Third, to address the problems existing in unsupervised defect detection, an algorithm with one model for all categories (OMAC) is proposed. Specifically, to reduce the need for a large number of samples in existing methods, this study proposes a representative-based sample selection algorithm to select representative samples for training the model. At the same time, a lightweight feature extractor is designed to extract patch features and global features to reduce the cost of model training and inference. In addition, a dual-memory bank algorithm based on a query mechanism is introduced so that one model can detect all categories of fabrics. Extensive experimental results on public datasets demonstrate that OMAC outperforms other state-of-the-art algorithms. Furthermore, OMAC can achieve a high frame rate of up to 58 FPS on the 3090 GPU, meeting the requirements of real factories. | en_US |
| dcterms.abstract | Overall, the research in this thesis enhances the understanding of density measurement of woven fabrics and provides new ideas for research on fabric defect detection. The proposed AYDNet is capable of processing not only woven fabrics with various complex patterns but also fabrics with curved yarns. Furthermore, the proposed OMAC solves the problem of training a separate model for each fabric in unsupervised defect detection and improves detection efficiencies. This study also constructs two datasets for training and evaluating the performance of the proposed density measurement algorithm and defect detection. Due to the complexity of fabric images, It is necessary to solve two main problems in the future. First, a large-scale density measurement dataset of various types of fabric should be constructed. Second, developing a more powerful defect detection model that can cope with various colors and types of fabrics. | en_US |
| dcterms.extent | xvi, 114 pages : color illustrations | en_US |
| dcterms.isPartOf | PolyU Electronic Theses | en_US |
| dcterms.issued | 2025 | en_US |
| dcterms.educationalLevel | Ph.D. | en_US |
| dcterms.educationalLevel | All Doctorate | en_US |
| dcterms.LCSH | Textured woven fabrics -- Defects | en_US |
| dcterms.LCSH | Computer vision -- Industrial applications | en_US |
| dcterms.LCSH | Hong Kong Polytechnic University -- Dissertations | en_US |
| dcterms.accessRights | open access | en_US |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/13833