Scholarlink

In the contemporary world of rapid textile advancement, identifying fabric defects is crucial to combat economic loss and to ensure fabric quality benchmarks. The majority of previous studies have been centered around the use of anchor-based methods. However, determining the appropriate dimension of an anchor box is difficult due to the vast variety of fabric defect sizes. To tackle this problem, we have put forth an alternative solution by employing an anchor-free approach. This approach is based on the capabilities of advanced YOLOv8 models to effectively detect fabric defects without relying on predefined anchor boxes. To achieve this, we have explored multiple pre-trained architectures, including YOLOv8n, YOLOv8m, YOLOv8s, YOLOv8l, and YOLOv8x.The YOLOv8x architecture outperformed other models with a remarkable Mean Average Precision (mAP) of 97.9%, a classification accuracy of 96.7%, and an average detection time of 22.2 ms.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhancing Textile Quality Assurance: An Advanced Deep Learning Approach for Efficient Fabric Defect Detection

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Sadia Rahman,
  • Sharmistha Chanda Tista,
  • Md Nazmul Hoq

摘要

In the contemporary world of rapid textile advancement, identifying fabric defects is crucial to combat economic loss and to ensure fabric quality benchmarks. The majority of previous studies have been centered around the use of anchor-based methods. However, determining the appropriate dimension of an anchor box is difficult due to the vast variety of fabric defect sizes. To tackle this problem, we have put forth an alternative solution by employing an anchor-free approach. This approach is based on the capabilities of advanced YOLOv8 models to effectively detect fabric defects without relying on predefined anchor boxes. To achieve this, we have explored multiple pre-trained architectures, including YOLOv8n, YOLOv8m, YOLOv8s, YOLOv8l, and YOLOv8x.The YOLOv8x architecture outperformed other models with a remarkable Mean Average Precision (mAP) of 97.9%, a classification accuracy of 96.7%, and an average detection time of 22.2 ms.