Scholarlink

<p>In the solid wood panel industry, combining lamella pieces of consistent quality is critical for final product performance. However, most manufacturers still rely on manual classification by quality control teams, a process that is labor-intensive, error-prone, and inefficient. This reliance often results in production delays, inconsistent quality, and even financial losses due to panels being sold below cost, ultimately eroding customer trust. To address these challenges, this study proposes an expert decision support system based on deep learning to automate lamella quality classification. A Mask R-CNN model with a ResNet-101 backbone was trained on 2972 beech wood images to detect key defects knots, ray cells, and cracks that determine lamella grading. The system then classifies lamellas into AA, BB, CC, and Crack categories, achieving an overall classification accuracy of 90.90%. This work introduces a novel approach, validated on a custom dataset, that reduces dependency on manual labor and significantly enhances efficiency and reliability in solid panel production.</p>

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

Deep Learning Based Defect Detection and Quality Classification on Lamella Pieces Used in Solid Wood Panel Production

  • 期刊论文

  • 发表日期：

    2025年10月22日

查看全文
  • Merve Ozkan,
  • Caner Ozcan,
  • Mahmut Selman Gokmen

摘要

In the solid wood panel industry, combining lamella pieces of consistent quality is critical for final product performance. However, most manufacturers still rely on manual classification by quality control teams, a process that is labor-intensive, error-prone, and inefficient. This reliance often results in production delays, inconsistent quality, and even financial losses due to panels being sold below cost, ultimately eroding customer trust. To address these challenges, this study proposes an expert decision support system based on deep learning to automate lamella quality classification. A Mask R-CNN model with a ResNet-101 backbone was trained on 2972 beech wood images to detect key defects knots, ray cells, and cracks that determine lamella grading. The system then classifies lamellas into AA, BB, CC, and Crack categories, achieving an overall classification accuracy of 90.90%. This work introduces a novel approach, validated on a custom dataset, that reduces dependency on manual labor and significantly enhances efficiency and reliability in solid panel production.