Scholarlink

Surface defect detection is a critical task in industrial production, involving the identification of defects in material surface images through advanced algorithms. While existing segmentation-based approaches struggle with issues such as weak anti-interference, susceptibility to noise, and poor performance in detecting minor defects, this paper introduces an enhanced approach using the R2U-Net framework combined with an attention mechanism. Our model integrates recurrent convolutional layers, residual connections, and attention gates within the U-Net framework to improve feature extraction, gradient flow, and focus on relevant regions. These enhancements allow for more accurate detection of small defects by iteratively refining features, improving network depth, and dynamically suppressing background noise. Compared to existing models, our approach demonstrates superior accuracy and precision in detecting small surface defects, making it a valuable contribution to industrial computer vision applications.

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

Recursive Residual Convolutional Neural Network with Attention Mechanism Based on U-Net (Attention R2U-Net) for Surface Minor Defect Detection

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Yongwen Zhang,
  • Jun Zhou,
  • Dongbin Ji,
  • Xin Wen

摘要

Surface defect detection is a critical task in industrial production, involving the identification of defects in material surface images through advanced algorithms. While existing segmentation-based approaches struggle with issues such as weak anti-interference, susceptibility to noise, and poor performance in detecting minor defects, this paper introduces an enhanced approach using the R2U-Net framework combined with an attention mechanism. Our model integrates recurrent convolutional layers, residual connections, and attention gates within the U-Net framework to improve feature extraction, gradient flow, and focus on relevant regions. These enhancements allow for more accurate detection of small defects by iteratively refining features, improving network depth, and dynamically suppressing background noise. Compared to existing models, our approach demonstrates superior accuracy and precision in detecting small surface defects, making it a valuable contribution to industrial computer vision applications.