Scholarlink

<p>The detection of surface defects in industrial products is crucial for ensuring quality. Enhancing detection accuracy of low-pixel-resolution defects amidst background interference while maintaining real-time performance poses a significant challenge. To address this, the CEYolo model based on Yolov8 introduces three key improvements: replacing the C2f module with an R2FN module to effectively reduce computational cost and enhance inference speed; integrating a Contextual Enhanced Transformer Attention (CoETA) mechanism to improve feature reuse and fusion, minimizing defect information loss during upsampling; and incorporating an enhanced Dynamic Head block (Dyhead) to increase focus on critical target areas, thus improving prediction accuracy. Case studies on NEU-DET and PCB-DET datasets demonstrate that CEYolo achieves mAPs of 80.9% and 95.8%, respectively, outperforming the baseline by 4.1% and 2.5%, with reduced parameters, GFLOPs, and increased FPS, showcasing its superior performance.</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

An industrial small surface defect detection technique based on CEYolo model

  • 期刊论文

  • 发表日期：

    2025年8月22日

查看全文
  • Biao Li,
  • Bing Wang,
  • Xiong Hu

摘要

The detection of surface defects in industrial products is crucial for ensuring quality. Enhancing detection accuracy of low-pixel-resolution defects amidst background interference while maintaining real-time performance poses a significant challenge. To address this, the CEYolo model based on Yolov8 introduces three key improvements: replacing the C2f module with an R2FN module to effectively reduce computational cost and enhance inference speed; integrating a Contextual Enhanced Transformer Attention (CoETA) mechanism to improve feature reuse and fusion, minimizing defect information loss during upsampling; and incorporating an enhanced Dynamic Head block (Dyhead) to increase focus on critical target areas, thus improving prediction accuracy. Case studies on NEU-DET and PCB-DET datasets demonstrate that CEYolo achieves mAPs of 80.9% and 95.8%, respectively, outperforming the baseline by 4.1% and 2.5%, with reduced parameters, GFLOPs, and increased FPS, showcasing its superior performance.