Scholarlink

<p>Due to imperfections in the welding process and external factors, weld defects can significantly impact the lifespan and reliability of equipment. Therefore, weld defect detection is a crucial step in industrial production. However, traditional weld defect detection algorithms suffer from low accuracy and speed. To address these issues, we propose GMVG-Net, a model based on YOLOv5, for efficient weld defect detection. First, we design the MSFEF module to extract features at multiple levels, then integrate the Ghost Bottleneck into the backbone network to reduce model parameters while maintaining accuracy. Next, we use the lightweight VoVGSCSP module in the neck and replace the original C3 with the GSConv module. This modification reduces model parameters and accelerates inference speed without compromising training performance. Finally, extensive experimental results demonstrate that GMVG-Net achieves a detection accuracy of 73.4 mAP for steel plate weld defects, which is 11.1% higher than YOLOv5. Our proposed model demonstrates outstanding performance in weld defect detection.</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

GMVG-Net: lightweight mul tiscale feature fusion network for welding defect detection

  • 期刊论文

  • 发表日期：

    2025年4月2日

查看全文
  • Kun Song,
  • Qiangxian Huang

摘要

Due to imperfections in the welding process and external factors, weld defects can significantly impact the lifespan and reliability of equipment. Therefore, weld defect detection is a crucial step in industrial production. However, traditional weld defect detection algorithms suffer from low accuracy and speed. To address these issues, we propose GMVG-Net, a model based on YOLOv5, for efficient weld defect detection. First, we design the MSFEF module to extract features at multiple levels, then integrate the Ghost Bottleneck into the backbone network to reduce model parameters while maintaining accuracy. Next, we use the lightweight VoVGSCSP module in the neck and replace the original C3 with the GSConv module. This modification reduces model parameters and accelerates inference speed without compromising training performance. Finally, extensive experimental results demonstrate that GMVG-Net achieves a detection accuracy of 73.4 mAP for steel plate weld defects, which is 11.1% higher than YOLOv5. Our proposed model demonstrates outstanding performance in weld defect detection.