Scholarlink

<p>Laser weld spot detection is vital in industrial manufacturing but faces challenges like multi-scale target recognition and imprecise bounding box localization. To address these issues, we propose the MMG-YOLO model, featuring a multi-scale attention fusion module (MSAFM) for better texture and scale recognition, and a ghost feature fusion module to enhance detection accuracy while reducing computational load. We also designed a novel regression loss function, MidInnerSIoU (MISIoU), to improve bounding box precision. For large-scale, high-resolution inspection on industrial production lines, MMG-YOLO can run in real time on GPUs and supports distributed scaling. It is suitable for large-scale inspection in actual production. Our enhanced model shows a 12.6% increase in accuracy, 5.7% improvement in recall, and 8.7% increase in F1 score. The mAP@50 increases by 5% and FPS increases 2.2, while parameters drop by 3.9M and GFLOPS decrease by 13.9.</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

Laser weld spot detection based on MMG-YOLO

  • 期刊论文

  • 发表日期：

    2025年10月18日

查看全文
  • Jianxin Feng,
  • Xinyu Zhao,
  • Jiahao Wang,
  • Zhiguo Liu,
  • Yuanming Ding

摘要

Laser weld spot detection is vital in industrial manufacturing but faces challenges like multi-scale target recognition and imprecise bounding box localization. To address these issues, we propose the MMG-YOLO model, featuring a multi-scale attention fusion module (MSAFM) for better texture and scale recognition, and a ghost feature fusion module to enhance detection accuracy while reducing computational load. We also designed a novel regression loss function, MidInnerSIoU (MISIoU), to improve bounding box precision. For large-scale, high-resolution inspection on industrial production lines, MMG-YOLO can run in real time on GPUs and supports distributed scaling. It is suitable for large-scale inspection in actual production. Our enhanced model shows a 12.6% increase in accuracy, 5.7% improvement in recall, and 8.7% increase in F1 score. The mAP@50 increases by 5% and FPS increases 2.2, while parameters drop by 3.9M and GFLOPS decrease by 13.9.