Scholarlink

<p>The expansion of the high-end product market in the tobacco industry and the increasing demand for product quality have posed significant challenges to tobacco online inspection technology. During the tobacco industrial production process, identifying defects in cigarette packaging and other areas is difficult. The current YOLO model lacks sufficient precision for detecting small target defects on high-precision tobacco production lines, making it unsuitable for the high-quality development of China’s tobacco industry. To address this issue, we propose an improved YOLOv9 (You Only Look Once version 9) model (SCS-YOLOv9) designed to enhance the detection performance for industrial tobacco surface defects. By introducing and integrating the SPD-Conv module, the improved algorithm’s ability to extract multi-scale and small defects on the tobacco surface has been significantly enhanced. Additionally, CARAFE technology replaces traditional upsampling methods, improving the accuracy and effectiveness of multi-scale feature fusion. The loss function is optimized to SIoU Gevorgyan (Siou loss: more powerful learning for bounding box regression. arXiv e-prints, 2022) Loss, enhancing both accuracy and convergence speed in bounding box prediction. Experimental results show that SCS-YOLOv9 significantly outperforms traditional detection methods, achieving an average precision of 97.6% and notable improvements in metrics such as accuracy, recall rate, and mAP@0.5. Experimental results show that our improved model provides an innovative solution for the quality detection of tobacco industry.</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

SCS-YOLOv9: a novel YOLOv9 framework for small object detection via edge computing

  • 期刊论文

  • 发表日期：

    2026年2月16日

查看全文
  • Xu Liu,
  • Gaofeng Wu,
  • Tao Zhou,
  • Ziyi Wang,
  • Kui Wang,
  • Jun Li,
  • Chong Xie,
  • Tingwen Liu

摘要

The expansion of the high-end product market in the tobacco industry and the increasing demand for product quality have posed significant challenges to tobacco online inspection technology. During the tobacco industrial production process, identifying defects in cigarette packaging and other areas is difficult. The current YOLO model lacks sufficient precision for detecting small target defects on high-precision tobacco production lines, making it unsuitable for the high-quality development of China’s tobacco industry. To address this issue, we propose an improved YOLOv9 (You Only Look Once version 9) model (SCS-YOLOv9) designed to enhance the detection performance for industrial tobacco surface defects. By introducing and integrating the SPD-Conv module, the improved algorithm’s ability to extract multi-scale and small defects on the tobacco surface has been significantly enhanced. Additionally, CARAFE technology replaces traditional upsampling methods, improving the accuracy and effectiveness of multi-scale feature fusion. The loss function is optimized to SIoU Gevorgyan (Siou loss: more powerful learning for bounding box regression. arXiv e-prints, 2022) Loss, enhancing both accuracy and convergence speed in bounding box prediction. Experimental results show that SCS-YOLOv9 significantly outperforms traditional detection methods, achieving an average precision of 97.6% and notable improvements in metrics such as accuracy, recall rate, and mAP@0.5. Experimental results show that our improved model provides an innovative solution for the quality detection of tobacco industry.