Scholarlink

<p>Reliable defect detection plays a vital role in preserving the quality and safety of strip steel. Traditional manual inspection is time-consuming, labor-intensive, and prone to inconsistencies, while many existing automated approaches suffer from high false detection rate, high miss rate, and slow processing speed. To address the challenges, this paper proposes SCSP-YOLO, an improved YOLOv5s-based algorithm for strip surface defect detection. First, a CSPDConv module has been introduced to substitute the C3 module in YOLOv5’s backbone feature extraction, improving performance, which enhances the feature extraction capability in low-resolution images. Second, the improved spatial pyramid pooling-fast structure broadens the sensory field through multi-gradient flow and effectively preserves semantic information across different scales, thereby improving the performance of multiscale feature fusion. Lastly, designed to address poor-quality anchor frames, the Wise-IoU loss function improves both stability and accuracy in detection tasks. On the NEU-DET dataset, SCSP-YOLO demonstrates an average precision of 80.1%, marking a 5.2% improvement over YOLOv5s.The frames per second reaches 104, which balances between the accuracy and speed of the algorithm. Therefore, the proposed algorithm can better satisfy the needs of practical industrial inspection.</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

SCSP-YOLO: a method for detecting surface defects on strip steel

  • 期刊论文

  • 发表日期：

    2025年8月9日

查看全文
  • Shiqi Jiang,
  • Yuebing Xu,
  • Hainan Lu,
  • Shuai Wang

摘要

Reliable defect detection plays a vital role in preserving the quality and safety of strip steel. Traditional manual inspection is time-consuming, labor-intensive, and prone to inconsistencies, while many existing automated approaches suffer from high false detection rate, high miss rate, and slow processing speed. To address the challenges, this paper proposes SCSP-YOLO, an improved YOLOv5s-based algorithm for strip surface defect detection. First, a CSPDConv module has been introduced to substitute the C3 module in YOLOv5’s backbone feature extraction, improving performance, which enhances the feature extraction capability in low-resolution images. Second, the improved spatial pyramid pooling-fast structure broadens the sensory field through multi-gradient flow and effectively preserves semantic information across different scales, thereby improving the performance of multiscale feature fusion. Lastly, designed to address poor-quality anchor frames, the Wise-IoU loss function improves both stability and accuracy in detection tasks. On the NEU-DET dataset, SCSP-YOLO demonstrates an average precision of 80.1%, marking a 5.2% improvement over YOLOv5s.The frames per second reaches 104, which balances between the accuracy and speed of the algorithm. Therefore, the proposed algorithm can better satisfy the needs of practical industrial inspection.