Scholarlink

<p>To address the issues of omission, and low algorithmic accuracy in detecting surface defects of hot-rolled strip steel in complex backgrounds, this paper proposes the PMSE-YOLO surface defect detection algorithm. First, to enhance the model’s ability to extract multi-scale features, the CSP_PMS structure is designed to optimize the C2f structure in the Backbone, improving feature extraction for multi-scale targets. Second, to preserve the semantic information of small targets in the Neck, the EfficientRepBiPAN structure is adopted, utilizing cross-layer connections to enhance multi-scale feature representation and achieve small target feature fusion. Finally, the Wise-ShapeIoU loss function is incorporated to enhance the model’s detection performance. Experimental validation on the NEU-DET dataset demonstrates that PMSE-YOLO reduces parameter by 17% and computational cost by 18.5%, while improving mAP@0.5 by 3.1 to 82.2% compared to baseline network. PMSE-YOLO balances lightweight design and real-time performance while enhancing surface defect detection accuracy for hot-rolled strips, facilitating deployment on edge devices. Furthermore, experimental results on the GC10-DET dataset confirm the superior generalization capability of the proposed model.</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

PMSE-YOLO: an efficient framework for detecting surface defects of hot-rolled strip steel

  • 期刊论文

  • 发表日期：

    2025年5月12日

查看全文
  • Mengran Zhou,
  • Ruiyan Wang,
  • Yue Chen

摘要

To address the issues of omission, and low algorithmic accuracy in detecting surface defects of hot-rolled strip steel in complex backgrounds, this paper proposes the PMSE-YOLO surface defect detection algorithm. First, to enhance the model’s ability to extract multi-scale features, the CSP_PMS structure is designed to optimize the C2f structure in the Backbone, improving feature extraction for multi-scale targets. Second, to preserve the semantic information of small targets in the Neck, the EfficientRepBiPAN structure is adopted, utilizing cross-layer connections to enhance multi-scale feature representation and achieve small target feature fusion. Finally, the Wise-ShapeIoU loss function is incorporated to enhance the model’s detection performance. Experimental validation on the NEU-DET dataset demonstrates that PMSE-YOLO reduces parameter by 17% and computational cost by 18.5%, while improving mAP@0.5 by 3.1 to 82.2% compared to baseline network. PMSE-YOLO balances lightweight design and real-time performance while enhancing surface defect detection accuracy for hot-rolled strips, facilitating deployment on edge devices. Furthermore, experimental results on the GC10-DET dataset confirm the superior generalization capability of the proposed model.