Scholarlink

<p>To improve the accuracy of defect detection in industrial components, this paper proposed an efficient defect detection model based on the improved RT-DETR, called CEH-RTDETR. To address issues such as missed detections and low precision caused by varying scales of the same type of defect in industrial component defect detection, We introduced the Efficient Local Attention Mechanism(ELA) into the High-level Screening-feature Pyramid Network(HS-FPN) module and replaced the CCFM module in the RT-DETR model. Additionally, by introducing the Cascade Group Attention module in the AIFI module, we split the input features and provided only a portion of the input features to each attention head to reduce computational redundancy. The experimental results show that our improved RT-DETR algorithm performs well on the industrial component defect detection dataset, achieving performances of 72.6% and 55.9% on the mAP@0.5 and mAP0.5-0.95 metrics respectively. Compared with the original RT-DETR model, it has improved by 5.1% and 5.6% respectively.In addition, the parameters and GFLOPs are reduced by 1786108 and 3.5 respectively. This algorithm significantly reduces computational load and effectively improves the mean average precision for industrial components, alleviating issues such as missed detections due to varying defect scales.</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

CEH-RTDETR: A lightweight and efficient industrial component defect detection model based on improved RT-DETR

  • 期刊论文

  • 发表日期：

    2025年7月30日

查看全文
  • Leda Qu,
  • Feng Wen,
  • Haixin Huang,
  • Fang Peng

摘要

To improve the accuracy of defect detection in industrial components, this paper proposed an efficient defect detection model based on the improved RT-DETR, called CEH-RTDETR. To address issues such as missed detections and low precision caused by varying scales of the same type of defect in industrial component defect detection, We introduced the Efficient Local Attention Mechanism(ELA) into the High-level Screening-feature Pyramid Network(HS-FPN) module and replaced the CCFM module in the RT-DETR model. Additionally, by introducing the Cascade Group Attention module in the AIFI module, we split the input features and provided only a portion of the input features to each attention head to reduce computational redundancy. The experimental results show that our improved RT-DETR algorithm performs well on the industrial component defect detection dataset, achieving performances of 72.6% and 55.9% on the mAP@0.5 and mAP0.5-0.95 metrics respectively. Compared with the original RT-DETR model, it has improved by 5.1% and 5.6% respectively.In addition, the parameters and GFLOPs are reduced by 1786108 and 3.5 respectively. This algorithm significantly reduces computational load and effectively improves the mean average precision for industrial components, alleviating issues such as missed detections due to varying defect scales.