Scholarlink

<p>To improve the efficiency and reliability of pipeline inner-wall defect inspection, this study proposes an enhanced YOLOv8n-based detection method. A synergistic coordinate and attention module (SCAM), integrating ECoordA and L-SimAM, is introduced to strengthen spatial localization and salient defect representation. A hybrid CIoU–NWD loss is further employed to improve the localization of small and low-contrast defects. On the ROC-DET dataset, the proposed YOLOv8n + SCAM model achieved a Precision of 88.3%, a Recall of 76.2%, an mAP@0.5 of 79.2%, and an mAP@0.5:0.95 of 60.6%, outperforming the baseline YOLOv8n by 3.5 and 2.7 percentage points in the two mAP metrics. On NEU-DET, the model achieved an mAP@0.5 of 79.9%. An integrated pipeline inspection crawler was also validated in an industrial production environment, reducing the average inspection time by approximately 10&#xa0;min per pipeline. These results demonstrate the effectiveness and practical applicability of the proposed method for intelligent pipeline defect 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

A synergistic coordinate and attention module for pipeline weld surface defect detection

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年8月1日

查看全文
  • Dingyu Sun,
  • Zhongguo Li,
  • Zheng Zhang,
  • Qi Wang,
  • Junlong Wang

摘要

To improve the efficiency and reliability of pipeline inner-wall defect inspection, this study proposes an enhanced YOLOv8n-based detection method. A synergistic coordinate and attention module (SCAM), integrating ECoordA and L-SimAM, is introduced to strengthen spatial localization and salient defect representation. A hybrid CIoU–NWD loss is further employed to improve the localization of small and low-contrast defects. On the ROC-DET dataset, the proposed YOLOv8n + SCAM model achieved a Precision of 88.3%, a Recall of 76.2%, an mAP@0.5 of 79.2%, and an mAP@0.5:0.95 of 60.6%, outperforming the baseline YOLOv8n by 3.5 and 2.7 percentage points in the two mAP metrics. On NEU-DET, the model achieved an mAP@0.5 of 79.9%. An integrated pipeline inspection crawler was also validated in an industrial production environment, reducing the average inspection time by approximately 10 min per pipeline. These results demonstrate the effectiveness and practical applicability of the proposed method for intelligent pipeline defect inspection.