Scholarlink

<p>In industrial manufacturing, precise and efficient detection of surface defects is crucial for ensuring product quality. This paper introduces MCAM-Net, a novel multi-scale convolutional attention network designed to address challenges in low-contrast defect detection. By integrating a channel-prioritized multi-scale subspace attention (CPMSSA) mechanism and a multi-scale convolutional block (MSCBlock), MCAM-Net enhances feature extraction and receptive field adaptability. Furthermore, the MCAM-Net model integrates the AKConv module, which improves the detection accuracy while reducing the number of parameters. Experimental results on the NEU-DET dataset demonstrate an mAP of 83.3%, outperforming existing benchmarks. On the GC10-DET dataset, MCAM-Net shows a 7.2% improvement in mAP, highlighting its effectiveness and speed in industrial surface defect detection. Here, we show that MCAM-Net offers a robust solution for enhancing defect detection accuracy in complex industrial environments. The code and model are available at <a href="https://github.com/zhangqianguang/MCAM-Net.">https://github.com/zhangqianguang/MCAM-Net.</a></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

MCAM-Net: multi-scale convolutional attention for enhanced industrial surface defect detection

  • 期刊论文

  • 发表日期：

    2026年2月26日

查看全文
  • Qianguang Zhang,
  • Xiangjun Dong,
  • Jianbin Xiong,
  • Hongbin Zhu,
  • Qi Wang,
  • Jing Wang,
  • Weikun Dai

摘要

In industrial manufacturing, precise and efficient detection of surface defects is crucial for ensuring product quality. This paper introduces MCAM-Net, a novel multi-scale convolutional attention network designed to address challenges in low-contrast defect detection. By integrating a channel-prioritized multi-scale subspace attention (CPMSSA) mechanism and a multi-scale convolutional block (MSCBlock), MCAM-Net enhances feature extraction and receptive field adaptability. Furthermore, the MCAM-Net model integrates the AKConv module, which improves the detection accuracy while reducing the number of parameters. Experimental results on the NEU-DET dataset demonstrate an mAP of 83.3%, outperforming existing benchmarks. On the GC10-DET dataset, MCAM-Net shows a 7.2% improvement in mAP, highlighting its effectiveness and speed in industrial surface defect detection. Here, we show that MCAM-Net offers a robust solution for enhancing defect detection accuracy in complex industrial environments. The code and model are available at https://github.com/zhangqianguang/MCAM-Net.