Scholarlink

<p>Bridge cracks, which are vital indicators of structural health, pose challenges in detection due to complex backgrounds and edge feature extraction difficulties. This study introduces CMC-Net, a dual-encoder network combining convolutional neural network (CNN) and Mamba for precise bridge crack segmentation. The CNN encoder, built with depthwise over-parameterized convolution, enhances local feature extraction, while the Mamba encoder captures global context information. A new feature fusion module (FFM) integrates these features, and the edge enhancement module (EEM) addresses edge feature extraction challenges. Experiments on CQBCD, DeepCrack, and CrackTree260 datasets show CMC-Net achieves MIoU of 89.47%, 87.36%, and 83.30%, respectively, significantly improving crack detection accuracy. Our code is available at <a href="https://github.com/DJ-hjk/CMCNet">https://github.com/DJ-hjk/CMCNet</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

Enhanced bridge crack segmentation via CNN–Mamba dual encoders with edge enhancement

  • 期刊论文

  • 发表日期：

    2026年6月22日

查看全文
  • Zhangli Lan,
  • Chuanhan He,
  • Hong Zhang,
  • Xin Ma,
  • Weihong Huang

摘要

Bridge cracks, which are vital indicators of structural health, pose challenges in detection due to complex backgrounds and edge feature extraction difficulties. This study introduces CMC-Net, a dual-encoder network combining convolutional neural network (CNN) and Mamba for precise bridge crack segmentation. The CNN encoder, built with depthwise over-parameterized convolution, enhances local feature extraction, while the Mamba encoder captures global context information. A new feature fusion module (FFM) integrates these features, and the edge enhancement module (EEM) addresses edge feature extraction challenges. Experiments on CQBCD, DeepCrack, and CrackTree260 datasets show CMC-Net achieves MIoU of 89.47%, 87.36%, and 83.30%, respectively, significantly improving crack detection accuracy. Our code is available at https://github.com/DJ-hjk/CMCNet.