Related Experiment Video
Updated: May 22, 2025
12:08
From Voxels to Knowledge: A Practical Guide to the Segmentation of Complex Electron Microscopy 3D-Data
Published on: August 13, 2014
24.5K
VM-UNet++ research on crack image segmentation based on improved VM-UNet
Wenliang Tang1, Ziyi Wu2, Wei Wang1
1School of Information and Software Engineering, East China Jiaotong University, Nanchang, 330013, China.
Scientific Reports
|March 16, 2025
Summary
A new VM-UNet++ model enhances crack segmentation accuracy by integrating Mamba architecture, outperforming existing methods. This deep learning approach offers improved performance with lower computational costs for structural defect detection.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Structural Engineering
Background:
- Cracks are critical structural defects requiring timely detection to ensure safety.
- Deep learning, including Convolutional Neural Networks (CNNs) and Transformers, has advanced crack detection.
- CNNs struggle with global information, while Transformers incur high computational costs.
Purpose of the Study:
- To propose an improved VM-UNet model (VM-UNet++) for enhanced crack segmentation.
- To address limitations of existing CNNs and Transformers in crack detection.
- To optimize deep learning models for practical crack segmentation tasks.
Main Methods:
- Modification of the VM-UNet architecture by integrating the Mamba architecture.
- Comparative experiments using the Crack500 and Ozgenel public datasets.
- Evaluation of segmentation accuracy, parameter count, and inference speed.
Main Results:
- VM-UNet++ demonstrated significant advancements in segmentation accuracy.
- Achieved a 3% improvement in mean Detection Score (mDS) and a 4.6-6.2% increase in mean Intersection over Union (mIoU).
- Showcased lower parameter count and floating-point operations with satisfactory inference speed compared to state-of-the-art models.
Conclusions:
- The improved VM-UNet++ model effectively enhances crack segmentation accuracy.
- VM-UNet++ offers a computationally efficient and accurate solution for practical structural crack detection.
- The integration of Mamba architecture proves beneficial for crack segmentation tasks.
