A Deep Learning-Based Watershed Feature Fusion Approach for Tunnel Crack Segmentation in Complex Backgrounds

Area of Science:

• Civil Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Highway tunnel structural defects, especially cracks, increase annually, overwhelming traditional manual inspection methods.
• Rapid tunnel network expansion necessitates advanced automated inspection solutions.
• Existing machine vision and deep learning methods face challenges in tunnel crack detection due to complex backgrounds and data labeling.

Purpose of the Study:

• To develop an automated labeling and optimization algorithm for crack sample sets in highway tunnels.
• To enhance the performance of deep learning-based crack segmentation networks.
• To provide an efficient and accurate solution for identifying structural defects in tunnels.

Main Methods:

• Utilized crack features and the watershed algorithm for efficient automated segmentation with minimal human input.
• Optimized deep learning crack segmentation networks by analyzing various network depths and residual structure configurations.
• Incorporated axis extraction and watershed filling algorithms to refine segmentation outcomes.

Main Results:

• Achieved a crack segmentation accuracy of 98.78% under diverse lining surface conditions and interference factors.
• Obtained an Intersection over Union (IoU) of 72.41% for crack segmentation.
• Demonstrated a robust solution for crack segmentation in tunnels with complex backgrounds.

Conclusions:

• The proposed automatic labeling and deep learning optimization algorithm effectively addresses limitations in tunnel crack detection.
• The enhanced segmentation accuracy and IoU provide a reliable tool for infrastructure monitoring.
• This approach offers a significant advancement for intelligent crack segmentation in civil engineering applications.