Concrete Crack Detection in Extremely Dark Environments Based on Infrared-Visible Multi-Level Registration Fusion and...

Area of Science:

• Materials Science
• Computer Vision
• Image Processing

Background:

• Concrete crack detection is challenging in dark environments due to poor illumination and modal differences.
• Existing methods struggle with heterogeneous image registration and segmentation accuracy.

Purpose of the Study:

• To develop an advanced framework for accurate concrete crack detection in challenging low-light conditions.
• To enhance both image registration/fusion and crack segmentation performance.

Main Methods:

• A two-stage framework: registration-fusion followed by decoupling-segmentation.
• Registration utilizes morphological priors and quadtree constraints for topological matching.
• Segmentation employs a Latent Frequency-Decoupled Topological Network (LFDT-Net) with Discrete Wavelet Transform (DWT).

Main Results:

• Achieved superior registration accuracy, fusion quality, and segmentation accuracy compared to mainstream methods.
• The LFDT-Net effectively decouples infrared and visible light features, reducing artifacts.
• Demonstrated a mean Intersection over Union (mIoU) of 81.7% for crack segmentation.

Conclusions:

• The proposed framework significantly improves concrete crack detection in dark, heterogeneous environments.
• The method accurately restores microscopic crack edges and topological structures while suppressing noise.
• This approach offers a robust solution for infrastructure monitoring and safety assessment.