A Bridge Crack Segmentation Algorithm Based on Fuzzy C-Means Clustering and Feature Fusion.

Area of Science:

• Civil Engineering
• Computer Vision
• Image Processing

Background:

• Traditional image processing for crack detection suffers from noise sensitivity and threshold dependency.
• Deep learning methods require extensive labeled data, posing a significant challenge.
• Existing methods struggle with fine cracks and misjudgments in noisy environments.

Purpose of the Study:

• To develop a novel, efficient, and unsupervised crack segmentation algorithm for bridge damage detection.
• To overcome the limitations of traditional and deep learning approaches in crack detection.
• To improve the accuracy and real-time efficiency of crack segmentation in structural health monitoring.

Main Methods:

• Utilized fuzzy C-means (FCM) clustering with c=3 in a 3D feature space (B-channel pixels) for preliminary segmentation.
• Employed connected domain labeling and a circularity threshold to distinguish linear cracks from noise.
• Implemented a 5x5 neighborhood search strategy based on crack pixel amplitude to restore fragmented crack continuity.

Main Results:

• Achieved an accuracy of 0.885 and a recall rate of 0.891 on Concrete Crack and SDNET2018 datasets.
• Outperformed the DeepLabv3+ algorithm by 4.2% in performance.
• Demonstrated real-time efficiency with a processing time of 0.8 seconds per image.

Conclusions:

• The proposed algorithm offers an efficient unsupervised solution for bridge crack detection.
• It effectively addresses challenges like missed fine cracks and misjudged broken cracks in noisy conditions.
• The integration of geometric features and pixel distribution characteristics enhances detection accuracy and robustness.