Concrete Crack Detection and Classification Methods Based on Machine Vision and Deep Learning.

Area of Science:

• Civil Engineering
• Structural Health Monitoring
• Image Processing

Background:

• Increasing underground space development necessitates robust structural crack monitoring.
• Existing methods face challenges with complex backgrounds and varying image resolutions.

Purpose of the Study:

• To develop a unified framework for automated crack analysis in underground structures.
• To enhance crack detection and classification accuracy for improved safety assessments.

Main Methods:

• An improved OTSU threshold segmentation algorithm using sliding windows and local statistics for preprocessing.
• Comparative analysis of Support Vector Machine (SVM), Convolutional Neural Network (CNN), ResNet-18, and K-means clustering for crack identification and classification.
• Validation through full-scale loading tests on metro shield tunnel segments.

Main Results:

• The improved OTSU method significantly enhances noise suppression and detail preservation compared to classical approaches.
• SVM achieved over 96% accuracy in crack classification, outperforming other models under limited data conditions.
• Real-world validation on tunnel segments confirmed SVM's robustness with 95.45% accuracy.

Conclusions:

• The proposed framework offers an efficient and reliable solution for automated crack detection and classification.
• SVM is particularly effective for crack classification in metro tunnel infrastructure and similar systems, especially with limited training data.
• The study highlights the importance of advanced image processing and machine learning for structural integrity monitoring.