Automated Crack Detection and Growth Monitoring: An Integrated SOM-KAZE Approach for Structural Health Monitoring

Abstract

This paper presents a novel approach for automated crack detection and growth monitoring in concrete structures, integrating Self-Organizing Maps (SOM) for pixel-level clustering with a KAZE feature descriptor for robust feature matching. The proposed method first employs SOM to cluster pixels in an initial crack image and derive a representative weight vector. Critically, this weight vector can be directly applied to subsequent images, thereby enabling the detection of crack growth without the need to retrain or rely on multiple reference images. Following SOM classification, noise is minimized through labeling, dilation, and superimposition, ensuring that thin or small crack portions are retained. Finally, the KAZE algorithm is employed to match features in different images, facilitating accurate crack localization and growth tracking across varying scales, rotations, and viewpoints. Experimental results demonstrate that the proposed method accurately detects crack evolution with minimal user intervention, offering a scalable and efficient solution for structural health monitoring.