Automated Crack Detection in Building Facades Using Deep Learning

— Building facades naturally develop cracks over time due to factors such as material degradation, environmental influences, and structural loads. If not identified early, these cracks can escalate into serious structural issues, compromising both safety and durability. Traditional inspection methods, which rely on manual assessment, are often inefficient, labor-intensive, and susceptible to human errors. To address these limitations, automated crack detection using deep learning presents a viable solution. For training CNN and VGG models for roof crack detection, researchers typically use image datasets containing labeled examples of cracked and non-cracked surfaces. This research explores the application of Convolutional Neural Networks (CNN) for detecting and categorizing cracks in building facades. Given the challenges of limited training data, transfer learning is employed to enhance detection accuracy. Experimental results demonstrate that while a conventional CNN model achieved an accuracy of around 89%, leveraging transfer learning significantly improved performance, achieving an accuracy of 94%. These findings highlight the effectiveness of transfer learning in enhancing detection capabilities even with minimal data availability. Implementing deep learning-driven automation in infrastructure inspection not only improves accuracy but also reduces manual intervention, ensuring timely maintenance and prolonged structural stability.