Deep Learning for Automated Crack Recognition: Insights, Challenges, and Future Directions

Abstract

Crack detection in concrete structures is a critical component of infrastructure health monitoring, as cracks pose significant risks to structural integrity. Traditional manual inspection methods, while effective, are resource-intensive, time-consuming, and prone to human error. Automated crack detection using vision-based techniques has become a focal point of research, aiming to enhance accuracy, efficiency, and consistency in inspections. Deep learning (DL) methods have revolutionized crack detection by enabling end-to-end systems that automatically learn from data and perform pixel-level segmentation. This study provides an overview of recent advances in deep learning-based crack recognition, highlighting the mainstream detection methods, and revealing ongoing challenges including the need for large and diverse datasets, the development of lightweight models for real-time performance, and model generalization across diverse conditions.