Enhanced concrete crack detection using YOLOv8: a multi-background approach

The integrity of concrete structures is paramount for the safety and longevity of civil infrastructure. Traditional methods for detecting cracks in these structures are often time-consuming, subjective and prone to human error. This study introduces an automated, high-performance crack detection system based on the YOLOv8 deep learning architecture, designed for real-time and accurate identification of cracks in concrete images. YOLOv8, renowned for its superior speed and accuracy, integrates a powerful convolutional backbone with decoupled head structures for detection and localisation, enabling the detection of fine and irregular crack patterns under varied lighting and background conditions. The system trained and tested on a comprehensive dataset of 4000 concrete images, achieved a precision of 91.80%, recall of 92.50% and overall accuracy of 93% in detecting cracked regions. For non-cracked images, the system demonstrated an accuracy of approximately 90%. These results underscore the model’s robustness in accurately identifying crack damages with minimal false positives and negatives, highlighting YOLOv8 as a reliable solution for automated structural health assessment.