Detecting structural cracks is vital for ensuring safety and preventing potential failures. However, manual inspection is time-consuming and subjective. As a result, researchers have turned to machine learning to automate the crack detection process. In this study, a deep learning-based approach is proposed to improve and boost the accuracy and efficiency of crack detection and health monitoring. The methodology involves creating a dataset of crack images and labelling them accordingly. Deep learning models, specifically YOLOv8, were trained on this dataset to effectively detect and pinpoint the location of cracks. Various preprocessing techniques such as denoising and color correction are applied to improve the quality of the images. Additionally, data augmentation techniques are used to diversify the dataset. Model performance was evaluated using Precision, Recall, and mean Average Precision (mAP). This research delves into investigating the advantages, challenges, and performance of machine learning algorithms (YOLOV8) for crack detection. Furthermore, it examines directional crack detection while comparing various instance segmentation models based on mAP scores. The study also discusses training results and presents graphs illustrating model performance and addresses dataset quality checks. Overall, this research contributes significantly towards evaluating object detection and instance segmentation methods in computer vision applications related to crack detection. The proposed deep learning approach shows promise in detecting cracks and analyzing them—an advancement that holds immense potential, for improving infrastructure integrity management systems.
Pshtiwan et al. (Wed,) studied this question.