Vision-based concrete crack detection using deep learning-based models

It is critical to detect cracks in concrete promptly and effectively to limit further deterioration and to perform timely repairs. Several convolutional neural networks have been proposed in recent years for identifying objects varying in their accuracy and speed. In this study, the YOLOv7, YOLOv5s, YOLOv5m, and YOLOv5x object identification models were trained for crack detection in concrete surfaces. The networks were trained using 1600 images of concrete cracks and analyzed. The different YOLOv5 versions and YOLOv7 are compared using assessment measures including F1 score, recall, and mAP. The research study found that all of the models predicted encouraging results in terms of crack detection in concrete images. According to the results, YOLOv5m and YOLOv5x achieved F1 scores of 0.87 and 0.86, respectively. Differently, YOLO5s and YOLOv7 acquired an F1-score of 0.85 and 0.84, respectively. As a result, this research verifies the recently introduced deep learning technology, which can replace conventional crack detection and identification techniques with more reliable and efficient alternatives. © 2023, The Author(s), under exclusive licence to Springer Nature Switzerland AG.