Supporting Data for 'Deep learning-aided parallel laser line-camera system for structural defect localization, quantification, and evaluation''

This research focuses on using images of structural defects to train and develop various deep-learning models aimed at detecting and segmenting defects. All images used in this study have been collected by the author from aging reinforced concrete buildings in Hong Kong.

• Chapter 3.1 includes a labeled dataset of 2,384 images for the development of the YOLOv9s model, which is designed for detecting common structural defects such as bulging, delamination, spalling, and cracking. This chapter contains a training folder, a validation folder, and a data.yaml file that defines the types of objects being detected.
• Chapter 3.2 features a labeled dataset of 1,304 images specifically for the YOLOv9s model to detect bulging and delamination. It follows the same structure as Chapter 3.1, with separate folders for training and validation, and matching filenames between the images folder and the labels folder for accurate image annotation.
• Chapter 6 presents a smaller dataset of 415 images used to train the U-Net model for segmenting tiny cracks. Similar to previous chapters, it includes training and validation folders. Each folder contains an images folder for the defect pictures and a masks folder for image annotations, ensuring that filenames correspond between images and masks for precise annotations.

All images across the datasets are standardized to a resolution of 640x640 pixels. The organization of data in each chapter is structured to facilitate efficient model training and validation.