Abstract

As an important part of construction engineering, the safety and durability of concrete structures directly affect the overall quality and service life of buildings. However, traditional concrete defect detection methods rely on manual work, which is time-consuming and laborious. It is urgent to explore automatic defect detection methods. This study aims to compare the performance and applicable scenarios of different target detection algorithms, and establish a more efficient and accurate detection method for the detection of apparent quality defects of concrete structures. Therefore, this study first constructed a concrete apparent quality defect image dataset containing 1085 images and 1760 defect instances; then selected four target detection models, YOLOv5, YOLOv9, SSD and EfficientNet, for training and testing; finally, the performance of the models was compared and analyzed by comparing the speed and efficiency of each model training, precision, recall rate and mAP. The results show that YOLOv9 has better robustness and accuracy among the four models for the problem of concrete structure apparent defect detection, and its mAP can reach 0.71. The relevant research provides useful guidance for the development of construction quality defect detection methods and equipment, and serves the improvement of construction inspection efficiency and construction quality.

Download paper here

Download preprint here

Financial Sources: