Defect Detection in Civil Structure Using Deep Learning Method

Abstract

With the usage time, undesirable defects like crack and spalling appeared in various types of civil structures including beam; wall and column. These defects have a great influence on the usage life of the structures and hence need attention and maintenance. Normally, defect detection in civil structures is the human visual inspection. The manual approach is time-consuming and has a low accuracy. In this paper, we use deep learning method to predict multiple defects in various types of civil structures. Crack and spalling are recognized using the convolutional neural network model in a pixel level. To build the inspection model, the U-net Plus Plus architecture is used for segmentation. In the dataset, 1300 images are taken for training and 600 images for validation. The obtained computer vison model is tested on a dataset of about 491 images. The recall for crack and spalling reach 0.89 and 0.85 respectively. This research work develops an automated approach for defects detection in civil structures. The results can provide helpful guidance for structure health monitoring and maintenance in civil engineering like buildings and bridges.