Affiliation:
1. The University of Memphis, USA
2. University of Engineering and Technology, Taxila, Pakistan
Abstract
These defects can occur in the form of cracks, pain deterioration, dampness, etc. due to mechanical and weathering effects. Crack identification and categorization must be part of the inspection procedure for civil engineering structures. Convolutional neural networks (CNNs), a sub-type of deep learning (DL), can automatically classify the defective images of wooden structures. In this research, 10 pre-trained models of CNN, including ResNet18, ResNet50, ResNet101, Inception-V3, GoogleNet, ShuffleNet, InceptionResNet-V2, MobileNet-V2, XceptionNet, and NASNet-Mobile are evaluated for classification and prediction of defects in wooden structures. After prediction of class, the algorithm calculated the angle, length, and width of cracks with quantification errors of 0.15%, 1.54%, and 4.2% for first testing image, and 2.3%, 0.99%, and 2% for second testing image.