Deep Transfer Learning for Image-Based Component Type Classification of Damaged Structural Members

Many areas of structural research, including the use of images to identify structural member types, are paying more attention to image processing. Additionally, deep learning has fundamentally transformed the field of image processing (DL). This research applies cutting-edge deep learning known as Transfer Learning (TL) technology to a civil engineering application that transfers knowledge from a source domain to a target domain, this application is about detecting the type of damaged structural component from pictures. Inspired by the ImageNet Challenge in computer science and computer hardware development, the structural image net project publishes a very limited number of images from structural components. Transfer Learning based on VGGNet (Visual Geometry Group) is presented and used on the mentioned dataset to minimize overfitting. Models created by the Transfer learning algorithm provide good recognition performance compared to training from scratch and may be utilized for preliminary analysis and future enhancement. These findings also point to the possible use of deep TL in image-based damaged structural component identification.

Deep Transfer Learning for Image-Based Component Type Classification of Damaged Structural Members Keywords:

Deep Learning , Transfer Learning , Computer vision , Structural components classification , Pretraind network.