PEER Hub ImageNet: A Large-Scale Multiattribute Benchmark Data Set of Structural Images

Abstract

With the rapid development of machine learning (ML) and deep learning (DL) in computer vision, adopting these learning tools in vision-based structural health monitoring (SHM) and rapid damage assessment is attracting interest in structural engineering. However, several critical issues become impediments, namely, no general automated detection framework, insufficient labeled data, lack of collaboration, and inconsistent research approaches. Thus, in this paper, the authors propose a general automated framework, namely, the Pacific Earthquake Engineering Research (PEER) Hub ImageNet (ϕ-Net), in which eight benchmark classification tasks are defined based on domain knowledge and past experience. A large-scale multiattribute ϕ-Net data set containing 36,413 pairs of images and labels was established, which was further separated into eight sub-data sets and open-sourced online. These pairwise images and labels can directly contribute to similar classification tasks and the raw structural images can further be used for labeling object localization and segmentation in future studies. Benchmark experiments with various DL models and training strategies were conducted with reported results. Two applications of ϕ-Net were further explored, namely, image-based postdisaster assessment of the 1999 Chi-Chi earthquake and the 2018 ϕ-Net Challenge. In conclusion, open-sourced data sets and benchmark results are the foundation for future studies where the extension applications reveal the great potential and contribution of ϕ-Net in vision-based SHM.