Seismic risk assessment using machine learning for the automatic identification of building features

In nations with a high seismic hazard and a significantly vulnerable built heritage, seismic risk assessment represents a serious challenge. In particular, when seismic risk needs to be analyzed on large scales, vulnerability and exposure evaluations can lead to time-consuming and expensive investigations. In this work, artificial intelligence techniques are leveraged to address this issue. Specifically, Convolutional Neural Networks (CNNs) are trained to automatically collect data about buildings from satellite imagery and street views. In this work, three CNNs are trained to recognize the following features: building height, material, and construction period, deemed to be the essential parameters for associating a specific seismic vulnerability level to a building. The following step of this study involves the combination of vulnerability and exposure with seismic hazard to evaluate seismic damage and risk. The latter is represented by potential losses in terms of reconstruction costs, number of unusable buildings, and displaced people. Emergency management organizations may find the results of this work useful for setting priority standards for seismic retrofit operations, as well as for allocating rescue resources after an earthquake.