Share

In a recent webinar focused on the May 2024 issue of Earthquake Spectra, Vitor Silva presented a promising solution to a longstanding frustration in seismic risk assessment: acquiring detailed building data.

Traditionally, modellers have relied on limited datasets like housing census information, which lack the granularity needed for accurate risk assessments. Vitor highlighted the potential of combining artificial intelligence with Google Street View imagery to automatically generate detailed exposure models for entire cities.

However, a major hurdle lies in the scarcity of open-source, labelled building image datasets for training these machine learning algorithms. To address this gap, Vitor presented his research on creating a database of geo-located building photos specifically designed for this purpose. This database allows researchers to train algorithms that can automatically assign vulnerability classes to buildings based on their visual characteristics.

During the webinar, Vitor walked through the creation process of the database of building photos and explained how it was used to calibrate machine learning models for classifying the vulnerability of 3,000 buildings in Lisbon, Portugal. This research holds significant promise for enhancing the accuracy and level of detail in exposure models, leading to more robust seismic risk assessments.

To learn more about Vitor’s paper “A building imagery database for the calibration of machine learning algorithms”: https://doi.org/10.1177/87552930241229103.