Revising the 2007 Peru Earthquake Damage Monitoring Using Machine Learning Models and Satellite Imagery

Adriano B.; Miura H.; Liu W.; Matsuoka M.; Portuguez E.; Diaz M.; Estrada M. · Journal of Disaster Research (2023) · DOI: 10.20965/jdr.2023.p0379

Resumen

We revised the building damage caused by the 2007 Pisco-Peru Earthquake using machine learning models and high-resolution satellite imagery. A framework for rapidly detecting collapsed buildings was proposed in the project “Development of Integrated Expert System for Estimation and Observation of Damage Level of Infrastructure in Lima Metropolitan Area” (JST-JICA SATREPS). The framework is based on a se-mantic segmentation model trained on freely available satellite and aerial imagery that does not include the target area. Thus, the generalization performance of the proposed framework was analyzed. Qualitative and quantitative analyses demonstrated that the proposed framework successfully predicted the distribution of collapsed buildings in Pisco. Moreover, it also reflects the ability to detect newly placed shelters. Our current trained model enables the rapid estimation of damaged buildings, crucial information for emergency response, and temporary refuges, which are also essential for fast rescue actions. © Fuji Technology Press Ltd. Creative Commons CC BY-ND:.

Instituciones: Universidad Nacional de Ingeniería