Citas bibligráficas
Portillo, A., (2024). Identificación de edificaciones colapsadas mediante imágenes satelitales, curvas de fragilidad, mapas de demanda y redes neuronales [Pontificia Universidad Católica del Perú]. http://hdl.handle.net/20.500.12404/28836
Portillo, A., Identificación de edificaciones colapsadas mediante imágenes satelitales, curvas de fragilidad, mapas de demanda y redes neuronales []. PE: Pontificia Universidad Católica del Perú; 2024. http://hdl.handle.net/20.500.12404/28836
@mastersthesis{renati/1662600,
title = "Identificación de edificaciones colapsadas mediante imágenes satelitales, curvas de fragilidad, mapas de demanda y redes neuronales",
author = "Portillo Cazorla, Aymar Raúl",
publisher = "Pontificia Universidad Católica del Perú",
year = "2024"
}
This thesis presents an innovative method for rapidly identifying areas affected by earthquakes. The method is notable for its ability to operate with limited information while still delivering precise results. It combines pre -and post- event satellite images, building inventories, fragility curves, seismic demand maps, and labeled samples collected shortly after the disaster. The central feature of this method is its use of neural networks, wich can classify large amounts of data on the fragility of buildings in affected areas with a minimal number of labeled samples- as few as six, as demonstrated in the 2016 Kumamoto earthquake. These samples are gathered from verificable sources, geolocated, and recorded for use in calibrating the neural network, allowing for a swift and accurate assessment of collapsed areas. The results obtained with this approach have proven to be effective, achieving an average accuracy of 87% in the 2016 Kumamoto earthquake and 82% in the 2023 Turkey-Syria earthquake. The method excels in its ability to efficiently analyze extensive areas, making it ideal for immediate response after significant seismic events. While the method is functional and fast, opportunities for future improvements have been identified. For instance, optimizing the neural network architecture and evaluating the impact of the number of labeled samples used for calibration are necessary. Additionally, testing the method in various contexts is crucial to validate its versatility and to identify any potential deficiencies that could be addressed subsequently.
Este ítem está sujeto a una licencia Creative Commons Licencia Creative Commons