24STUDENT Understanding the Physical Structure of Communities for Improved Disaster Impact Predictions

Accurately predicting and assessing disaster impacts on communities requires a comprehensive understanding of their physical structures, particularly buildings and critical infrastructure. This study developed Convolutional Neural Network (CNN) model to classify buildings into categories such as residential, commercial, and industrial, following building type classification in the Enhanced Fujita Scale. We collected imagery of various building types from multiple open-source databases, providing a diverse dataset for classification. Two types of CNN models were compared: a basic CNN built from scratch and a pre-trained model fine-tuned using transfer learning techniques. The goal was to evaluate the effectiveness of these models in categorizing buildings based on structural features. The models were assessed using performance metrics such as accuracy, precision, recall, and F1 score, ensuring a robust comparison. It will support the development of a geodatabase cataloging diverse building structures within a community, enabling more precise predictions of wind-induced damage. This geodatabase will serve as a valuable tool for disaster response and resilience planning, offering insights into the vulnerability of different building types. By improving damage prediction models, communities will be better equipped to plan for and respond to disasters, enhancing overall resilience.