Student Research Reports

View Research Report

Abstract The U.S. is an ethnically and economically diverse nation containing exceptional differences in socioeconomic and urbanization levels throughout the nation. In order to maximize the quality of life, the prediction of developmental and civil data is crucial for urban planners as it provides the basis for the problems within a community that should be addressed. Our goal was to create a tool that could utilize images taken by citizen scientists to supplement the existing geospatial data often used in civil development. Though pictures gathered using citizen science are able to provide extra context on their own, the tool created would involve automated classification, generating more information at a faster pace. Building and Infrastructure Recognition using AI at Large Scale (BRAILS) is a Python library that utilizes deep learning and computer vision. The BRAILS modules, Occupancy Classifier, and Year Built Classification, fine-tune a model that outputs a score (1-100) for urbanization level. The Convolutional Neural Network (CNN) model, supplemented by the BRAILS library, increased accuracy by accommodating additional urban features predicted by the BRAILS modules. Our CNN model was trained on a dataset of 50 high-quality representations of each urbanization level (urban, suburban, rural) from our Earth System Explorers, SEES2024, GLOBE Observer database using TensorFlow, reaching exceptional accuracy (90%) in determining urbanization level. Currently, the model only expands on the urbanization information given from geospatial imaging; however, in the future, the model could be expanded by incorporating national databases like the Socioeconomic Data and Application Center (SEDAC) and the US Census. This would provide a multitude of civil data that can provide additional context to the imaging, helping civil developers further determine the best course of action. Keywords Area Classification, Urbanization, CNN, BRAILS