Using machine learning to detect blue plastic tarps on roofs in Puerto Rico damaged by Hurricane Maria leveraging drone and Sentinel-2 imagery

Authors: Kyle M Monahan*, Tufts Data Lab, TTS, Tufts University, Medford MA 02144, Yuehui (Aurora) Li, Urban and Environmental Policy and Planning, Tufts University, Medford MA 02144
Topics: Remote Sensing, Spatial Analysis & Modeling, Drones
Keywords: deep learning, Hurricane Maria, Puerto Rico
Session Type: Virtual Poster
Day: 4/10/2021
Start / End Time: 3:05 PM / 4:20 PM
Room: Virtual 52
Presentation File: No File Uploaded

Hurricane Maria made landfall in southeast Puerto Rico on September 20th, 2017. This powerful storm affected millions of people, causing a partial collapse of the power and healthcare systems, and placed much of the affected population at risk for adverse mental and physical health outcomes. In this work, a remotely sensed indicator of damage to infrastructure was developed for Puerto Rico using remotely sensed blue-colored tarps placed over damaged roofs using a machine learning technique. A random forest and deep learning model were created using ArcGIS Pro, training on both drone imagery (1 meter resolution) and Sentinel-2 imagery (15 meter resolution). Results of models were compared using accuracy, F1, and sensitivity values for each class. Future work using change detection should be performed.

Abstract Information

This abstract is already part of a session. View the session here.

To access contact information login