Skip to main content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Facilitating Large-Scale Data Collection by Synthesizing Remote Sensing, GIS, and Machine Learning

Published

Author(s)

Patrick S. Crawford, Judith Mitrani-Reiser, Andrew Graettinger

Abstract

Post-disaster studies are critical for measuring impact to communities, from physical infrastructure performance and vulnerabilities, to interdependencies with social and economic systems of the affected communities. The data collected in these studies additionally are used to drive community resilience modeling. Disaster studies aimed at capturing perishable data are often time-sensitive and intensive, logistically rigorous, and expensive in terms of dollars and person-hours due to the typically large impacted areas. Research entities regularly undertaking these studies, would benefit from enhanced methods for the execution of these activities. Additionally, development of enhanced data collection and processing tools allows disaster researchers to prioritize time in the field, capture larger datasets expending less effort and fewer resources, and streamline the extraction of meaningful information from collected data. A methodology is presented to capture a large, passive, geospatial dataset in affected areas using vehicle-mounted 360° video cameras, elicit serviceable datasets by extracting images from the videos at locations of interest (LOI), and apply these images to a machine learning approach for image classification. Algorithms have been developed to extract images from the videos using an input suite of LOIs, e.g. damaged buildings, and associate the extracted images with their LOI. Extracted images can be labelled with identifying features (e.g. level of damage) to facilitate the creation of machine learning algorithms for image classification, which typically requires large amounts of labelled data before accuracy convergence can be achieved. A case study is presented for Dauphin Island, Alabama, a barrier island in the Gulf of Mexico prone to hurricane winds and flooding. All roads of the island were driven and images were extracted at all building locations. A subset of extracted images were labelled either “elevated building”, “non-elevated building”
Conference Dates
July 14-18, 2019
Conference Location
broomfield, CO
Conference Title
natural hazards workshop

Keywords

machine learning, remote sensing
Created July 14, 2019, Updated September 5, 2019