Communications of the ACM
Combining News Media, AI to Rapidly Identify Flooded Buildings
Researchers at Japan's Tohoku University have developed a machine learning model that can help identify flooded buildings within 24 hours of a disaster using news media photos.
Credit: MLIT/Shikoku Regional Development Bureau.
A machine learning model developed by researchers at Japan's Tohoku University can help identify flooded buildings within 24 hours of a disaster using news media photos.
The model was applied to Mabi-cho, Kurashiki city in Okayama Prefecture, which experienced heavy rains and flooding in 2018.
After identifying press photos and geolocating them based on landmarks and other visual cues, the researchers used synthetic aperture radar (SAR) PALSAR-2 images from the Japan Aerospace Exploration Agency to approximate the conditions of unknown areas.
Buildings surrounded by floodwaters or within non-flooded areas were classified using a support vector machine.
About 80% of the buildings classified by the model as flooded were actually flooded during the event.
Tohoku's Shunichi Koshimura said, "Our model demonstrates how the rapid reporting of news media can speed up and increase the accuracy of damage mapping activities, accelerating disaster relief and response decisions."
From Tohoku University (Japan)
View Full Article
Abstracts Copyright © 2021 SmithBucklin, Washington, DC, USA
No entries found