People around the world are suffering from the nerve-wracking anxiety of waiting weeks or months to find out if a home has been damaged by an intensifying wildfire. Now that the smoke disappears for aerial photography, researchers have found a way to identify damage to the building within minutes.
Through a system they call Damage Map, the team at Stanford University and California Polytechnic State University (Cal Poly) has brought an artificial intelligence approach to building evaluation. Instead of comparing the previous and next photos Machine learning Depends only on post-fire images.The survey results will be displayed in International Journal of Disaster Risk Mitigation..
“We automate the process, First responder “Even for citizens who want to know what happened to their home after a wildfire,” said Marios Galanis, a senior research author and graduate student in the Department of Civil and Environmental Engineering at Stanford University’s Faculty of Engineering. Human accuracy. “
The current method of assessing damage is for people to make door-to-door visits to check all buildings. DamageMap is not a replacement for face-to-face damage classification, but it can be used as a scalable supplemental tool by providing immediate results and the exact location of the identified building. Researchers have tested with a variety of satellite, aviation, and drone photographs with at least 92% accuracy.
“This application will probably allow you to scan the entire paradise town in a matter of hours,” said G., a senior author who is an assistant professor at Carpoli, referring to the town of Northern California destroyed by the 2018 campfire. Andrew Flicker said. “This provides more information in the decision-making process of firefighters and emergency responders, and fire damage by filing insurance claims and obtaining information to help bring life back on track. I hope we can help you. “
Most computing systems cannot efficiently classify building damage because AI compares post-disaster photos with pre-disaster images that require the same satellite, camera angle, and lighting conditions. According to researchers, current hardware isn’t sophisticated enough to record high-resolution surveillance daily, so the system can’t rely on consistent photos.
Instead of looking for differences between the front and back images, DamageMap first uses all sorts of pre-fire photos to map the area and locate the building. The program then analyzes post-wildfire images to identify damage due to features such as blackened surfaces, collapsed roofs, and lack of structures.
Co-author Krishnarao, a graduate student in Earth Systems Science at Stanford University’s Faculty of Earth Sciences, said: , Energy and Environmental Sciences (Stanford Earth). “This can be a powerful tool for quickly assessing damage and planning disaster recovery efforts.”
Structural damage from wildfires in California usually falls into four categories: little damage, minor damage, major damage, or destruction. Because DamageMap is based on aerial photography, researchers quickly realized that the system couldn’t make such a detailed assessment and trained the machine to easily determine if there was fire damage. Did.
The team used a deep learning technique called supervised learning, which allows them to continuously improve their models by providing more data. They tested the application using damage assessments in Paradise, California after a camp fire and Whiskeytown-Shasta-Trinity National Recreation Area after a 2018 car fire. Researchers said the open source platform can be applied to wildfire-prone areas. We also hope that you will receive training to classify the damage caused by other disasters such as floods and hurricanes.
“So far, it’s been suggested that this can be generalized, and if you’re interested in actually using it, you can continue to improve,” Galanis said.
Galanis and Rao developed the project at Stanford University’s 2020 Big Earth Hackathon: Wildland Fire Challenge. He then worked with Cal Poly researchers to improve the platform. This is the result of Rao and Frickers attending Google’s 2019 Geo For Good conference, and they created their first prototype as part of the conference Build-A-Thon.
The co-authors tested the model results against damage data collected in the field by California Department of Forestry and Fire Protection (CAL FIRE) agents. This is the information that made the research possible.
“Damage inspectors have worked hard to make door-to-door visits, damage investigations, geotagging locations, and ultimately to the public,” Lao said. “Future technology research or innovation depends directly on access to such data.”
Marios Galanis et al, DamageMap: A classifier for buildings damaged after a wildfire, International Journal of Disaster Risk Mitigation (2021). DOI: 10.1016 / j.ijdrr.2021.102540
Quote: AI system damaged by wildfire (September 16, 2021) acquired from https://phys.org/news/2021-09-ai-wildfire.html on September 16, 2021 Identify the building
This document is subject to copyright. No part may be reproduced without written permission, except for fair transactions for personal investigation or research purposes. The content is provided for informational purposes only.
https://phys.org/news/2021-09-ai-wildfire.html AI system identifies buildings damaged by wildfires