Natural Disaster Damage Assessment
In 2018 the United States experienced 14 disasters that cost the economy as much or more than $1 billion each. According to the U.S. National Oceanic and Atmospheric Administration, from 2016 to 2018, the average number of billion-dollar disasters totaled 15 each year, while the average for 1980-2018 was just 6.2 events per year.
The numbers don’t lie. The frequency of natural disasters is increasing, but the question remains, how do we better plan for the aftermath of these impending disasters? Post-disaster, it is crucial to estimate the damage quickly to support response and recovery services adequately. Current response and recovery efforts are often slowed by the need to manually review post-disaster status in-person or by individually reviewing aerial imagery. REI Systems has created a solution to this manual and time-consuming challenge. The solution can automatically and quickly detect and assess damaged buildings from satellite images with less need for human effort.
REI Systems is using semantic segmentation techniques to train custom deep learning models to detect buildings in satellite imagery and identify damage to those buildings by using images from before and after a disaster event. From there, one can assess damage based on the imagery comparison. This tool, if implemented, will result in much faster identification of damaged neighborhoods and specific buildings, so that response teams could be sent immediately to help disaster victims without the need for a time-consuming aerial or terrestrial survey of the damaged region. Also, the tool could help assess the severity of damage that would prioritize where funding is needed, support insurance claims adjustments, and other recovery tasks.
REI Systems will be presenting this case study at Deep Learning World Las Vegas 2020. Find out more about the event or when REI Systems will be presenting online here.
Name: Ankit Mittal, Senior Manager