Deep-learning tools assess earthquake damage
Deep-learning tools assess earthquake damage
Author: | Eric Bender |
---|---|
Magazine Section: | Change The World |
College or School: | CoE |
Article Type: | Issue Feature |
Page CSS: | #article-banner { background-position: center top !important;} |
Feature Intro: | Shirley Dyke and her team are the first to employ deep-learning technologies for rapid assessment of structural wreckage. |
To streamline the grueling task of post-disaster data collection and organization, Purdue researchers use computers to automate the process. Shirley Dyke, professor of civil engineering and mechanical engineering, and Chul Min Yeum, postdoctoral researcher, are developing deep-learning algorithms to classify images and identify regions of concern.

(Image source: DataHub)
“Rather than having reconnaissance teams spend several hours trying to organize their data and figure out how to collect essential data during the next day, we’d rather enable doing this automatically and rapidly with our algorithms,” she says. “We want to automatically determine if an instance of structural damage is there, and understand what additional data should be collected in the affected community. This work will allow teams to collect more data and images, knowing that it can be made available for use.”
Dyke’s project builds on work by Yeum both in computer-vision technologies and deep-learning methods. Deep learning draws on neural network techniques and large data sets to create algorithms and design classifiers for assessing the data. “Deep learning has been applied to many everyday image classification situations, but as far as we know, we’re the first to employ it for damage assessment using a large volume of images,” Yeum says.
Training the system requires a painstaking process of manually labeling tens of thousands of images from diverse data sets according to a chosen schema. The computer system learns about the contents and features of each class, and then it trains the algorithms to find the best ways to identify the areas of interest.

(Photo: Mark Simons, Purdue University)
So far, Dyke and Yeum have gathered about 100,000 digital images for training the system, contributed from researchers and practitioners around the world. Most of these are from past earthquakes, but other hazard images from tornadoes and hurricanes also are collected.
“We will seek opportunities to test our system in the field, perhaps by working with researchers that are examining the tens of thousands of buildings exposed to significant ground motions in Italy in the past months,” Dyke says. “The classifiers trained using our database of past images would be applied to new images collected on-site to directly support teams in the field.”
“As we develop the methodologies to automatically organize the data, we are engaging researchers who deal with reconnaissance missions and building codes,” Dyke says, “so that these classifiers can be used by people around the world.”
Comments