A deep-learning approach from Stanford University detects property damage caused by wildfires. The system, known as DamageMap, could help responders focus their recovery efforts and offer more immediate information to displaced residents.

Most AI-based computational systems cannot efficiently classify building damage because the artificial-intelligence systems traditionally compare post-disaster photos with pre-disaster images that must use the same satellite, camera angle, and lighting conditions — parameters that can be expensive to obtain or unavailable.

Instead of analyzing before-and-after photos, researchers from Stanford University and the California Polytechnic State University trained a program using machine learning to rely solely on post-fire images.

DamageMap first uses pre-fire photos of any type to map the area and pinpoint building locations. Then, the program analyzes post-wildfire images to identify damage through features like blackened surfaces, crumbled roofs, or the absence of structures.

“People can tell if a building is damaged or not – we don’t need the before picture – so we tested that hypothesis with machine learning,” said co-author Krishna Rao, a graduate student in Earth system science at Stanford’s School of Earth, Energy & Environmental Sciences (Stanford Earth), in a recent press release . “This can be a powerful tool for rapidly assessing damage and planning disaster recovery efforts.”

With a deep learning technique called supervised learning, Rao's model can continue to be improved as it receives more data.

Rao and colleague Marios Galanis, a graduate student in the Civil and Environmental Engineering Department at Stanford’s School of Engineering, developed the project during Stanford’s 2020 Big Earth Hackathon: Wildland Fire Challenge . The duo later collaborated with Cal Poly researchers to refine the platform.

The team tested the application using damage assessments from Paradise, California, after the Camp Fire and the Whiskeytown-Shasta-Trinity National Recreation Area after the Carr Fire of 2018.

In a short Q&A below, Krishna Rao tells Tech Briefs where they expect to try out DamageMap next.

Tech Briefs: Your press release mentions that the AI is not meant to replace door-to-door checks made by humans. Can you provide an example of how AI can supplement the human effort?

Krishna Rao: Our AI system can supplement human effort by helping damage inspectors prioritize the inspections. For example, if there are thousands of buildings in a town, we could pinpoint the damaged buildings within minutes, and the damage inspectors can visit those before the other buildings. Why our system cannot entirely replace the manual inspection is that it is not perfect. It doesn't work well on buildings which have canopies covering it, and it doesn't work at all when the damage is only internal to the property.

Tech Briefs: What characteristics of damage is the AI trained to detect? What other characteristics are the most challenging for the system to recognize?

Krishna Rao: We trained our AI model using red-green-blue images from satellites, flights, and drones. Our AI model is trained to detect damage to roof of buildings. Our tool doesn't quite work well when the colors of roofs are similar to the ground color, and when there are tree canopies partially covering the roof.

Tech Briefs: How can fire fighters and emergency responders best use a tool like this?

Krishna Rao: I think emergency responders can use this tool to plan the allocation of relief resources. For example, to get a first-order sense of the scale of damage to a town, they wouldn't be required to drive through the entire town and subjectively gather the scale of the damage. Our tool can provide information like the number of damaged properties, addresses of them, very rapidly. Such information can be used to locate interim relief centers, and call for additional equipment and/or staff if the scale of damage is too big.

Tech Briefs: What inspired this idea?

Krishna Rao: My research is on wildfire hazard prediction, which involves understanding the effects of forest dryness of where wildfires could occur. But ever too often I read news about the disastrous consequences of wildfires. When Camp Fire demolished Paradise, that was the tipping point.

Although I couldn't save the town from collapse, I wished to work on a tool that rapidly collects and infers information about wildfire damage so that rescue efforts can be expedited, residents can get a sense of the overall damage to their town, and property insurance claims can be made smoother. If owners are able to prove damage to their properties by citing our damage estimate, insurers might consider providing some interim payout.

Tech Briefs: What are you working on now?

Krishna Rao: We are in conversation with fire agencies and insurers to understand how best to adapt this technology to suit their needs. Our conversations so far suggest there is a lot of excitement towards using technology to aid disaster relief.

What do you think? Share your questions and comments below.