• Subscribe

Deep learning looks through the flames

Speed read
  • Wildfires take a heavy toll in lives and property loss
  • Emergency responders are challenged to make timely, life-saving decisions
  • Deep learning is changing the game for disaster preparedness teams

The intelligence of machines is a topic that generates a lot of anxiety. But what if artificial intelligence could save lives? That’s the hope of DeepEye.

An analytical tool designed to aid emergency response teams, DeepEye uses remotely sensed data, photo databases, and a real-time stream of social media data, to give disaster responders a head start.

Ghost in the machine?

When a machine ‘learns,’ it means a computer program has been immersed in sufficient fragments of data (e.g. striations on a shark fin, blotches on a giraffe’s neck) that it can predict and recognize an item. 

What is deep learning? A discussion with Damian Borth and Janis Keuper at ISC 2017 in Frankfurt.

When a pile of leaves is exhumed from an archeological site, the computer reads the features of a digital image and identifies and informs the human researcher about the content of that pile of leaves.

In contrast, humans learn by seeing a horse, a second horse, a third horse and so on, and then coming to grasp what a horse is (as opposed to, say, a camel).

Machine learning looks at fragments (mane, nostrils, fetlocks, hooves) and eventually pieces them together to identify a horse whenever it is presented with another fragment.

Computers can 'remember' a virtually unlimited set of examples, and they have a nearly infinite capacity to recall and quickly associate. 

Deep learning, a subset of machine learning, attempts to simulate the collocation of cognitive concepts humans hold in mind.

Whereas machine learning may learn about fragments of one aspect (a corner of an image, a syllable) and then build a base of knowledge, ‘deep’ learning plunges through many more layers. 

Accordingly, deep learning requires much larger stores of memory and more computing cores to make the associations required for a desired output. 

Fire? Ready, aim!

DeepEye uses deep learning to great effect and social benefit, finding a welcome partner in disaster management teams.

Scientists at institutions like the National Center for Atmospheric Research (NCAR) have previously used computer models to simulate where wildfires will strike. These models, however, are mostly unable to give emergency response teams a minute by minute prediction of how a wildfire may spread throughout a region.  <strong>Glowing report. </strong> Displacing 90,000 people and causing $3.6 billion in property damage, the Fort McMurray widfire is the costliest disaster in Canadian history. With DeepEye, emergency responders have artificial intelligence to help them predict where fires will spread. Courtesy Chris Schwarz; Government of Alberta. <a href= 'https://creativecommons.org/licenses/by-nd/2.0/legalcode' > (CC BY-ND 2.0)</a>

But what if there were a way researchers could chart potential paths of destruction in real time? With lives hanging in the balance, every single second counts.

Real time monitoring of natural disasters is the ability Damian Borth wants. He could achieve it, using deep learning programs to predict future wildfire routes.

“For me, it’s very important to look into areas where we can use artificial intelligence for good,” says Borth, director of the Deep Learning Competence Center at the German Research Center for Artificial Intelligence (DFKI). “We want to create applications that can protect our planet and save lives.”

To track nascent wildfires in real time, Borth’s team developed DeepEye, a multimodal analytical tool to aid emergency response teams. DeepEye had to ingest a large cross section of data to identify wildfires.

The team used LandSat satellite imagery, a dataset of images from Flickr, and over 18,000 posts from Twitter to train the program to recognize wildfire markers.

The images collected from LandSat were grouped into various spectral bands. DeepEye categorized the satellite data in three various areas: RGB image, vegetation health, and fire scars.

<strong>A drop in the bucket. </strong> A helicopter carries fire retardant to a wildfire burning in a Fort McMurray neighborhood. Deep learning is informing responders so they can react to natural disasters more efficiently — and save lives. Courtesy Premier of Alberta; Chris Schwarz; Government of Alberta. <a href= 'https://creativecommons.org/licenses/by-nd/2.0/legalcode '>(CC BY-ND 2.0) </a>

The RGB images of the area were unaltered, while the vegetation health images measured the well-being of the province’s plant ecosystem. Fire scar pictures tracked where the wildfire had marred the landscape.

The combination of data is a potent weapon for the DeepEye project. It combines pictures that users take of the wildfire on the ground with satellite images tracking the fire's progress from above as inputs for its network. Then, it  produces outputs of where a wildfire will strike next.

“You can give a response team information like ‘these are areas that are burnt already. Here's the probability of where new burns will come,’” Borth says. “With this added knowledge, emergency response teams can then dispatch resources more effectively.”

Diving into data with deep learning might be the next step for disaster response teams. It could soon prevent a single flame from turning into an all-encompassing inferno.

If you'd like a closer look, you can preview the DeepEye project at the MediaEval Workshop in Dublin, Ireland from September 13-15.

Join the conversation

Do you have story ideas or something to contribute? Let us know!

Copyright © 2017 Science Node ™  |  Privacy Notice  |  Sitemap

Disclaimer: While Science Node ™ does its best to provide complete and up-to-date information, it does not warrant that the information is error-free and disclaims all liability with respect to results from the use of the information.

Republish

We encourage you to republish this article online and in print, it’s free under our creative commons attribution license, but please follow some simple guidelines:
  1. You have to credit our authors.
  2. You have to credit ScienceNode.org — where possible include our logo with a link back to the original article.
  3. You can simply run the first few lines of the article and then add: “Read the full article on ScienceNode.org” containing a link back to the original article.
  4. The easiest way to get the article on your site is to embed the code below.