The Challenge | Spot That Fire!

Build a crowdsourcing tool for citizens to contribute to early detection, verification, tracking, visualization, and notification of wildfires.


NoFire is an interactive service which predicts and estimates the probability of fire existence in a worldwide scale by image and text processing on social media, NASA APIs and reports it to people near the incidents.


Say No To Fire


Wildfires can cause serious damage to property and human life as well as local ecosystems. No_Fire is an interactive service that supports real-time monitoring and detection of wildfires to prevent their destructive impacts. It encourages citizens to participate in wildfire early detection, tracking, verification, and reporting.

This project is being developed with :

  • flask (python)
  • tweepy (python)
  • firebase (framework)
  • java (for android)
  • scikit-learn (python)
  • tensorflow (python)
  • tflearn (python)

What are we doing?

  • Gathering global information anywhere related to fire from social media such as Twitter
  • Developing a system for detecting tweets informing a fire incident with an accuracy of 98% using Natural Language Processing
  • Checking users location and evaluating the rate of danger
  • Getting alerts for probable fire incidents from users through pictures, videos, or texts
  • Image processing on user photos and Natural Language Processing on sent texts
  • Predicting prone positions for catching fire through exponential ascending temperature
  • Predicting the direction of fire extending and estimating its speed of spreading

Social Media

The goal is to include reports by people all over the world which are also not using our app but using social media to share fire incidents. We manage to filter tweet messages talking about a fire incident using Natural Language Processing (NLP) and obtain their locations[1]. In addition, a fire detection Convolutional Neural Network [2] (CNN) is used to verify the possible photos in the tweets.

User Reports

We have implemented a feature for all users to commit a fire incident through app easily by uploading related pictures and texts describing the situations. There is an idea to report data after our process to firefighting centres and letting them verify and publishing public warnings for any necessary considerations. Also, we created an awareness feature for users to inform any danger in their current location. Changing the background color shows the high possibility of danger.


Extracting valuable NASA’s API gives us access to a ten-minute-update of global map temperature. The exponential ascending temperature creates a high probability of fire dangers. We use this feature to predict any likely fire incidents.

In order to predict growth and existence of the fire, various information from social media, NASA satellites API, and notifications received from people using developed app combined together. NASA satellite API has the largest probability to show the existence of fire in an area. People can send a message or upload an image or video inside the developed application. If they send the message, their message processes and once detected as emergency fire it gets a high probability. On the other hand, If photo or video uploaded to the application, it processes with a convolutional neural network to extract context of the image and see whether it shows fire or not. After measuring how danger is the image, its probability uses as the weight. Social media (Twitter) messages are also being processed every 10 minutes to detect the location of tweets related to the fire. After extracting all these information, Density-Based Spatial Clustering and Application with Noise (DBSCAN) method applies to extract clusters demonstrating fire and nearby people notify to know the existence of fire.

In order to predict the fire, every ten minutes, using all the information, new cluster center measures and its direction, and speed for future steps predicts and we alert firefighters to handle an emergency situation.

It is worth mentioning that unfortunately, we didn't have enough money to buy VPS; however, all of our codes are available in a link below which are tested in local server.


[1] Schulz, A. Guckelsberger, Ch. Janssen, Frederik : “Semantic Abstraction for Generalization of Tweet Classification” , Semantic Web, 2015

[2] Dunnings and Beckon : “Experimentally Defined Convolutional Neural Network Architecture Variants for Non-temporal Real-time Fire Detection”, In Proc international conference on image processing IEEE, 2018

you can find our apk file in the following link :

Open Source Code :


SpaceApps is a NASA incubator innovation program.