Inefficiency in disaster response in low income countries, particularly in rural areas, causes tremendous loss because of difficulties in communication. And it is becoming even severer as global climate change continues.
Our idea is to improve disaster detection, evacuation and rescue process by analyzing satellite images.
What it does
Detect areas affected by disasters (so far only flood is supported)
How I built it
Built it using transfer learning, a technique used in machine learning
Challenges I ran into
Scrapping the web for satellite images Fine tuning parameters on a small dataset
Accomplishments that I'm proud of
Getting it to work in a day!
What I learned
Supervised techniques require a lot of data focus on unsupervised techniques for next steps
What's next for
Extend the model for earthquakes, eruptions, man-made disaster, wild-fires etc. Localizing detection to help in finding access routes. Divide areas with severity of event to allow for proper resource allocation