DisastersAI

Inputted Image
User can choose from types of forests
Probability prediction
Mark Predicition

Inspiration

This past summer wildfires spread through thousands of acres across Northern California, leaving people with no other choice than to evacuate. It is crucial for these disasters to be predicted on short notice and with great accuracy, in order to ensure not only public safety, but the safety of our environment and Earth. Currently there are thousands of satellites orbiting the Earth, however, only less than 10 of them can detect wildfires, and because of these scarcely small number, wildfires have continued to persist as a huge issue for years on end. This immense problem led me to create DisastersAI, a Real Time Wildfire Detection system that uses AI to instantaneously detect wildfires around the globe from satellite imagery, in order to instantly alert citizens to evacuate, as well as predict the negative consequences of the event.

What it does

DisastersAI takes a user inputted satellite image, and utilizes a deep learning model to determine the predicted probability, mask, CO2 emission, total burnt area, and the amount of daily electricity power will be wasted during the event of the inputted wildfire.

How I built it

First I used satellite images of wildfires to train the deep learning model. The model created was a simple architecture with 17 input features, the first hidden layer with 64 neurons, the second hidden layer with 32 neurons, and the final regression node. The final model was then saved on s3 and deployed on the front-end which was developed using python's streamlit API.