More than 4.5 million acres of land have burned on the West Coast in the past month alone Experts say fires will worsen in the years to come as climate change spikes temperatures and disrupts precipitation patterns Thousands of families have been and will continue to be displaced by these disasters
What it does
When a wildfire strikes, knowing where there are safe places to go can bring much-needed calm in times of peril Mildfire is a tool designed to identify higher-risk areas with deep learning analysis of satellite data to keep people and their families out of danger Users can place pins at locations of themselves or people in distress Users can mark locations of fires in real time Deep learning-based treetop detection to indicate areas higher-risk of forest fire Heatmap shows safe and dangerous zones, and can facilitate smarter decision making
How I built it
User makes a GET request w/ latitude/longitude value, which is then handled in real time, hosted on Google Cloud Functions The request triggers a function that grabs satellite data in adjacent tiles from Google Maps Static API Detects trees w/ RGB data from satellite imagery using deep-learning neural networks trained on existing tree canopy and vegetation data (“DeepForest”, Weinstein, et al. 2019) Generates a heat map from longitude/latitude, flammability radius, confidence from ML model Maps public pins, Broadcasts distress and First-Responder notifications in real-time Simple, dynamic Web-interface
Challenges I ran into
Completely scrapped mobile app halfway through the hack and had to change to web app.
Accomplishments that I'm proud of
Used a lottt of GCP and learned a lot about it. Also almost finished the web app despite starting on it so late. ML model is also very accurate and useful.
What I learned
A lot of GCP and ML and Flutter. Very fun experience overall!
What's next for Mildfire
Finish the mobile and web app