RaymondDashWu / ai4crowdsourcing-wildfire-hackathon

A crowdsourcing app intended to allow users to be able to detect and report wildfires while incorporating machine learning

Crowdsourcing4Mankind

Frontend Contribution by Jennifer Ma:

https://github.com/jenniferhm

Machine Learning by Raymond Wu:

https://github.com/RaymondDashWu

Note: Model has been converted from PyTorch to ONNX to Tensorflow for the purposes on the hackathon. Things may not work! https://github.com/onnx/onnx-tensorflow

ONNX model in case things don't work: TODO Link

Note: This screenshot is not 100% representative of the end product. Could not find a way to expose the Swift class to React Native!

API Design by Hong Tran:

https://github.com/Jessiehongtran

Hong's branch can be found here: https://github.com/RaymondDashWu/ai4crowdsourcing-wildfire-hackathon/tree/image-API-branch

Description

Crowdsourcing4Mankind is a mobile app geared towards the early detection of wildfires. The goal is to reduce the time it takes for a fire to be reported to authorities after it begins. The app was created as part of a hackathon run by AI for Mankind. Specifically their Challenge IA: Smoke vs No Smoke using Entire Image

Disclosure:

Crowdsourcing4Mankind is not complete. Key features and their current status are highlighted below. Please clone and create a pull request if you would like to contribute.

How it Works:

• A user can use the app to find their location and see the likelihood of a fire forming in their area based on historical data and the current climate.
• If the user spots smoke/fire, a picture can be taken through the app.
• The image will be analyzed by the machine learning model, which will return a rating corresponding to the likelihood that there is a fire.
• If the image scores above 70%, the user has the option of sending the image to the local authorities or to retake the image.
• If the image scores below the threshold, then the user can choose to keep the image or delete it.

Primary Technologies:

• React Native
• Google Maps API
• React Native Maps
• Core ML
• PyTorch
• fastai

Key Features & Status:

• Location finder
  • Status: WIP - form needs to be completed and connected to backend
• Camera
  • Status: WIP - need to use react-native-camera or other camera API
• Polygon overlays of region with % of likelihood of fire based on current climate and historical data
  • Status: WIP - need to import historical fire data, temperature API, and add additional polygon overlays
• In-app ML model to recognize smoke in an image
  • Status: WIP
• Option to send image to local authorities if fire is detected
  • Status: WIP - need to add form to allow users to send image once threshold is reached
• Train model to detect wildfires
  • Status: Completed - model can be found here. Note that this is a PyTorch model.
• Convert PyTorch model to Tensorflow
  • Status: WIP
• Convert Swift Core ML example to work `with React Native.
  • Status: WIP