Crisis Vision AI

CrisisVision AI: Real-time wildfire detection, alerting, and data insights powered by AI and satellite intelligence

Comment

Inspiration

I am from Nepal, where wildfires are a recurring and devastating issue. Every year, we lose forests, wildlife, property, and even human lives to wildfires that are beyond our control. This isn’t just a local problem—Australia, the Amazon Rainforest, California, and many other parts of the world face similar catastrophes due to wildfire outbreaks.

This inspired me to build a system that not only visualizes wildfire data, but also empowers users to take action, contribute to early detection, and potentially save lives.

What it does

CrisisVision AI is a real-time wildfire detection and alert platform that combines satellite data, machine learning, and user participation to tackle global fire disasters.

Key features include:

  • Live Map showing the 200 most recent fire events using NASA VIIRS API, filterable by region, timeframe, and crisis type.
  • Data Charts (line, bar, pie) to analyze trends, compare regions, and understand the distribution of wildfires and other crises.
  • AI Image Upload: Users can upload fire images. If the model detects fire with more than 90% confidence, it sends location-based alerts to others nearby.
  • Downloadable Reports: Users can export insights as PDF or CSV for analysis or distribution.
  • All data is stored in MongoDB Atlas, and services are powered by Google Cloud, Node.js, Flask, and React + Firebase.

How we built it

  • Built the frontend with React, hosted on Firebase.
  • Used NASA VIIRS API for real-time fire detection data.
  • Stored fire metadata and user inputs in MongoDB Atlas, enabling fast geo-queries.
  • Developed a Node.js backend to fetch and process fire data.
  • Created a custom fire detection model trained using both Infrared (IR) and RGB images.
  • Implemented Learning Without Forgetting (LwF) to retain IR knowledge while fine-tuning on RGB data, so the model performs well in both day and night conditions.
  • Deployed the AI model via Flask on Google Cloud Run.
  • Used Google Cloud Storage to manage uploaded images and processed results.

Challenges we ran into

  • Integrating thermal (IR) and RGB data into a single robust model required thoughtful training and architecture.
  • Learning and implementing the Learning Without Forgetting algorithm for cross-modal learning was both complex and rewarding.
  • Handling large, geospatial datasets in MongoDB and optimizing performance for real-time querying.
  • Building a full-stack, cloud-native, AI-powered platform on a tight timeline while ensuring a clean and responsive UI.

Accomplishments that we're proud of

  • Successfully implemented a custom dual-trained AI model capable of detecting fire in both RGB and IR images.
  • Enabled real-time community alerts based on AI predictions and user location.
  • Created a platform that doesn’t just present data—but allows users to interact, respond, and contribute.
  • Built a full cloud-hosted stack integrating MongoDB, Google Cloud, Flask, Node.js, and React.

What we learned

  • Advanced model training techniques like Learning Without Forgetting (LwF) for multi-phase training.
  • How to manage geospatial fire datasets and optimize data flow from backend to frontend using MongoDB Atlas and Google Cloud.
  • The importance of intuitive UI/UX when presenting disaster-related data to users.
  • How to make AI practically useful—by connecting it to real-world actions like alerting people in crisis zones.

What's next for CrisisVision AI

  • Add support for other disasters such as floods, earthquakes, and pollution using similar real-time data feeds.
  • Integrate SMS/email-based alert systems for broader community outreach.
  • Implement vector search in MongoDB Atlas to improve fire pattern detection.
  • Collaborate with disaster relief agencies to deploy the app on a national or international scale.
  • Launch a mobile app version for on-the-go fire reporting and alerts.

Built With

Share this project:

Updates