An advanced system for near-real-time greenhouse gas (GHG) anomaly detection using a fusion of Edge AI and satellite data.
Airguard VisionEdge empowers environmental researchers and on-site analysts to detect, visualize, and interpret GHG anomalies with unprecedented speed and accuracy. By fusing powerful, low-power Edge AI with comprehensive satellite imagery, VisionEdge provides actionable insights directly in the field.
- On-Device AI: Runs optimized TinyML models on edge nodes for local, real-time emission pattern detection.
-
Sensor Fusion: Combines satellite raster tiles with ground sensor data (
$CO_2$ ,$PM_{2.5}$ , etc.) for higher confidence alerts. - Interactive Dashboard: An Android-first web dashboard visualizes fused insights through maps, charts, and anomaly markers.
- Deep Analysis: Seamlessly syncs with Google Colab notebooks for advanced analytics and deep-dive visualization.
- AI-Powered Assistant: An on-device LLM, the VisionEdge Copilot, explains observed trends and provides context for data anomalies.
The system is designed with three core, interconnected layers:
-
Edge Impulse Node:
- Runs highly optimized Machine Learning models (TinyML) for local emission pattern detection.
- Performs direct inference on satellite raster tiles and live ground sensor data.
- Operates with low power, enabling deployment in remote locations.
-
Web Dashboard (Android-first):
- Displays fused insights through interactive maps, charts, and clear anomaly markers.
- Syncs with Google Colab notebooks for advanced, customizable analytics.
- Provides a central hub for monitoring all connected edge nodes.
-
LLM Assistant (VisionEdge Copilot):
- An on-device Large Language Model assistant that interprets and explains observed data trends in natural language.
- Recommends research insights and provides crucial context for data anomalies (e.g., "This spike correlates with regional agricultural burn patterns observed last year.").
- Secure sign-in via Google or institutional accounts.
- Sync connected Edge Impulse devices via Bluetooth/WiFi.
- The "Add New Station" feature automatically detects and registers a local AI node.
-
Top Bar: Quick filters for
Region | Model | Timeframe. -
Live Map Panel: Displays satellite raster data with overlay layers for GHG,
$NO_2$ , and temperature. Edge inferences are highlighted as colored, pulsating hotspots. -
Mini Stats Bar: Shows key metrics like
Emission Index,Confidence Level, andAnomaly Count. Tapping expands the view.
- Organized into clear tabs:
AI Inference|Time Series|Correlations|Ground Data. - Features interactive plots generated directly from edge node outputs.
- An "Open in Colab" button instantly launches a pre-populated notebook session for deeper analysis.
- A floating chat widget allows users to "Ask VisionEdge Copilot."
- Users can query the system with natural language, e.g., “Explain today’s emission spike in the Cairo region.”
- Download comprehensive reports as PDF or GeoTIFF files.
- Push results directly to a shared research group or an institutional drive.
- Theme: A modern space black background with vibrant green-cyan gradients to represent emission heatmaps.
- UI Style: Sleek and minimal, following Material 3 design principles with a clean, card-based layout.
- Data Visualization: 2D raster overlays with opacity controls and dynamic graphs for comparing local inferences with historical data.
Instructions for setting up the project locally will be added here.
View interactive mockup designing or use the repository index.html
Contributions, issues, and feature requests are welcome! For significant contributions, please open an issue first to discuss what you would like to change.