Climate-Vision/ClimateVision

What it solves

ClimateVision automates the detection of environmental changes from satellite imagery, removing the need for manual analysis. It specifically targets the monitoring of deforestation, arctic ice melting, and flooding to provide conservation NGOs and researchers with timely, automated alerts.

How it works

The platform integrates with Google Earth Engine to pull Sentinel-2 and Landsat data, which is then processed through a pipeline involving cloud masking, normalization, and tiling. It employs deep learning architectures, specifically U-Net for semantic segmentation and Siamese networks for change detection, to analyze the imagery. The system is delivered as a FastAPI REST API with a React-based dashboard for visualization and monitoring.

Who it’s for

It is designed for conservation NGOs and environmental researchers who need to monitor specific geographic regions for climate-related changes.

Highlights

• Multi-hazard monitoring: Active detection for deforestation, ice melt, and floods, with drought and wildfire detection planned.
• Automated alerting: Allows organizations to subscribe to regions and receive alerts via webhooks or email based on defined thresholds.
• Production-ready ML: Includes MLflow for experiment tracking and ONNX export for optimized inference.
• Interactive Dashboard: Features a React frontend with a bounding box region selector, analytics charts, and confidence gauges.