Inspiration
We built FarmScan to tackle one of India’s biggest agricultural challenges — accurate crop identification. With millions of small, fragmented farms and diverse conditions, traditional methods often fail. By combining satellites and AI, we aim to empower farmers, policymakers, and researchers with actionable crop insights.
What it does
FarmScan classifies crops across Indian states using radar and optical satellite imagery. It generates district-level crop maps, helping farmers with sowing, irrigation, and harvesting decisions, while supporting policymakers and agritech companies in planning and yield forecasting.
How we built it
Collected and preprocessed radar (Sentinel-1, EOS-4) and optical (Sentinel-2, ResourceSAT-2/2A) data.
Integrated AgriFieldNet India crop labels.
Extracted features like vegetation indices, radar backscatter, and seasonal growth patterns.
Built ML models and deep learning architectures (Temporal CNNs, Transformers).
Fused radar and optical data to handle cloud cover.
Fine-tuned the Prithvi foundation model for India-specific conditions.
Challenges
Handling large, diverse satellite datasets.
Managing cloud interference in optical imagery.
Adapting global AI models to fragmented Indian farms.
Computational limits for temporal deep learning.
Accomplishments
Created accurate district-level crop maps.
Improved accuracy by fusing radar and optical data.
Successfully fine-tuned Prithvi for local conditions.
Developed a scalable workflow for multiple states.
What we learned
Working with multi-source satellite data.
Benefits of combining radar and optical imagery.
Implementing temporal deep learning models.
Adapting foundation models to regional challenges.
Next steps
Expand coverage pan-India.
Partner with government and agritech firms.
Improve real-time efficiency and farmer-friendly tools.
Launch web or mobile platforms for direct farmer access.
