AgriFieldNet India

Inspiration:

Agriculture is the backbone of India’s economy, yet smallholder farmers face challenges in crop monitoring and planning due to fragmented landholdings and diverse cropping patterns. Unlike large farms in the USA or Europe, Indian fields require specialized solutions. Inspired by this challenge, we set out to build a scalable crop identification system that leverages multi-sensor satellite data and modern AI models to better understand Indian farmlands.

What it does:

This solution classifies crop types across fragmented agricultural fields in Northern India using Sentinel-1 SAR, Sentinel-2 optical, and ResourceSAT-2 observations. By fusing structural and spectral information, the system learns to differentiate between crop types and provides accurate crop maps at the field level. This can support policymakers, researchers, and farmers in making data-driven agricultural decisions.

How I will build it:

-Data Integration – We combined SAR and multispectral imagery from Sentinel-1, Sentinel-2, and ResourceSAT-2. -Preprocessing – Applied cloud masking, radiometric calibration, and temporal alignment. -Modeling – Fine-tuned a geospatial foundation model using the AgriFieldNet India dataset enhanced with field-collected ground truth. -Training & Evaluation – Trained on diverse districts across Uttar Pradesh, Rajasthan, Odisha, and Bihar to ensure generalization. -Validation – Compared predictions with labeled datasets and optimized for accuracy across different crop types.

Challenges:

-Handling fragmented field boundaries and mixed cropping. -Integrating multi-sensor data with varying resolutions and noise. -Cloud cover in optical data required relying more on SAR backscatter. -Achieving generalization across diverse agro-climatic regions.

What's next for AgriFieldNet India:

-Expand crop coverage to include more varieties across different seasons. -Integrate weather and soil data for richer context and improved predictions. -Build a real-time monitoring platform accessible to farmers and policymakers. -Collaborate with government agencies and NGOs to scale impact on agricultural planning, food security, and climate resilience.

Built With

• datasets:-agrifieldnet-india-(radiant-earth-foundation-+-ibm-research-enhancements).-sentinel-1-(sar)
• gdal
• hugging-face-transformers
• lightning
• opencv.-visualization:-matplotlib
• pandas
• rasterio
• resourcesat-2/2a-imagery.-tools-&-libraries:-geospatial-&-remote-sensing:-google-earth-engine
• scikit-learn
• seaborn
• sentinel-2-(optical)
• sentinel-hub.-ml/deep-learning:-pytorch
• torchgeo.-preprocessing:-numpy