Inspiration
Small farms produce nearly 35% of the world’s food, yet most of them are in low- and middle-income countries where reliable agricultural data is scarce. This lack of data makes it difficult for governments and organizations to provide timely support, monitor crop health, and ensure food security. We were inspired to leverage satellite data and AI to create a scalable solution for monitoring crops in India, where smallholder farming dominates.
What it does
Our solution uses multispectral satellite imagery from Sentinel-2 combined with machine learning models to classify different crop types across smallholder farms in Northern India. By doing so, it helps in generating crop maps that can support agricultural planning, disaster risk management, and policy-making.
How we built it
-Collected Sentinel-2 multispectral imagery for districts in Uttar Pradesh, Rajasthan, Odisha, and Bihar.
-Preprocessed the data using Google Earth Engine and Python libraries.
-Extracted vegetation indices like NDVI and spectral band features.
-Trained a machine learning model (Random Forest / XGBoost / CNN) for crop classification.
-Validated results against the AgriFieldNet India dataset.
Challenges we ran into
-High cloud cover during monsoon season reduced the availability of clean optical data.
-Small and fragmented farm sizes made classification harder.
-Limited labeled ground-truth data for certain crop types.
-Balancing computational cost with the large volume of satellite imagery.
Accomplishments that we're proud of
-Successfully created a working crop classification pipeline using Sentinel-2 data.
-Achieved promising accuracy in differentiating multiple crop types across diverse regions.
-Built a scalable solution that can be adapted to other smallholder farming regions globally.
What we learned
-How to handle and preprocess large-scale satellite data using Google Earth Engine.
-The importance of combining different features like spectral indices and temporal data for better classification.
-How agricultural challenges (cloud cover, small fields, mixed cropping) directly affect AI model performance.
What's next for AgriVision : “Smart Crop Mapping for Small Farms”
-Incorporate Sentinel-1 SAR data to overcome cloud cover issues.
-Expand to include real-time monitoring and early warning systems for floods/droughts.
-Build a web dashboard for farmers and policymakers to visualize crop type predictions.
-Scale the project to other regions in India and eventually to other developing countries.
Log in or sign up for Devpost to join the conversation.