🌾 #H1 CropMosaicAI 🌱 Inspiration In India, most farms don’t grow just one crop — they grow a blend of two or more within the same field. Yet, most satellite-based crop mapping models are designed for monocropping. This mismatch inspired us to build a solution that reflects India’s mixed-farming reality, empowering farmers, policymakers, and agri-tech companies with more accurate insights.
🌾 What it does CropMosaicAI detects intercropping and mixed farming patterns within the same field boundary by: Using spectral unmixing + temporal SAR data to separate crop signals. Providing multi-crop classification instead of single-crop assumptions. Enabling accurate yield estimation, fairer crop insurance, and better policy design.
🛠️ How we built it Data Collection – Satellite SAR + spectral imagery with ground-truth references. Spectral Unmixing – Instead of one label per pixel, we decomposed signals into fractions of multiple crops. Classification Pipeline – Combined supervised learning + clustering to detect intercropping. Validation – Tested on Indian farm data across different crop seasons.
⚡ Challenges we ran into Limited availability of labeled datasets for intercropping. Complex & noisy SAR signals needing heavy pre-processing. High computational cost of spectral-temporal analysis. Ensuring generalization across regions and seasons.
🏆 Accomplishments that we're proud of Built a first-of-its-kind pipeline tailored to Indian farming patterns. Successfully applied spectral unmixing to real-world farm signals. Demonstrated potential for policy-level impact in yield, insurance, and subsidy systems. Learned to integrate AI + remote sensing + agriculture into one solution.
📚 What we learned The importance of ground realities vs. academic datasets. Combining SAR + spectral imagery can unlock powerful agri-insights. Cross-domain collaboration (AI, agriculture, policy) strengthens outcomes.
🚀 What's next for CropMosaicAI Expand datasets across Indian states & crop seasons. Improve scalability for real-time farm advisories. Partner with agri-tech startups, insurers, and policymakers. Build a farmer-facing platform for localized, multi-crop insights.
✨ Tagline: “Mapping the real farm, not just the field.”
Built With
- geopandas
- google-earth-engine
- jupyter
- python
- scikit-learn
- sentinel-1/2-data
Log in or sign up for Devpost to join the conversation.