PrakritiLens - AI-powered crop identification

Inspiration

India’s farms are like a mosaic of millions of tiny fields, each with its own story. Unlike the vast, uniform farms of the West, these fields are fragmented, diverse, and dynamic. We asked ourselves: “Can satellites, orbiting hundreds of kilometers above, really understand this complexity?” That question became our spark. That’s why we re-imagined crop monitoring for the tiny, fragmented fields of India—something most global models aren’t built for.

What it does

PrakritiLens uses satellite data + AI to automatically identify crop types in India’s small, fragmented farms.

-Inputs → Multispectral (Sentinel-2, ResourceSAT) + SAR (Sentinel-1, EOS-4) imagery.

-Preprocessing → Data cleaning, feature extraction (NDVI, EVI, SAR backscatter), and temporal analysis across crop cycles.

-Model → Machine learning & fine-tuned geospatial foundation models classify fields into crop types.

-Outputs → A color-coded crop map and insights dashboard that show which crops are grown in each region (e.g., wheat, rice, mustard). Unlike conventional tools, PrakritiLens works on sub-acre fragmented fields and explains its predictions visually—so even smallholder farmers can trust and act on the insights.

How we built it

-Data Harmonization – stitched SAR + multispectral signals into temporal “cubes” of each farm.

-Feature Extraction – NDVI, EVI, backscatter, growth stages.

-Modeling – from quick wins (XGBoost) to deeper exploration (CNNs/LSTMs for time series, foundation models like Prithvi).

-Demo – a visual dashboard where users can pick a district and see predicted crop maps appear — turning abstract pixels into real farming insights. This makes PrakritiLens one of the few solutions purpose-built for smallholder farms, rather than retrofitting global models.

Challenges we ran into

-Clouds vs. Crops – optical data blocked by clouds forced us to rely on SAR as a complement.

-Fragmented Fields – where one farmer grows wheat, the neighbor grows mustard — models must separate these fine lines.

-Generalization – training in UP and predicting in Odisha is no small feat.

-Time Crunch – balancing deep ML experiments with hackathon deadlines.

Accomplishments that we're proud of

-Decoding satellite signals – We successfully combined SAR and multispectral data to create a unified dataset despite noise, cloud cover, and fragmented field boundaries.

-Working prototype in limited time – Built a baseline-to-advanced pipeline (from XGBoost to deep models) that already shows promising classification results.

-From zero to domain understanding – In just days, we learned how vegetation indices, temporal signals, and foundation models can be harnessed for Indian agriculture.

-Impact-first mindset – Designed not just code, but a demo dashboard that makes our model’s predictions easy to understand for farmers and policymakers.

-Collaboration & speed – As a 3-member team, we divided roles efficiently and delivered faster than expected.

What we learned

We discovered that satellite data is more than just images — it’s hidden language.

-SAR shows how fields feel (structure, moisture).

-Multispectral shows how they look (color, vegetation health).

-Together, they reveal a crop’s fingerprint, if decoded correctly. We also learned the importance of temporal rhythms — a paddy field in August doesn’t look like one in November.

What's next for PrakritiLens - AI-powered crop identification

-Expand Coverage – Extend the model beyond Northern India to pan-India, fine-tuning for local cropping patterns and diverse agro-climatic zones.

-Yield & Stress Prediction – Move from “what crop is here” to “how much yield will it produce” and “what stress (drought, pest, flood) is it facing?”

-Farmer-Facing Tools – Integrate with FPOs and government portals, offering field-level insights through mobile dashboards in regional languages.

-Climate & Policy Applications – Support policymakers in monitoring food security, crop insurance, and climate adaptation strategies.

-Global Scalability – Build PrakritiLens as an India-first solution that can adapt to smallholder farms across Africa, SE Asia, and beyond.