Big Blue Crew

From Space to Soil

Submitted to

IndiaAI Impact Gen-AI Hackathon

Created by

I contributed by processing Sentinel-2 data, engineering vegetation indices, and extracting field-level features. I developed and fine-tuned gradient boosting models (XGBoost, CatBoost, LightGBM) and integrated a U-Net for pixelwise classification. I designed the ensemble blending strategy using a weighted geometric mean to improve robustness. I also handled data imbalance with SMOTE and clustering, visualized feature importance and performance metrics, and ensured smooth integration of the entire pipeline.

ABHINAV RAJPUT
Aman kumar
Rudransh Singh Rathore

Updates

ABHINAV RAJPUT posted an update — Aug 30, 2025 10:48 AM EDT

Big Blue Crew Update

We’ve been making exciting progress on crop classification using Sentinel-2 satellite data. Here’s what’s new:

Improved ensemble blending with weighted geometric mean, boosting accuracy across smallholder farms.
Added more vegetation indices (like RECI, CVI, and EVI2) for better crop differentiation.
Optimized our U-Net model and other gradient trees for pixelwise classification on large Sentinel-2 tiles.
Built cleaner visualizations of feature importance and model outputs.
Tackled class imbalance using SMOTE and clustering to enhance minority crop detection.

Log in or sign up for Devpost to join the conversation.

ABHINAV RAJPUT started this project — Aug 30, 2025 10:45 AM EDT

Leave feedback in the comments!

Log in or sign up for Devpost to join the conversation.