Track 1: Crop Identification Using Satellite Data

Team Name: DAS

Our Journey

Inspiration

India’s agriculture is the backbone of food security, yet monitoring crops remains a major challenge due to fragmented farms, diverse cropping practices, and frequent cloud cover during the monsoon. We were inspired by the idea that satellite data + AI could empower farmers, policymakers, and researchers with reliable crop information. This problem felt meaningful because it connects technology with social impact: food security, climate resilience, and farmer empowerment.

What We Learned

Through this hackathon, we deepened our knowledge of multi-satellite remote sensing. We learned how SAR (Sentinel-1) can see through clouds, how multispectral imagery (Sentinel-2, ResourceSAT-2/2A) captures vegetation dynamics, and how temporal patterns like NDVI ($\frac{NIR - Red}{NIR + Red}$) reveal crop cycles. We also explored geospatial AI models, including CNN+LSTM hybrids and Transformers, and gained hands-on experience in preprocessing pipelines.

How We Built It

Our system processes raw satellite images through:

Preprocessing & Fusion – cloud masking, atmospheric correction, SAR speckle filtering
Feature Extraction – vegetation indices, radar backscatter, temporal patterns
AI Modeling – CNN+LSTM and Transformer-based models for classification
Deployment – a cloud-native pipeline with a web dashboard for AOI-based insights

Challenges We Faced

The main challenges were:

Data Heterogeneity: Indian farms are small and diverse, making model generalization difficult
Cloud Cover: Monsoon-season optical data is unreliable, forcing us to fuse SAR and optical
Computational Resources: Handling large multi-temporal datasets required efficient cloud pipelines

In the end, our project showed us that combining advanced AI with satellite data can bridge the gap between space technology and farmers’ fields. This experience taught us not only technical skills, but also the importance of designing solutions that are practical, scalable, and impactful.

Built With

agrifieldnet-india-dataset
bootstrap
cams/merra-2-atmospheric-correction-apis-visualization:-interactive-web-dashboard
cloud-storage-(aws/gcp/azure)-frontend:-vue.js-/-react
fastapi-(python)-ai/ml-models:-cnn-+-lstm-hybrids
gdal
geojson
geoserver
gis-compatible-exports-(geotiff
html/css/js-backend:-node.js
javascript-frameworks-&-libraries:-pytorch
ndvi-&-sar-feature-extraction-apis-&-data-sources:-sentinel-1-sar
numpy
openlayers-databases:-postgresql-with-postgis-cloud-services:-docker
programming-languages:-python
rasterio-geospatial-tools:-google-earth-engine
resourcesat-2/2a
saliency-maps
scikit-learn
sentinel-2-multispectral
serverless-apis
snap-toolbox
tensorflow
transformer-based-geospatial-foundation-models