Gen AI hackathon - T1

Inspiration

As MSc Geoinformatics students at SPPU, this is our very first hackathon experience.
We were inspired by the importance of agriculture in India and the role of geospatial technology in addressing challenges like food security and resource management.
The specific problem of crop type identification caught our attention since it can help in monitoring agricultural patterns in the North Indian region using SAR and optical datasets.

What it does

Our project aims to classify crop types by integrating:

• Synthetic Aperture Radar (SAR) data for structural crop information.
  
• Optical datasets for spectral indices and vegetation health assessment.
  

By combining these datasets, we aim to develop a workflow that can provide accurate crop type maps to support agricultural planning and decision-making.

How we built it

Although the implementation is ongoing, our methodology involves:

1. Data acquisition: Collecting SAR and optical datasets (e.g., Sentinel-1, Sentinel-2).
   
2. Preprocessing:
   
   • SAR calibration and speckle noise reduction.
     
   • Atmospheric correction for optical imagery.
     
   • Harmonization of spatial resolution and projections.
     
3. Feature extraction:
   
   • Vegetation indices (e.g.,
     $$ NDVI = \frac{NIR - Red}{NIR + Red} $$).
     
   • Backscatter coefficients from SAR data.
     
4. Classification: Designing supervised learning models to identify crop types.
   

We use Python with libraries like Rasterio and GeoPandas, and plan to integrate machine learning frameworks in the next stage.

Challenges we ran into

• First time working with coding for spatial datasets, despite prior theoretical knowledge.
  
• Handling large raster and vector data formats (.tif, .shp).
  
• Limited experience in integrating AI/ML frameworks with geospatial workflows.

Accomplishments that we're proud of

• Designed a structured methodology for crop type identification.
  
• Learned to practically use Rasterio and GeoPandas for spatial data handling.
  
• Successfully transitioned from theory to a coding-first approach in geoinformatics.

What we learned

• Practical coding with spatial data is significantly different from theoretical study.
  
• Preprocessing SAR and optical datasets is crucial for reliable outcomes.
  
• AI integration offers great potential but requires careful data preparation.
  
• Teamwork and problem-solving under time constraints.

What's next for Gen AI hackathon - T1

• Implement the drafted methodology with actual datasets.
  
• Extend the workflow to handle larger regions and multiple crop types.
  
• Scale the solution for real-world applications in crop yield estimation, and agricultural planning.

Built With

• machine-learning