The Crop Classification model is built upon multi-temporal analysis of Sentinel-2 satellite imagery combined with vegetation growth indicators to distinguish crop types at the farm level. It follows a modular process pipeline that transforms raw input data into actionable crop labels, applicable across diverse geographic regions and growing conditions.

  • Input Processing : The model ingests farm boundaries along with user-specified crop season timeframes, extracting vegetation growth profiles derived from normalized difference vegetation indices (NDVI) over time. These temporal profiles capture phenological patterns — i.e., how vegetation develops and changes throughout the growing season.
  • Pre - Processing : Vegetation time series data are filtered and smoothed to focus on meaningful growth cycles representative of valid crops. This enhances signal quality for subsequent analysis while managing noise and data irregularities inherent to satellite observations.
  • Classification Paradigms :
    • In the supervised setting, the model leverages similarity measures between temporal vegetation profiles of unknown farms and those of known reference farms provided by the user, enabling direct crop type inference.
    • Alternatively, the unsupervised approach utilizes clustering techniques on the time-series vegetation growth profiles of farms to discover natural groupings based on phenological similarity, which is useful when labeled data is not available.
  • Scalability and Adaptability : The model architecture is designed to be adaptable across different agricultural regions and climate zones, enabling it to operate effectively on diverse crop types and phenological patterns found worldwide.
By incorporating mechanisms to preprocess and filter crop growth cycles, the model maintains robustness against noise and incomplete data, which are common challenges in remote sensing applications. The focus on meaningful vegetation growth periods ensures that the classification remains accurate and relevant for operational use cases like farm management, crop insurance assessments, and food supply chain planning.