Satellite-Enabled Precision Agriculture: Machine Learning Classification of Cropland Using Sentinel-2 Imagery

A machine learning project that classifies agricultural cropland versus non-cropland areas using Sentinel-2 satellite imagery and spectral analysis techniques. This project demonstrates the application of remote sensing and machine learning techniques to classify land use patterns in the Brong-Ahafo region of Ghana. Using Sentinel-2 satellite data, the analysis achieves 100% accuracy in distinguishing between agricultural cropland and non-cropland areas through spectral band analysis and vegetation indices.

Technologies & Tools

Remote Sensing & Data Collection

• Google Earth Engine (GEE) - Cloud-based platform for satellite data processing
• Sentinel-2 Satellite Imagery - European Space Agency's multispectral satellite data
• JavaScript - GEE scripting for data extraction and preprocessing

Data Analysis & Machine Learning

• Python 3.8+ - Primary programming language
• Jupyter Lab/Notebook - Interactive development environment
• Pandas - Data manipulation and analysis
• NumPy - Numerical computing
• Scikit-learn - Machine learning algorithms and evaluation metrics

Visualization & Analysis

• Matplotlib - Static plotting and visualization
• Seaborn - Statistical data visualization
• PCA - Principal Component Analysis for dimensionality reduction

Dataset

• Source: Sentinel-2 Level-2A surface reflectance data via Google Earth Engine
• Study Area: Brong-Ahafo Region, Ghana
• Time Period: 2023
• Sample Size: 600 data points (300 crop, 300 non-crop)
• Features: 6 spectral bands + 3 derived vegetation indices
• Balance: Perfectly balanced dataset (50% crop, 50% non-crop)

For data collection (optional)

• Open Google Earth Engine Code Editor
• Copy and Run script: brong_ahafo_crop_classification.js file

Spectral Bands Used:

• B2 (Blue): Atmospheric and water analysis
• B3 (Green): Vegetation and water features
• B4 (Red): Chlorophyll absorption
• B8 (NIR): Vegetation health and biomass
• B11 (SWIR1): Soil and vegetation moisture
• B12 (SWIR2): Geological features

Key Insights

• Vegetation indices are the most discriminative features (78.5% total importance)
• Perfect class separation achieved through spectral differences
• NDWI shows highest importance, indicating water content differences
• Clear spectral signatures between crop and non-crop areas