Crop Classification with ML

This project applies Random Forest classification to analyze and classify NDVI (Normalized Difference Vegetation Index) raster images specifically for crop types throughout the year.

Folder Description

This project consists of several folders that contain the necessary data and scripts for the classification process:

• 0_NDVI_Rasters: Contains NDVI raster images for each month from Sentinel 2. However, due to their large file sizes, these raster files cannot be shared in this GitHub repository.
• 1_Training_Samples: Holds training and testing shapefiles with known crop classes.
• 2_Outputs: Stores the output classified raster images and results.

Key Steps

1. NDVI Rasters

• Load NDVI raster images for each month (January to December).

2. Data Preparation

• Stack the NDVI rasters into a multi-layered dataset.
• Use training data (sample points with known crop classes) and testing data (points for validation).

3. Training the Models

• Train a Random Forest model using the stacked NDVI data and training samples.
• Train an XGBoost model using the same data for comparison.
• Both models learn to classify different crop types based on the input NDVI values.

4. Reducing Sample Size

• Reduce the training data to 10% of the original size while maintaining representative samples from each class.

5. Model Evaluation

• Evaluate the model's performance by checking the Out-of-Bag (OOB) error for Random Forest and using cross-validation for XGBoost.
• Analyze the importance of different variables (monthly NDVI data) in crop classification for both models.

6. Image Classification

• Apply the trained Random Forest and XGBoost models to classify the entire NDVI raster stack.
• The results are classified images that map different crop types.

7. Accuracy Assessment

• Validate the classifications using testing data.
• Use Confusion Matrices to evaluate the performance of both models in predicting each crop class.

8. Output

• Save the final classified images as GeoTIFF files for further analysis.

---

In summary, this project uses satellite NDVI data and machine learning to map different crop types over time, producing a classified raster image of the area. This project was part of a practical exercise for the Advanced Image Analysis elective course at ITC, University of Twente.