Usage

1. Place your satellite image tiles in the data/satellite directory
2. Place training data tiles in the data/training directory
3. Place mask files in the data/mask directory
4. Update the paths in the script:

   dir_sat = "your_path\\"
   dir_train = "your_path\\"

5. Run the script to:
   • Train the Random Forest model
   • Generate classified map tiles
   • Create probability map tiles

Model Details

The Random Forest classifier is optimized using RandomizedSearchCV with the following parameters:

• Number of trees: 200-2000
• Maximum depth: 5-25
• Minimum samples split: 2-15
• Minimum samples leaf: 1-10
• Maximum features: 'log2', 'sqrt'
• Bootstrap: True/False

Outputs

The script generates:

1. Classified image tiles
2. Probability map tiles
3. Model performance metrics
4. Feature importance analysis
5. Confusion matrix visualization

Data Format

• Input images: GeoTIFF format
• Training data: GeoTIFF format
• Output classifications: GeoTIFF format
• Probability maps: GeoTIFF format (float64)

Notes

• Make sure to have sufficient disk space for output files
• Processing time depends on the number and size of input tiles
• The script includes memory management for large datasets
• Model parameters can be adjusted in the hyperparameter section

Performance

The model's performance can be evaluated through:

• Classification accuracy
• Confusion matrix
• Feature importance analysis
• Cross-validation results

Tips

• Ensure consistent CRS across all input data
• Check memory requirements for large datasets
• Monitor processing progress through console outputs
• Backup important data before running large batch processes