Classification

• In situ data preparation
  • 1. Apply 10m inner buffer to each polygons
  • 2. Add polygon area
  • 3. Add pixel count
  • 4. Remove unnecessary in situ polygons
  • Plot histogram
• In situ split calibration-validation
  • 1. Open in situ data prepared
  • 2. Add number of pixel of each class for each polygons
  • 3. Split polygons in training and validation datasets
• In situ split calibration-validation with complex strategies (optional)
  • 1. Open in situ data prepared
  • 2. Get “pixel ratio” for each polygons
  • 3. Assign sampling design strategy for each polygons
• Random Forest Classification
  • 1. Prepare classification features associated to in situ data
  • 2. Train the Random Forest
  • 3. Predict the rest of the image
  • 4. Reclassify classification
  • 5. Filter classification with moving window
  • 6. Write classification products into GeoTIFF files
• Validation
  • 1. Prepare data
  • 2. Confusion Matrix
  • 3. Accuracy metrics