Improving Machine Learning Training Data

Zach Torres
Zach Torres

November 18, 2022

Improving Machine Learning Training Data

In this project, I called the Google Earth Engine API within the JavaScript code editor to use cloud and date filtered imagery from Sentinel-2 . I then built training data points and added to the feature engineering table for improved supervised classification of landcover compared to training data on reflectance values at points alone.


My workflow is outlined below:

  1. Import Urban Areas feature collection and filter system:index to desired feature
  2. Import ALOS elevation surface model
  3. Import Sentinel-2 harmonized data
  4. Run cloud mask
  5. Filter the image collection to desired date, bounds, and bands
  6. Define a median and clipped Sentinel-2 composite
  7. Define visualization parameters
  8. Add composite to map
  9. Create training data points for four classes as feature collections and export to Assets
  10. Add relevant indices bands to composite image
  11. Add elevation and slope bands from ALOS to composite
  12. Normalize the data
  13. Merge the classifications as individual GCPs
  14. Train a classifier
  15. Classify the improved composite image and add new layer to map


Check out the demo link below to review and run my script.

Demo

Tools used

ALOS PRISMGoogle Earth Engine(GEE)sentinel-2

Plug-ins used

Javascript

tags

image classificationland coverRemote Sensing

You might also like

Join the community!

We're a place where geospatial professionals showcase their works and discover opportunities.