Create unsupervised classification

Use the Create unsupervised classification block to classify images without supplying training data. 

This block runs a cluster analysis to look at the pixels in the images and identify where pixels cluster together with similar data properties. These clusters of pixels are then assigned a class.

Algorithm

Earth Engine uses the Weka software to implement unsupervised classification using optimised machine learning. Weka refers to a collection of algorithms made available by Waikato University. The classification methods available are:

• Cascade K Means
• Cobweb
• K Means 
• LVQ
• X Means

Click on the cog icon to see the parameters available to refine the algorithm. 

Setting the number of classes

You may want to experiment with the Number of classes variable. LVQ and K Means will ask for an exact number of classes, which is appropriate if you already have some idea of how many classes you expect to see in the Area of Interest. Cascade K Means and X Means give you the option to specify a minimum and a maximum number of classes, giving the algorithm some flexibility in the analysis of the clusters.  

Input data

To use a single dataset as an input, drag and drop the Use this dataset container block into the Create unsupervised classification block.

To add multiple datasets as inputs, use the Reuse saved dataset block.

Outputs

You can output the results of the classification as a map layer, table, or dataset.

Learn more: How to run a classification