How to use k-means clustering to saparate the 4 groups of cells

so basically I have a image that was being masked

It consist on 4 groups of data P, N, poly and defects

This is P group

This is N group

This is poly group

and this is the defect.

How do I use kmean clustering to automatically separate the masked image to each individual image of P, N, poly and defect itself. Use the top 10% value of each cell for clustering data for accuracy.

The masked file is being attached here.