I've performed unsupervised classification (K-means) on a hyperspectral image. I had got 10 different classes and I brought it to bring it down to 6. For some parts of the image, the merging of the classes has worked out just fine, but in other parts of the classified image, even though I know that certain regions in the image belong to the same class, but they have been classified into multiples classes. I can't merge these classes, because they'll affect the rest of the image. So, my question is can I merge different classes in certain areas of the image, without affecting the rest of the image? Thank you!!