Training classifier on multiple images


I am trying to develop a supervised KDTree classifier to use for the purpose of classifying multiple images. At the moment I’ve trained it on one image, but land cover types across other images are diverse, and the classification results are not good enough. Is it possible to train the classifier across multiple images to increase its effectiveness?


You can create vector containers from different images and import them into your model for training.

Could you please tell me how to do so? When I try to import 2 images with vector containters into the classifier, I get “Source products are of different dimenstions” error.

If you are working with several images on different areas, it would be better to mosaic the scenes first. If you have several images of the same area on different dates, you should be able to import the vectors in the classifier window.

The scenes are from different dates and are really far away from each other. Making a mosaic resulted in an unbearably big file with a bad spatial resolution, so my it’s not a viable solution in my case.

Maybe you could train your model with a multitemporal scene that contains all types of classes and their variability, and apply the model to the other images later.