I am trying to do unsupervised classification on time series hyperspectral bands using K-means clustering techniques in Qrfeo toolbox plugin of Qgis to mask out shadow area.

But my problem is I am not able to compare the results to each other because the classes assigned are different from image to image.

Could any one of you help to resolve this issue?

  • Can you add some more information, maybe screenshots to help everyone understand the problem? Welcome to GIS SE. please take the tour. gis.stackexchange.com/tour – Mattropolis Jun 30 '17 at 19:44
  • I am trying to mask out shadow of plants from rest of the classes such as vegetation, soil etc., in time series hyperspectral images. I got reliable result from k means clustering technique in orfeo toolbox. But my problem is how can I get same class number for shadow region for all the classified output generated from time series data using this algorithm. – Shree Jun 30 '17 at 20:03

I think that to be able to compare results you need to ensure that the number of classes and the initialization are identic between each run. The K Means application you used in OTB use random initializations which leads to different results.

Since last stable release of OTB you can use the pixel based classification framework including the sampling strategy initially used for supervised classification methods to sample, initialize and run kmeans classification:


Note to finish that there is ongoing development in OTB 6.2 to update the kmeans application to provide a "all in one" application to perform unsupervised classification and giving access to kmeans parameters:


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.