if it is correct that Maximum Liklihood classification algorithm takes into account RGB values of the bands only, then is it useless to stack other bandslike NIR SWIR etc prior to go ahead and use stacked image to retrieve landcover classes?

  • 1
    You need to state what piece of software that you are using. There are many implementations of Maximum Likelihood, however, it should be easy for you to test if the one you are using is only using RGB by simply running it once on a 3 band stack and again with the other bands stacked in as well. – Mikkel Lydholm Rasmussen Apr 30 '16 at 12:38
  • I am using ENVI 5.3 to extract my Landcover classes, So my questions is explicitly confined to ascertain whether Maximum Liklihood Classification considers RGB bands only? – Rex Apr 30 '16 at 13:21
  • In the future, remember to add a ENVI 5.3 tag to your question. As for the question itself, I'm quite certain that the implementation of Maximum Likelihood Classification takes into account all available bands. – Mikkel Lydholm Rasmussen Apr 30 '16 at 14:50

Maximum Likelihood Classification, in any remote sensing software, will consider all of the bands passed to the tool and not be limited to the RGB spectral space.

However, one thing to keep in mind is that Maximum Likelihood does not do very well with data in different scales so, for the best results, you want to match the bit-depth of your data. For example, if your NDVI is in a floating point (-1 to 1) and your source spectral data is 8-bit you would want to rescale NDVI to 0-255. Maximum Likelihood is also not a very good choice in algorithms when including ancillary data (eg., slope).

  • @ Mikkel Lydholm Rasmussen, in view of this ambiguity I should classify an extract of the scene using all of spectral bands with me and with RGB separately to check. – Rex May 2 '16 at 6:04
  • Not sure what the source of ambiguity is. If you are using ENVI there is absolutely no doubt that all of the bands passed to the maximum likelihood classifier will be used. I am really wondering where you got the impression that this was not the case. – Jeffrey Evans May 2 '16 at 14:22
  • Jeffrey Evans, the source of ambiguity is the result of this algorithm gis.stackexchange.com/questions/191158/… although there is vivid contrast between the spectral responses of these two materials (see spectral profile of the both) why does they were not segregated using the algorithm. – Rex May 2 '16 at 14:56
  • @Ben, but the post you are pointing to has nothing to do with a ML algorithm using only the RGB bands but, rather about spectral separability. Again, I have no idea where you got the impression that ML, implemented in remote sensing software, only uses the RGB bands. I have been working with remote sensing data for 25+ years and this has never been the case with any software I have worked with, which have been numerous. Your assertion that only the RGB is considered is just incorrect, otherwise a massive history of remote sensing would have just not been possible. – Jeffrey Evans May 2 '16 at 15:51
  • you are right, thank you for a brief outline for the understanding. It has helped me what I can't understand in the past. – Rex May 3 '16 at 4:21

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.