ROC

Hi all. To start off I am fairly new to all of this so this question may not be very clear. Any ideas you can give me will be great. I am trying to classify habitat on my rural study site. I have 2014 1 m resolution aerial photography available to me. I would like to separate habitat categories into woody (trees, shrubs, etc.), a heavy and light herbaceous vegetation category (very thick grass vs. sparse grass/forbs), and bare ground (mostly bare with some vegetation growth). I've been advised to do unsupervised classification in ERDAS IMAGINE. I've tried that, a combination of NDVI and texture, and now I'm working with a NDVI only classification. The woody category separates out well in all the different methods. Up until now the two herbaceous categories are not well defined and they match my ground truthing 50% of the time. Under advisement, at the moment I am using the NDVI classification to try to separate out a single thick grass species that is characterized by a center of green growth surrounded by a mass of dead growth. This species tends to grow over large areas and often covers 85-99% of the area, so I am not trying to pick out small clumps from this photography. I am trying to remove this as it may confound the NDVI classification, so the goal is to identify areas that this species exists and clip them out to run a NDVI on all other herbaceous vegetation. This classification to identify this species does not seem to be working well. If anyone is able to provide guidance on methods I could try to differentiate these vegetation classes or pick out this certain species from the imagery I would greatly appreciate it. Thank you!