Mixing classes *drama*

[cvamko]

Hello everybody!

I am trying to make a relatively simple LULC classification of a region with only 5 categories - grassland, agricultural, built-up, forest and other. I have the data for summer seasons.

I tried with several RGB combinations, PCA, band ratios from every possible tutorial, still I can't differentiate barren land (in this case "agricultural" category") and built-up.

Is there any algorithm or some procedure which would select some of the possible good combinations? I am working with ArcGIS 9.3 and am very frustrated.

Also, when I tried with several procedures, I did it parallel on 1991 and 2011 datasets, LANDSAT same bands and same month of the year, unsupervised classification, I got completely different results, some classes were impossible to distinguish on one dataset, and were separated in the other.

Thank you in advance.

[SIGIS]

Hi you are using a tool which doesn't you have a landsat dataset why don't using a remote sensing tool such as ENVI ? If you know C or C++ language you can also use the Orfeo Toolbox which is a beautiful library mixing topology and remote sensing.