X
PrevPrev Go to previous topic
NextNext Go to next topic
Last Post 04 Feb 2012 08:19 AM by  anon
Problem with Mapping Methods: Endmember Collection
 2 Replies
Sort:
You are not authorized to post a reply.
Author Messages

anon



New Member


Posts:
New Member


--
04 Feb 2012 08:19 AM
    I have been attempting to apply the spectra collected via ROIs in one image to another image with MLC classification. The procedure that was recommended on this forum was to calculate stats with covariance from the ROI tool, save the results to an ENVI stats file, and then use the Spectral->Mapping Methods->Endmember Collection tool to import the saved spectra and apply them to a new file. This has worked fine for a Landsat TM image, however when I create a file which is the 6 TM bands stacked with a few vegetation indices and a topographic variable the procedure no longer works. Specifically, when I get to the final "apply" step in the endmember collection tool I get an error message that: "Problem: The selected algorithm requires that the collected endmember spectra all contain an associated covariance. ENVI is unable to continue because some of the endmembers collected do not have their covariance" I have repeated the process many times verifying that the "calculate covariance with stats" option is checked prior to saving the stats files. I have also exported the stats files to text instead and the covariance is listed there for all bands. Additionally, the full covariance matrix is displayed in the stats tool prior to saving as ENVI stats files. So, I'm positive that the covariance information is there, but it is somehow not being recognized by the endmember collection tool. Even more confusing is the fact that the described procedure works fine for a 6 band image. Any ideas?

    Deleted User



    New Member


    Posts:
    New Member


    --
    04 Feb 2012 12:23 PM
    Alright, so after some testing I figured out the problem. If you don't select the algorithm FIRST in Endmember Collection, apparently only the stats necessary to compute with the selected algorithm are imported. So, if you import the spectra with the default selection (parallelpiped) then switch to MLC or something which requires the covariance information, then you'll get the problem I was having. You must select the algorithm first, and then import the spectra, then there's no problem. It's an easy enough workaround, and maybe this would be obvious to someone else, but it wasn't to me and it created a rather perplexing error situation. It seems that it would be an easy fix to either import all the stats regardless of the selected algorithm, or to re-import the necessary info when the algorithm is switched. My argument here is that it's an interactive tool, and the order of operations here should be inconsequential. I don't know if it qualifies as a "bug" but it's certainly an easily fixed nuisance.

    Deleted User



    New Member


    Posts:
    New Member


    --
    15 Jan 2013 01:50 AM
    Dear Joshua, thanks for the advice! For now, I saved 6 .stat files for the 6 categories of land cover classes. I have imported them in the Endmember Dialog and selected the MLC first. Pressing the apply buttom, everything seems normal and the screen about 'ML parameters' appears. If I leave everything on default, the output image will only have the value 2 so something is going wrong. Do you have some advice? Is it necessary to give in probability threshold values for multiple values or to change the scale factor (255)? Thanks!
    You are not authorized to post a reply.