X
PrevPrev Go to previous topic
NextNext Go to next topic
Last Post 06 Aug 2015 01:08 PM by  anon
Using Band ratios in Decision trees
 1 Replies
Sort:
You are not authorized to post a reply.
Author Messages

anon



New Member


Posts:
New Member


--
06 Aug 2015 01:08 PM
    Hello. I am trying to separate submerged vegetation from sediment within an estuary using some CASI hyperspectral data. I first ran a Max-likelihood classification with defined ROIs. I am now using decision trees to further separate the classes and improve overall accuracy. So I am using the classification file as the initial input data and the imagery(band-ratios), elevation, NDVI data to refine the classification I have identified a band ration that should separate the classes but it's not. I have used the code: (b25/b15) LE 0.3 where b25 and b15 have been assigned to the original hyperspecral data. Any ideas why this is isn't working? I have attached some photos to my website: https://sites.google.com/site/chadstellern/image-classification

    Deleted User



    Basic Member


    Posts:228
    Basic Member


    --
    07 Aug 2015 01:00 PM
    It looks like the hyperspectral data is stored as signed integers, instead of floating point values. Sometimes hyperspectral data are scaled by some constant factor (say, 10,000) so that it can be stored as integers to save space. In that case, the expression you show would result in an integer as well. Integer division is performed. The remainder is discarded. It looks like when you discard the remainder, all of your pixels end up with a value of 0 or less, and so answer Yes to the decision "is the value less than 0.3". If this is the problem, then you can resolve it by changing your expression to be something like this: float(b25)/float(b15) lt 0.3 - Peg Exelis VIS
    You are not authorized to post a reply.