X
PrevPrev Go to previous topic
NextNext Go to next topic
Last Post 11 Apr 2013 05:59 PM by  anon
Gathering training data across images
 1 Replies
Sort:
You are not authorized to post a reply.
Author Messages

anon



New Member


Posts:
New Member


--
11 Apr 2013 05:59 PM
    I'm currently carrying out a project for a class. I need to carry out a multi-temporal classification of two landsat scenes from Central America (both the same R/P). After some pre-processing I am now ready to classify, and I would like to use the same training data for both images as they are both corrected for atmospheric affects. Is there a way I can select training data pixel by pixel from both images? For example, I have individually classified images in the past, but that makes change detection somewhat difficult, as I am unsure what changes are due to differences in classification parameters due to having different training data for each image's classification (I will be using an nearest means method). I have been told that the only way to do this would be to find regions in both images that are unchanged, and apply the ROI's from one image to the other, but would prefer if there is a way to build training data that is then applied to both images for a simultaneous classification. Another question, once ENVI has carried out change detection and produced contingency tables, is there a way I can instruct ENVI to turn the pixels detected as having changed, into change classes of their own? I have been told of one method that involves some arduous band math, but is there perhaps another and more streamlined method? Thank you for any assistance.

    MariM



    Veteran Member


    Posts:2396
    Veteran Member


    --
    12 Apr 2013 10:50 AM
    It seems that if the images have overlap, you could create ROIs of the overlap area that contain statistics of all bands in all images. You would need to layer stack the images and then create ROIs or Shapefiles of the training areas that are common in both images. Then we performing classification, use a spectral subset of each image in the classifier. Or, calculate covariance on the ROIs (using the ROI tool in classic) and export the ROIs to a statistics file. You can use the statistics to apply a classification to each image separately using the Endmember Collection dialog. This would be primarily performed in ENVI Classic or a previous version of ENVI.
    You are not authorized to post a reply.