X
PrevPrev Go to previous topic
NextNext Go to next topic
Last Post 24 May 2013 12:52 PM by  anon
ENVI Error/Chain Classification Question*Edited*
 2 Replies
Sort:
You are not authorized to post a reply.
Author Messages

anon



New Member


Posts:
New Member


--
24 May 2013 12:52 PM
    I am working to create a landcover classification map of an area that crosses several Landsat tiles and that I have limited reference data across the whole area (although I have quite a bit in one of the tiles). I was able to classify the tile with the most reference data successfully and would like to utilize overlap between the tiles to 'chain' the classification. I want to use the successful classification that is in the overlap area as the training data for the next tile. I assumed that I could do that by creating a resized image 'slice' of the overlap area on the classification and creating ROIs using the "Generate Random Sample from Ground Truth Image" tool to create a pixel training ROI for the classification classes. I am able to get the pixel ROIs created and reconciled to the next image, but whenever I run the classifier itself all that returns is a blank screen with all the pixels showing a value of "0-Unclassified". I am using ENVI Classic. Any ideas what I might be doing wrong or any suggestions would be greatly appreciated. ***Edit*** I backtracked a bit to try to determine if the issue was with the images, the ROIs, or some combination. I am working with a stacked image of multiple dates and bands. If I break it out into the various parts (one image thats all spectral bands (36 bands total), and several vegetation indices (6 bands and 18 bands) images, for the multiple dates) I can get the ROIs to classify the individual stacks (stacks 1, 2, 3) and on combinations of stacks 1&2 and 1&3, but not on 2&3 or when I restack all of the different bands (stacks 1, 2, &3) into a single image it gives the black screen output from the classifier. Running a seperability report on the various image stacks also works fine, but when I try the seperability report on the totally stacked image I get an error "Singular value encountered in calculation for ROI:Random Sample" for each of the ROI classes. The totally stacked image has a total of 60 bands and I'm using at least 250 ROI points per class, in case that matters. I also tried creating totally new ROIs in the image itself and had the same problem (individual stacks work, but not the complete stack). I tried looking for a better description and ways to resolve this error on-line and was unable to find anything. I've used this same image/band stacking procedure on other images/tiles without a problem.

    MariM



    Veteran Member


    Posts:2396
    Veteran Member


    --
    28 May 2013 04:23 PM
    Your process is fairly complicated but it sounds like you have performed this kind of analysis before. The error you are running into with the ROI separability is essentially the same as described in this help article: http://www.exelisvis.com/Support/Help... There is something in your ROI data that is causing a singularity in the covariance matrix which stops the process. It could be duplicate bands or ROIs with bands that are too similar. You might try testing the classification process with just the original spectral data, then start adding in the products you created (you mentioned vegetation indices). I also think classifications on images with combinations of bands that represent the original reflectance data and products can cause strange behavior with some algorithms because the data ranges can be very different in those kinds of images. What classification algorithm are you using? Does it work with other such as SAM (typically it is the algorithms such as maximum likelihood that are sensitive to data ranges)?

    Deleted User



    New Member


    Posts:
    New Member


    --
    29 May 2013 06:46 AM
    The classification does work when using an SVM classification algorithm while ML throws the blank screen error. Ultimately I intend to use SVM for the classification anyway, so that part may be a 'worry about it later' issue, but it would be nice to have the seperability reports for reference, plus it'll be helpful to know where the error is coming from so I can avoid it/fix it in the future. I have done this type of stacked image analysis before and never had this problem, so thanks for the reference article. I'll try some of the suggested techniques and see what I can find out.
    You are not authorized to post a reply.