X
PrevPrev Go to previous topic
NextNext Go to next topic
Last Post 05 Oct 2015 12:59 PM by  anon
NDVI Usage in Confusion Matrix
 6 Replies
Sort:
You are not authorized to post a reply.
Author Messages

anon



New Member


Posts:
New Member


--
05 Oct 2015 12:59 PM
    I am working on NDVI in ENVI... the output file is in Envi Standard format... I required it in Envi Classification format, which can be done through Envi Edit Header tool... This requirement is for its use in Confusion Matrix using Ground Truth Image... but when i performed this, the resulted NDVI image transformed into BLACK color showing as 'unclassified OR class1' portion (what ever name was suggested at the time of conversion into classification format). (As by the editing header, from standard format to classification, it require at-least 2 classes). I need this to perform Tasseled Cap comparison with NDVI. KINDLY SUGGEST SOMETHING EFFECTIVE.

    Deleted User



    New Member


    Posts:
    New Member


    --
    05 Oct 2015 02:27 PM
    When performing NDVI in ENVI, the output file is simply a grayscale image in which each pixel has a NDVI value computed for it (typically values in the output NDVI image range from -1.0 to 1.0 with vegetation having values from somewhere around 0.2 to 1.0). This is not a classification image, so simply changing information in the header would not work. To create a classification image from an NDVI image you would have to apply thresholds to the image in order to delineate separate classes. An easy way to create a classification image from your NDVI result would be to right-click on the file in the Layer Manager and select 'New Raster Color Slice...'. In the Edit Raster Color slices dialog, you can slice (A.K.A. threshold) the image any way you choose. So, for example: Slice Min: -1, Slice Max: 0.2 (no vegetation) Slice Min: 0.2, Slice Max: 0.4 (slight vegetation) Slice Min: 0.4, Slice Max: 0.6 (moderate vegetation) Slice Min: 0.6, Slice Max: 0.8 (high vegetation) Slice Min 0.8, Slice Max: 1.0 (very high vegetation) Once you are finished, you can select 'OK' to close the dialog. The new slices will appear in the Layer Manager. Right-click on the newly-created 'Slices' folder in the Layer Manager and select 'Export Color Slices... > Class Image...'. This will create a classification image for you from your NDVI file using the threshold values you applied in the Raster Color Slice tool. You could do something similar with the 'Green Band' from your Tasseled Cap Transformation if you wanted to compare them. One thing to keep in mind is that the values in an NDVI image that represent vegetation do not necessarily correspond to the values that represent vegetation in the greenness band of Tasseled Cap. You will have to choose how you handle this to determine which algorithm actually works best to delineate healthy vegetation. I hope this helps!

    Deleted User



    New Member


    Posts:
    New Member


    --
    07 Oct 2015 11:37 AM
    OMG! Thanku so much... i ll apply this and let u know... Can i do the same with T-cap image?

    Deleted User



    New Member


    Posts:
    New Member


    --
    07 Oct 2015 02:15 PM
    Yes, you can do something similar to this with a single band from your Tasseled Cap images (i.e. the Green band for vegetation). Like I mentioned previously, Tasseled Cap and NDVI values do not necessarily correspond to one another, so you would need to fiddle with the threshold values you apply to your NDVI and Tasseled Cap images to pick out similar features in each. Good luck!

    Deleted User



    New Member


    Posts:
    New Member


    --
    07 Oct 2015 09:34 PM
    Thank you. I have tried this method... my NDVI images ranges from -0.5 to 0.6 (its a more of urban area)... and im working on Landsat 7. What if the value ranges for NDVI and TCT dont correspond much? I have to use confusion matrix for its accuracy assessment.

    Deleted User



    New Member


    Posts:
    New Member


    --
    09 Oct 2015 01:48 PM
    Yes, unfortunately, these are different algorithms, and although NDVI and the greenness band of TCT both emphasize vegetation observed in Landsat imagery, they do not produce an apples-to-apples comparison because the computed values do not match. Unfortunately, that's going to be up to you to determine where you set those threshold values. There might be some literature on this topic, but if so, it is not something that I am aware of. Good luck!

    Deleted User



    New Member


    Posts:
    New Member


    --
    18 Oct 2015 04:54 AM
    Thank you so much for kind assistance... i ll manage further.
    You are not authorized to post a reply.