Morgan Bond
New Member
Posts:2  
| 22 Aug 2019 02:13 PM |
|
|
I tested pan-sharpening on a LV3D (orthorectified) Worldview-3 image from DigitalGlobe. At this point the resulting pan sharpened image looks fine. Then I converted the multispectral image to radiance for FLAASH, then run the FLAASH tool. The resulting atmospherically corrected image also looks fine. But, when I then run NNDiffuse pan sharpening, the resulting pan sharpened image looks terrible. There are magenta halos and water loses it's gradient of color. I am not sure what is wrong. I have tried with, and without converting the panchromatic image to radiance, and it doesn't seem to matter.
|
|
|
|
MariM
Veteran Member
Posts:2396 
| 22 Aug 2019 02:48 PM |
|
|
To be clear - are you saying you are using NNDiffuse Pan sharpening on the original WV-3 data (ms+pan) in its original format and this result looks as expected?
You then perform a radiometric and atmospheric correction on the *original* Wv-3 ms data set and then perform an NNDiffuse pan sharpening on the atmospherically corrected data. This results in color distortions.
It is my opinion that if you are getting a good pan sharpened result with the data in their original integer format, then I would use that product for spatial analyses and visualization..
I would not perform a pan sharpening on atmospherically corrected data because it would alter the spectral response of the pixels. I would use this product for spectral analyses.
I have encountered color distortions using NNDIffuse pan sharpening with images that have a lot of shadow (ENVI-70441) but I believe the distortion occurred even with images in their original integer format.
|
|
|
|
Morgan Bond
New Member
Posts:2
| 22 Aug 2019 03:26 PM |
|
|
Yes, that is correct. I am trying to use various spectral indices to classify water and sandbars in stream habitats. Pan sharpening helps with identification of those fine-scale habitats, but I would like to create a classification model that can be used among images taken at different times/places. So, I was hoping to pan sharpen the FLAASH corrected data
Original pan sharpening of imagery as received from DG (if I correct to radiance, it also looks like this):
[img]https://i.imgur.com/4cfvwMK.jpg[/img]
Pan sharpening of FLAASH corrected data:
[img]https://i.imgur.com/EBNNV6X.jpg[/img]
|
|
|
|
MariM
Veteran Member
Posts:2396
| 23 Aug 2019 05:39 AM |
|
|
Yes, I can see that the blue band is emphasized in the FLAASH corrected data. Unfortunately, I believe this is a limitation in the algorithm since the pan band is so wide and ranges up to the NIR range.
I wonder if you scaled your FLAASH corrected data to integers in the original range (I believe 0-2046 or something like that) it would produce a better result.
In any case, for comparing time-series data corrected to reflectance, the integrity of the spectral response is very important. It is my opinion that it is better to not use pan sharpened data other than for location assistance.
|
|
|
|