| 25 Jun 2013 07:50 AM |
|
|
Good afternoon,I am doing the Post-classification Change detection analysis in ENVI 4.7. I am using LANDSAT images from four different time periods. I did the classification on all four of them, and there are certain issues. For example: some agricultural fields that are not used anymore, or some concrete surfaces appear as the same class as the sandy beach (for example). Probably they have the same/similar reflectance. I tried to remove these spurious pixels, i.e assign them to their true classes (agriculture/urban) but had no success. I've tried it with Majority analysis. It worked to some extent but then also some of the 'real' pixels of sandy beach were removed :(My question is: Is there a way to remove these pixels 'manually', e.g. can I assign these pixels to their true class one by one (they are not so many but still they can affect the Change Detection procedure).Would be very grateful for your help.Kind regards
|
|
|
|
Deleted User
New Member
Posts:   |
Deleted User
New Member
Posts:
| 25 Jun 2013 01:30 PM |
|
|
Thanks Alex. I really appreciate it
|
|
|
|
Deleted User
New Member
Posts:
| 02 Jul 2013 11:05 AM |
|
|
You might write a majority filter to run over the class image. It could be guided by rules such as "if center pixel is class A, then change assignment to majority of background". Or, "if ctr pix is class B, then don't change". And so forth.
Good luck!
|
|
|
|
Deleted User
New Member
Posts:
| 03 Jul 2013 02:40 AM |
|
|
Cheers Jim!I did my first post-classification 'cleaning' with majority analysis. It worked to some extent. However, since I am working in coastal zone, there are always some abandoned agricultural fields and concrete surfaces which appear as a sandy beach (I guess they have very similar spectral signatures). I need to say that I've tried the recommended Classification Editor extension, and it worked PERFECTLY!!!Thanks to everyone
|
|
|
|