Hi all!
For my work I am interested in identifying a single kind of vegetation from LANDSAT Images. If I use signature file with just one signature class(I take many training sites all over the area), the classified image with i.maxlik appears full pink. On the other hand if I randomly pick training sites for one more class(water) and have two classes in the signature file this does not happen. So I am wondering will it not be possible to use a single class’s signature sites for supervised classification.
Thanks and regards,
Chethan S.
On 23/05/11 06:38, Chethan S. wrote:
Hi all!
For my work I am interested in identifying a single kind of vegetation
from LANDSAT Images. If I use signature file with just one signature
class(I take many training sites all over the area), the classified
image with i.maxlik appears full pink. On the other hand if I randomly
pick training sites for one more class(water) and have two classes in
the signature file this does not happen. So I am wondering will it not
be possible to use a single class's signature sites for supervised
classification.
Classification will attribute each pixel to the class with the signature closest to the one of the pixel. If you only have one class then this will, by definition, always be the closest.
I don't know if there is another way, but if you have a series of threshold values for your class in each landsat band, then you could just use r.mapcalc to identify those pixels that are within the defined thresholds in each band, i.e. something like this:
r.mapcalc myareas = "if((band1>X1 && band1<Y1) && (band2>X2 && band2<Y2) && etc), 1, null())
Moritz
Another easy option would be to try an unsupervised classification and
see if the vegetation cover you are interested in falls inside one of
the classes... who knows...
Cheers
Daniel
On Mon, May 23, 2011 at 10:09 AM, Moritz Lennert
<mlennert@club.worldonline.be> wrote:
On 23/05/11 06:38, Chethan S. wrote:
Hi all!
For my work I am interested in identifying a single kind of vegetation
from LANDSAT Images. If I use signature file with just one signature
class(I take many training sites all over the area), the classified
image with i.maxlik appears full pink. On the other hand if I randomly
pick training sites for one more class(water) and have two classes in
the signature file this does not happen. So I am wondering will it not
be possible to use a single class's signature sites for supervised
classification.
Classification will attribute each pixel to the class with the signature
closest to the one of the pixel. If you only have one class then this will,
by definition, always be the closest.
I don't know if there is another way, but if you have a series of threshold
values for your class in each landsat band, then you could just use
r.mapcalc to identify those pixels that are within the defined thresholds in
each band, i.e. something like this:
r.mapcalc myareas = "if((band1>X1 && band1<Y1) && (band2>X2 && band2<Y2) &&
etc), 1, null())
Moritz
_______________________________________________
grass-user mailing list
grass-user@lists.osgeo.org
http://lists.osgeo.org/mailman/listinfo/grass-user