Help doing supervised classification using texture data

Prev Go to previous topic

Go to next topic

Last Post 15 May 2015 10:55 AM by anon

2 Replies

Sort:


You are not authorized to post a reply.

Author

Messages

anon

New Member

Posts:

15 May 2015 10:55 AM
I have a panchromatic image, of which I created a texture map. I want to use this information to do a supervised classification. I was going to create three ROI classes: bare ground, buildings, vegetation. I've performed supervised classification in ENVI using spectral data but never pan data. More specifically I'm having trouble with my classification recognizing my ROIs in the entire scene. For example, I have if I use a texture map to identify buildings vs. trees. vs. ground in the ROI tool, when I do classification I miss most of the trees. Can someone offer some insight?

MariM

Veteran Member

Posts:2396
Veteran Member

15 May 2015 02:35 PM
I would look at the statistics of your classification ROIs. Do the statistics for the trees overlap one of the other classes? Try plotting the mean spectrum of the training ROIs. Is there overlap here? It may be that a grayscale texture image does not provide enough information to classify these features using the standard classifiers. Do you have access to the Feature Extraction module? Feature extraction uses a segmentation process along with spectral information for classifying. I was thinking it might do a better job with your dataset and classes.

Deleted User

New Member

Posts:

16 May 2015 09:13 AM
I do have access to the feature extraction module, but since I'm not using spectral data I wasn't sure if it would work. I'll try using it on the texture analysis to see what it can pull out. Thanks! Part of the problem is also the spatial resolution with what I'm trying to extract, which would be a problem regardless.

You are not authorized to post a reply.