Hi,
I would like to better understand the function of the Edit bands for segment signature extraction option. Is there any example of advantages in doing the preliminary SEGMENTATION task with a set of inputs and subsequent signature extraction with added inputs?
Luca,
A good question. I believe you are referring to the Segtrain module that creates the signatures from the training segments.
We have the option to edit the bands use fro creating signatures to give more flexibility to the user, after the segments were created. In this case, if you were to enter additional bands to the ones that were used to create the segments, this would give additional information during the classification process, in line with object-oriented processing.
You could use the additional information during the segmentation process as well, but you will need to experiment to see if the resulting segments are reasonable.
An example may be the use of soil information, which tends to be much coarser than typical multispectral bands. If you use soils during the segmentation stage, it will create boundary artifacts, rather than using soil data during the signature creation stage.
I hope this helps.
Best,
James Toledano
Clark Labs
Thank You, James,
it is helpful, I just try to reformulate Your answer, to check my level of comprehension of it.
You mean that when one has to face with the (spatial) segmentation, the choice of the kind of inputs should be driven by the spatial resolution (not in the sense of pixel size but of the characteristic dimensions of the objects that You are looking for). Instead, when the task is to characterize the segments in terms of signatures there could be an input feature, not good in terms of segmentation, but useful to better separate signatures. Is it right ?
Hello,
I have a problem now with segtrain. I have created a first group of signatures (segtrain_1) on the basis of the inputs used in the segmentation module. Now I would like to mantain this first result and make a new experiment (segtrain_2) on the same samples selected before but adding new inputs for extracting the signatures.
It seems that is not possible to mantain segtrain_1 and create a new segtrain (segtrain_2), but mantaining the same samples (I mean the vector file containing the polygons to identify the samples).
Is it right?
If Yes, I tried an alternate solution by using makesig with segtrain_1 as input in order to create segtrain_2 with the additional features to extract signatures. But this time the problem arise when I give names to signatures because they are the same names of segtrain_1, and there is a pop_up alert, asking if I want to overwrite the original signature files, and I don't want; because my objective is to create two indipendent results (segtrain_1 and segtrain_2) that have in common only the Segmentation file and the polygons selected to create the signatures.
was it clear my question ? Is there a solution?
Yes, on your first response. I think you understand it correctly. I hadn't thought of it in terms of the feature size, but it makes sense.
On your last response, again, you understand the process correctly. In Makesig, you just need to rename the signature names and the output signature group file. Then you can compare the outputs.
James