I am trying to perform a supervised classification with "i.maxlik".
I have three bands: 1) elevation (a dem); 2) reflectance (derived from Landsat ETM+ band4); 3) cumulative solar radiation (derived using "r.sun"). I am using rock glacier outlines (polygonal vectors) as my region of interest (ROI). I want to find which area is similar to my ROIs over the image in the three above bands. The image was previously masked to remove all regions other than bare soil (e.g. water bodies, vegetation, etc.), so I cannot define other classes for my classification easily.
My workflow was:
- I grouped the three bands with "i.group" into
- I added a new column to the rock glacier vector, giving a common value of 1 (int)
- I converted the vector to raster with
v.to.rast in=rg_visible out=rg_visible use=attr attribute_column=IDmaxlik
- I used "i.gensig" to generate the signature file (
i.gensig trainingmap=rg_visible group=perma_max subgroup=perma_max signaturefile=rg_sig)
- I run
i.maxlik group=perma_max subgroup=perma_max signaturefile=rg_sig output=classification01 reject=classification01_reject)
Each pixel of the output map ("classification01") is 1, as if the everything was classified as "rock glacier". Moreover, the reject map ("classification01_reject"), has higher values (closer to 16, or 100%) over the same rock glaciers used as ROI than elsewhere.
I already post this question here, and they suggest me to add classes to my classification, but this is not quite possible, as in my case I have no idea what is around the rock glaciers. Considering the pixels of my RGB image within the rock glaciers as a proxy of permafrost presence, the final aim would be to classify teh pixels of bare soil outside the rockglaciers with the supervised classification. Those which are similar to the rock glaciers would be my "high permafrost probability" pixels.