Change name of "Classification from vector" process
What "Classification from vector" does is not a classification, this denomination is very misleading. This process is a vector to raster conversion in which pixels take the value of the integer IDs recorded in the vector table of attributes.
Also, in the Classification workflow the entry “Vector Classification” should be “Vector training set” or “Labels in vector format”
Comments (10)
-
-
Also, in the Classification workflow the entry “Vector Classification” should be “Vector training set” or “Labels in vector format”
The term “Vector Classification” is now explained in the help text:
-
reporter But this is an inconsistency, that’s not a classification. Users understand a classification layer as the result of a classification process. Also, note that all classification processes in EndMap box produce a raster layer, so rasterizing a classification does not make sense. Actually, most people would look for the reverse, vectorizing a raster classification which has been produced by EnmapBox so that it can be overlaid on the original input multi- or hyper-spectral raster.
-
Ok, yes, “Rasterize Classification” would be inconsistent, but “Rasterize Vector Classification“ would be fine, right?
Users understand a classification layer as the result of a classification process
That is not always the case: we also want a vector layer to be a valid classification, e.g.
or
Such a vector classification can be used together with a raster (holding the features) to train a classifier.
-
reporter I think that the point is that I disagree with your use of the term “classification”. In a raster-based processing toolbox such as EnMap Box, users will assume that a “classification” is a raster layer resulting from applying a classification process to an input multi-band raster. Users will often want to vectorize their raster classification and thus get a classification in vector format. But it seems that you are using the term “classification” also for training sets (aka labeled data sets), and there I disagree because it becomes very confusing. It is true that you can use part of a classification as a training set, but this is not the general case. Inputs for a supervised classification are the image (raster, by definition) and the training set, which can be in vector or raster format.
-
We’ve recently introduced a gloassary, where we try to explain all relevant terms. Regarding the term classification: we now only use it for raster layer, as you also suggested. Feel free to browse trough it: https://enmap-box.readthedocs.io/en/latest/general/glossary.html
-
"Classification from vector” is now called “Rasterize categorized vector layer“
-
- changed status to resolved
-
reporter In the machine learning literature, the term “labelled dataset” is often used to refer to a dataset that can be used for training, validation and test. Perhaps “Rasterize labelled vector layer”? But “Rasterize categorized vector layer“ is also ok.
-
Yes, we also use labelled as an overarching term:
- Log in to comment
Agreed. I was already reworking the whole algorithm. Will rename that into “Rasterize Classification”: