Train classifier


I have a train and test samples in tif format, how can convert this raster data to vector data to using it in a train classifier ?

Thank you.

You have two tif files containing label values for the different classes, one for training and one for testing ?

In this case, I don’t think we have an application in OTB that would convert them to vector data. However, you can use the gdal script (this script is in the binary package of OTB) that will convert each 4 connected component (areas with the same label) into a polygon. The resulting shapefile can then be used in the TrainImageClassifier.


I will try this thank you so much.

I don’t know if this is the right place to mention, but I think that adding the possibility to train with an image instead of a vector would be very useful for many applications. I had to do it for myself in a custom app because converting a large labelled image into vector is not always possible. I can share some code (multi-threaded sampler) if needed, but if I understood well there is already an internal conversion of the polygon to rasters at some stages of the sampling, so I guess that the existing framework would not need too much changes.