Random Forest Classification (training on raster)

snap
#1

I’d like to conduct a supervised Image classification of Sentinel 2 Scenes (multitemporal). I’d like to use the Option ‘Train on raster’ in order to use raster data as Training data.

I’ve read about the EuroSAT Datasat which includes 27000 64x64 tiles of manually chosen Training data as raster (10 classes). https://arxiv.org/pdf/1709.00029.pdf

Is it possible to use this dataset within ESA SNAP in order to Train the Random Forest Classifier?

#2

The train on raster option is intended for land cover or bands that cover your product scene. This dataset is provided as labeled chips in JPG format without geocoding. You would be better off producing a lot of your own chips with your features and then using RapidMiner to train and classifiy.

1 Like
#3

Thanks for the helpful answer! But if I still wanted to use this dataset, is there a way in Snap to do so?