Export preprocessed data to unit8

snap
#1

Dear All,

Sorry if there is a similar topic somewhere else, I did search but I didn’t find myself.

Preprocess data S1, for an instance is the first way towards moving to machine learning or/and deep learning algorithm, Exporting data as Geotiff/BigTiff Or/and Geotiff is resulted into to float32 which ultimately and according to my experience I didn’t find any effects of train the model in ML, However,
it really reduces the capability of the model sometimes to train optimally in DL in Unet model, simultaneously I found unit8 type is the best option to train the model.
Try out converting data type using any of the python packages rasterio, cv2, numpy, might be resulted in a loss of precision of the data.

Any solution in SNAP to save the results as unit8 directly?, Or any other suggestion!

@marpet @ABraun

#2

Any conversion from float32 to uint8 will result in loss of precision.
And the original Sentinel-1 data is provided as int16. So, I guess this would be the best data type.
In SNAP you have the option to convert the data type. Right-click and select ‘Convert Band’. But this works only on computed bands (virtual or filtered).
You can try to convert to int16 in python. This might give better results.