Random Forest Classifier struggles with long names

s1tbx Problem Reports
#21

maybe you accidentally removed one of the brackets while renaming the rasters
grafik

Fore every open bracket there must be one closed.

#22

no missing brackets I guess.

image
image.png1242×692 24.6 KB

#23

could be as well at any other location where you edited the names. Otherwise this error wouldn’t occur.

Is it an option to repeat the whole stacking. If you rename the bands before the stacking, the manual editing afterwards would not be neecssary.

#24

I ckecked but it is not.

I did it but gives same error.

image
image.png1924×1080 236 KB

I think the problem causes about the band I ve created by Band maths. After stack, I cannot open the band I ve created via Band Maths.

image
image.png1922×1080 212 KB

Maybe it is better to do all preprocessing of radar images from the beginning and try again.

#25

if you create a band in the band maths, make sure you uncheck ‘create virtual band’ because otherwise it will not be available after renaiming the origin rasters it was based on.
Uncheck ‘create virtual band’ and save your product afterwards. Only if it appears as a raster in your data folder, you can use it in later products.

This could indeed be the reason for errors.

1 Like
#27

I pre-processed all my S1 images and created new bands by using Graphbuilder (it doesn’t create virtual bands). Then I stacked one S1 product with one NDVI product and tried classification. Finally it worked. Most probably, the problem was due to virtual band as you stated. However I haven’t tried time series classification yet. I hope It will work as well

thanks.

#28

good to hear - good job.

#29

Hello,

For time series classification, first, I exported all my images to GeoTIFF format and created layer stacks in Erdas Imagine. Since Erdas gives unique names to each band. Then I opened these stacks in SNAP again and saved as DIMAP. However, when I try RF classification, It takes very long time. My screen is stable like this for one hour (I have 160 bands and 4 training sets).

Is this normal?

image
image.png1918×1036 158 KB

#30

this is the evaluation of feature importance. Uncheck “Evaluate Feature Power Set” to reduce the processing time. You won’t get a text file with the importance measures (permutation importance, gini decrease) however.