Classification of GRD product

Did you try up train on raster and then identify your thresholds in Quantize class value ,

Did you check up this post

Source of the post

Before calibration, should I do “Thermal Noise Removal” and “Apply Orbit-File” as graph below?

I don’t think it makes a large difference. But to be honest, I don’t know.


By the way, I will create ( sigma0VH / sigma0VV ) ratio band. Should I do it before GLCM or after?

After I calculate GLCM, I will do classification with using VH, VV and VH/VV ratio bands in dB scale. So I am a little confused. Should I create VH/VV band before GLCM to have entropy, energy, contrast etc values for VH/VV band and check if it improves my classification results.

first calibrate to Sigma, then calculate the texture measures.

sir as you mentioned above I tried on train on raster, but got the blank outputs

could you please help me to do classification by giving the backscatter values as training samples.
thank you in advance

What is the minimum class value did you apply?

The thing you should take in your consideration, this value whatever you selected doesn’t meet the minimum of all your input raster, that’s why I think some of your input will be out, I think the better solution is to switch to vector training,

what training bands did you use?

I wanna to use RF to classfy the GRD of sentinel-1 ,the images can processed in the snap software whole now?please reply me as soon as possible.thanks in advance!

It is still need to train the classfy use the pyimpute?

SNAP supports random forest classification in the meanwhile.

Please check these discussions:

thank you very much!

just as your advices ,I had do it on snap with RF to classfy the GRD products,my steps like :GRD-calibrated- GLCM-filtering-Muitilook-Terrain correction-RF(not train the samples),is it right?I wnna to classfy the product to 3 types.if not what should i do?
please reply me ,thanks in advance!

To do a supervised classification you need vector training areas.
You cannot use the textures as training bands because they do not store classified information.

If you have no training vectors you can only do an unsupervised classification.

Dear @ABraun,
Can you please share with me your script processing image with scikit-learn module to learn processing image in Python?

Thank you very much!
Kind regards,

It is this one:

1 Like

Thank you @ABraun!

Hi. I have a question. What is recommended: calculate GLCM with bands in dB scale or with original scale? I did a test and results were very different visually. Could you help me @ABraun? Thank you so much

both is possible, I’d select those with better contrasts. You can visually check where the classes you aim at are better separable.

1 Like

Thank you so much for you reply @ABraun