Classification of GRD product

thank you for your quick response
sir every image in the same coordinate system, because i done the batch processing for these all images.
sorry sir I don’t know hoe to check thisgrafik

thank you for your quick response,

yes sir my all inputs are S-1 GRD.
you mean my input vector also should be in the same coordinate system of S-1 TC products.

Yes, exactly

Thank you for your good response
I got the solution for that but, While doing the RF classification for the GRD products getting the error like bound must me positive, sorry I don’t know what it is


could you please help me for my classification.

and also sorry for my lack of knowledge, I have one doubt the classification which we are doing is on what based will be classify?
is it color based or backscatter based? please give me some guidance.

thank you in advance.

your data must be reprojected into WGS84. The solution has been given here: Rndom forest classification steps

radar data is not measurung colours, it operates at higher wavelenghts. The signal you retrieve (and the classification is based on) represents physical characteristics of the surfaces (roughness and moisture, among others). Rice, water and built-up structures have different types of backscatter mechanisms, so they should be separable quite well. Build-up structures cause corner reflection (very high backscatter), water is mostly smooth and rice produces volume scattering once crops are developed. Especially if you are using images of different dates, its signature will change and be therefore quite unique.

The solution is already solved in here,

Source of the post

and in here,

source of the post

If I understand your question well, Your input data are GRD, so the only information available is intensity.

thank you sir for your quick response
I will go through your steps

thank you so much.

By default the “number of training samples” is 5000 in any classifier in SNAP. How does it effect the results exactly and if I increase it’s value, will the accuracy improve ?

please have a look at my comments here: Number of training samples at Random forest classifier

This is related to the size of your study area, please also read here: Issue with Supervised random forest classification

hi team,
sorry for lack of knowledge,as per my concern while we are doing the classification by any technique, we need take vector samples, then only we can classify our products.
is there any technique to classify the products by giving the back-scatter ranges (like -14 db to -19 db & -24 db to -30 db ) only based on our requirement without giving any vector samples??
if any technique is there please help me to classify the products.
and one more question is after done the classification how can we see the statistics based on area wise.
please help me to do that process.

thank you in advance.

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.

Thanks.

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?