The solution is already solved in here,
and in here,
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
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?
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,
Giang