I am doing sea ice classification using sentinel data with HH and HV pol image(over arctic region). Which parameter should be chosen for this classification since i can not apply decompositions 
Hi,
Dual pol data offer better seperability compared to single polarization. So, when performing classification, both of the polarizations should be chosen as input.
May I ask, which algorithm are you going to use for sea ice classification?
1 Like
any supervise classification algorithm
what are the features that can be extracted with this dual pol data.
can you tell me where to get Quad pol data over the ocean? thanks
So, you have not decided yet what kind of algorithm you are about to use.
If you have knowledge about programming language (such as python), i would suggest you to use either Support Vector Machine (http://scikit-learn.org/stable/modules/svm.html) from Scikit learn library or Convolution Neural Network (https://keras.io/layers/convolutional/) from keras library. There are many machine learning algorithms but these two (i have mentioned) are considered very good compared to other algorithms.
If you are not familiar with programming language, you can use Random Forest from SNAP. SVM and CNN outperform RF (do not expect to see huge difference though)
To achieve higher accuracy in your classification results, i would suggest you to combine texture analysis (GLCM) with the classification algorithm.
Why do you need Qaud pol data? your research is based on dual po data, right?
I do not know how easy it is find qual pol data over Arctic region
3 Likes
Can you tell me how to step in python from snap i know basics of SVM and CNN in python using tensorflow. how to open this file in python and any idea about ground truth or how to generate the same pixel by pixel
thanks
1 Like
It seems you have not decided yet what methodology to follow for doing a sea ice classification. It is good to have a plan beforehand (what kind of algorithms you are going to use and what you are trying to achieve) and then you start digging further into the algorithm implementation.
tensorflow is a nice library for applying NN. If you want to use SENTINEL-1 images on python, you have to export the image into tiff format first using SNAP toolbox. You can use either rasterio or Gdal python libraries to open up your image.
If you are not familiar with programming language and you need to perform a classification so that you can deliver your project on time, then you better use Random Forest from SNAP.
In terms of ground truth data, you can have a look at Danish Meteorological Institute (DMI) and Norwegian Ice Service .
3 Likes
Thanks a lot man. I will look at it.