I am attempting to run a decomposition on RADARSAT-2 quad pol data to obtain the H / A / alpha plane information. I am wondering if I have the correct model:
My first question is, should I be speckle-filtering the image prior to or post polarimetric decomposition. My second question is, is this the correct model/model order to be using? My third and final question is, if the model is correct, and since I am looking to extract the H / A / alpha information for a subset of vectorized regions, how can I go about plotting the information on the H / alpha (H/ entropy) plane (2-d plot)?
Perfect, thank you Iveci! I have one theoretical question for you: why perform the terrain correction post-polarimetric decomposition? Would it make a difference if it comes prior-to or post-decomposition?
Iād say, mostly because Terrain Correciton involvs resampling of the raster to the new pixel size. This changes your values (depending on the projection). It is advisable to leave the complex values (i+q) as they are until the last step.
Perfect, that is what I thought. Thanks! In the model progression outlined by Iveci, we are still performing the pre-processing step of speckle filtering prior to decomposition. Is this advised as well, or should we be running this somewhere after the decomposition too?
both steps involve a moving window. But if you remove speckle before applying the decompositions, changes are higher to get larger areas of similar scattering meachanisms.
If your desired targets are very small and you strongly rely on the detection of their polarimetric features, you might as well skip the filtering and first conduct the decomposition with a small image size. But you get nicer results with less noise if you filter first.
Thank you ABraun, this is very helpful. I am looking at rather small areas, maybe about 100 pixels within each ROI. I am comparing the decomposition to the error results within the processed data. Do you believe speckle filtering is the proper practice in this instance?