Hi,
I Have a problem with the Unsupervised H-Alpha DualPol classification. Wenn I am doing the classification I am getting 9 classes. Thats how it suposed to be, because the H-Aplpha Plane is devided in to 9 zones, BUT the 9 classes and there percentage of the classification don’t fit to the distribution in the H-Alpha Plane from the H-Alpha Dual Decomposition.
So, did I undersatnd you right?.. There should be a relationship between zones and classes? But the progremm is doing a mistake and is asigning the zones falsely to the classes of the classification?
I am also intrested in this topic, because you mentioned it is for dualpol data. (Kefeng et Yonghui, 2015) noticed that for dualpol data division of H/alpha plane is different as for quadpol.
Yes, that’s right. The Non-Feasable Regions (or irrelevant region) in the Plane of DualPOL data and FullPOL data are different.
From : Kefeng Ji and Yonghui Wu
I think the Alpha-Image is inverse. Because for example waterbodies should have a quit high Alpha. But in my images the Waterbodies have the lowest Alpha.
Is this right? And if yes, is there anything I can do to invert the Scale of Alpha in the Plot?
And the Entropy in Urban Areas in my Image is quit low.
But in a lot of books the Entropy of citis is high.
Is the distribution of DualPOL data and the H-Alpha Plane with the QuadPOL Data comparable at all?
I am really confused.
Im worried about the article of Kefeng et Yonghui, 2015. It seems that, of all dual pol combinations, only HH-VV works for H alpha decomposition Does someone know or understand this thing?
I am not sure, because Kefeng and Yonghui talking about the Cloud-Pottier Decomposition. And on SNAP we are doing the H-Alpha Decomposition. I think they are different. And when you look into the Unsupervised Classification you can choose between H-Aplha Classification and Cloud-Pottier, both are fore DualPOL. I tried the H-Alpha and it works (still the problem with the classes from the top of the dicussion) for VV HV and also tried the Cloude-Pottier Classification, that dosent work for VV VH.
For anyone who is still interessted in this topic:
It is possible to use the H/Alpha Decomposition with DualPol Cross-Polarimetry Data (VV VH / HH HV), but the boundaries of the 9 zones in the H/Alpha Diagramm are not the same as for the Quadpol-Data. So you have to determine your own boundaries for the Dualpol H/Alpha Plane.
I hope that the SNAP-Team will develope something for a better Classification.
Hi,
maybe you can find some interesting stuff in here:
From : A. Tishampati Dhar*, B. Doug Gray
Title:
COMPARISON OF DUAL AND FULL POLARIMETRIC ENTROPY/ALPHA
DECOMPOSITIONS WITH TERR AS AR-X, SUITABILITY FOR USE IN CLASSIFICATION
or
From: Zheng-Shu Zhou
Titlel:
DUAL POLARISED ENTROPY/ALPHA DECOMPOSITION AND COHERENCE
OPTIMISATION FOR IMPROVED FOREST HEIGHT MAPPING
@tim.patelscheck, I’m interested in determining my own boundaries for dualpol H/Alpha Plane. Do you have any suggestions on how to do this practically?
if you want to adress this empirically, you could use an image where you have sound knowledge about (field information would be best) and manually select regions with Double bounce scattering (urban), multiple scattering (e.g. forest canopy), surface scattering, dipole scattering, multiple scattering and bragg scattering and then systematically compare their entropy/alpha values. This could lead to mathematically derived thresholds, which seperate them best.
Does someone know or understand this thing?