[stevexu]

I do got this quest for quite a long time, as we use the typical solution of selecting PS candidates by choosing

1.Amp Stab. 1-sig/mu, which is a modified version of amplitude dispersion index. What about other choices?

2.RCS, is that for the ratio of clutter and signal?

3.Temp-phase, is that for the phase dispersion index?

4.Amp Stab mu/sigma, what is the difference between that and the 1-sigma/mu.

5.Amp Stab. plus Sp.., I can not see the following words, what that stands for?

6.Tem. Coh., I do not quite understand how to calculate the coherence in temporal domain.

7.Syn. Coh., what the synthetic here mean? A weight of the spatial and temporal?

I not intend to challenge about the algorithm you use, but I do discern there are some PSCs not so trust-able, some vegetation along the road was assigned, some pixel on the lake(I am quite curiosity), some pixel positioned in the crop area(even when I set the 1-sigma/mu as high as 0.8). Do you have any suggestion to filter these fake PS out? Or any suggestion to depress the effects.

Thanks for you patience.

[periz]

I am tempted to answer in the following way:
whatever the initial PSC you select, there is no better way to identify the good targets as with the temporal coherence. That is, the best way to state if a point is reliable or not is to look at the phase dispersion after analyzing the connections. With “analyzing the connections” I mean processing the time series in the APS module by choosing the most appropriate model.
In fact, in Sarproz, after analyzing the connections you use non-linear weights to discard bad points based on the temporal coherence. That is the most reliable index to keep good PSC.

The initial selection of PSC is just a matter of optimizing the processing time. But you should not be so afraid of having also some bad points inside. You will discard it later on.

Anyway, given this introduction, in Sarproz you still have ways to better tune up the initial selection of points.
You can do it by generating your own index to decide which points you like and those to discard.
Instead of loading one of the parameters you mention, you can scroll all the list down to “choose a real file”. You can thus pick up your own index.
How to create your own file containing your selection?
a. select an initial set of points via “Load Mask” in “Site Processing”. E.g. Refl Map>1 just to take a lot of points.
b. go to “Scatter Plots”
c. in Scatter Plots you can apply as many criteria as you want with the list in the right column. You can apply boolean operations (and/or), you can select any/all the parameters you mention above. You can select points with good coherence, points at low elevation, points in a given position in your image, or use amplitude statistics or whatever you want.
e. after tuning up your sub-selection, you can save it using the “save selection button” (1.4.9.2.1. section of the manual). Be careful there: if you simply save a sub-selection of points, you will get a matrix with 1s and 0s. If, before saving the sub-selection, you select a parameter as “color” such parameters will be saved in the matrix in the non-zeros elements. This is useful for further points sorting (the APS module makes use of it).

About your further questions (full parameter names are reported in section “1.4.4.1. Load Mask” of the manual):
– RCS: Radar Cross Section (to make it meaningful you need to calibrate the data. otherwise it’s just indicative)
– Temp. Phase: constant of proportionality phase vs temperature (thermal expansion coefficient)
– mu/sigma hope you can understand the difference w.r.t 1-sigma/mu. the second one is normalized
– Amplitude stability index plus spatial coherence. it’s a way to consider both amplitude and phase stability at once
– temporal coherence. better you read some basic paper about the persistent-scatterers approach. there you will find it.
– synthetic coherence: sorry, we will remove it. it’s obsolete

[Author]

Posts