APS processing and coherence

This topic contains 1 reply, has 2 voices, and was last updated by  periz 11 months ago.

  • Author
  • #2631



    I have read a few topics on the forum on the APS module and I think I am getting there. However, I am still not sure which is the optimum for my analysis and if there is something wrong. I have been playing with different graphs, amp stab index thresholds, weights and so on. Often, I got into cases like this with the suggested NLW2.

    Basically, using the automatic NLW, The coherence of the connections get much lower (peaking on zero). On the other hand, if I play the NLW (i.e. setting m=0) the coherence of NLW gets better (gaussian towards one).

    Results get bad again at the APS estimate (coherence very low).

    I would like to know which is the objective (or the correct way to approach this): trying to keep the automatic values? Not decreasing them? TRying to shift as much as possible the Gaussian of the coherence towards higher values?

    dataset: 26 S1B, Canada, mine site, I would say from 500 to 1200m. There is a small town, but most of the AOI (80%) would be bare soil and vegetation (I am not interested in the forested area).

    I would have other questions on other aspects, but it might be better addressing them separately.

    Thanks for the great work

    You must be logged in to view attached files.
  • #2636


    By looking at your images I have to make the following comments:
    1. I am not sure if 0.7 ampl stab is a good threshold with 26 images. it would be useful to see the reflectivity map and the ampl stab map to have a better idea.
    2. the range of height you are trying to estimate is too big. consider that you are processing connections (nearby points)
    3. why aren’t you removing the external DEM (mistake: always do it)?
    4. why are you using a MST graph (generally, not good. too little lever arm for estimating anything)?
    5. why do you use the MST graph and you don’t use the coherence as weight? this is a kind of contradiction.
    The non linear weights are used to decide the meaningful values of the coherence. To make an example, 0.7 with 100 images is a high value of coherence. 0.7 with 20 images is very low. But if you use 0.7 as a weight, it is high in any case. So, non linear weights are helpful to stretch values from 0 to 1. However, if they are not properly chosen, even if the stretched coherence is close to one, it does not mean the result is good…

You must be logged in to reply to this topic.