AD adaptive filtering with PS

Viewing 2 reply threads
  • Author
    • #6356

      Hello everyone!

      What is the correct way to use Anderson-Darling Adaptive filter on the PSInSAR chain?
      Is the following process chain correct:
      1) Choosing AD statistical test in the Multi-Temporal Adaptive Mask.
      2) Setting MT adapt. on the InSAR parameters.
      3) Running Interferograms Processing. Sign convention in this step is opposite to SPP, there is an option to avoid this?
      4) Running APS estimation.
      5) Running Sparse Points Processing. Do I need to load manually the filtered Interferograms? if so, how it can be done?

      Is this the correct way to do it?

      On another question, does the file “AutoMISP-PRJ-Atmo.mat on the directory RESULTS/MATLAB/ contains the final atmospheric phase screens?

      Thanks for your replies in advance,

    • #6358


      1. the so-called PS approach is based on the following concept:
      A. star-images graph (single master).
      B. no spatial filter (full resolution images, no need to generate interferograms).
      C. no weights (the phase of each image weighs the same).
      In such context, in Sarproz you do not need to generate interferograms. follow the tutorials online to apply the full processing chain (APS estimation first, MISP later)

      2. if on the contrary you want to pursue a more complex approach (we called it Q-PS or partially coherent targets, others name it squeesar, weighted approach or whatever) you have to:
      A. choose a suitable image graph. ideally, you should use all possible pairs (full graph). a smaller set of interferograms could still be used, however, this depends on your dataset.
      B. apply spatial filtering (and an option here could be using the MT adaptive filter, after generating the MT adaptive mask with the proper statistical test). This means, here you need to generate interferograms with the options you want.
      C. use the coherence as weight (this option triggers the loading of interferograms). there is an option for this in the time series processing panel, in APS, MISP, TS module or Small Area MTINSAR modules.

      3. consider anyway that, even in the QPS approach, we suggest you to firstly estimate the APS using a conventional PS approach. You can later apply different processing steps in the MISP, after removing the previously estimated APS. This is anyway a suggestion. In Sarproz, in fact, you can do what you want, you are totally free to apply the processing chain you prefer with the options you prefer.

      4. yes, the AutoMISP-PRJ-Atmo is matrix Np * NI, containing the APS for the vector of points and for the images graph you selected

      5. the sign of interferograms is a convention that cannot be changed. however, if you look at a single interferogram, you could convert it into millimeters to have them consistent independently from the sign of the temporal baseline

      6. consider that soon we’ll deliver a new online sarproz course. write to courses at sarproz dot com if you are interested


    • #6392

      Hello periz,

      Thanks for the detailed response.

      When applying spatial filtering, in InSAR parameters, should be selected “Output Coherence = post-filtering”? So that this post-filtered Coherence was used on MISP module as Weights. I would appreciate it if you clarify this doubt.

      Regarding AutoMISP-PRJ-Atmo.mat file, the image shows APS for 5 dates and three models. The same APS estimation (Full graph and linear model) was used on the three models, but the models differ on the MISP processing.
      Model3 -> Full graph and liner model
      Model5 -> Star graph and linear model
      Model6 -> Star graph and Non-linear (Smart 5), not sure what 5 stand for?

      My question is: Why Model 3 APS are different from Model 5 and 6? Note that I only estimated APS for model 3 and load the same APS in model 5 and 6.


      You must be logged in to view attached files.
Viewing 2 reply threads
  • You must be logged in to reply to this topic.