Strange artifacts after Goldstein filtering with CSK

Microwave Toolbox Interferometry
1

Hi, I am trying to do InSAR with CSK data. After Goldstein filtering I see strange artifacts in shapes of stripes and squares, that do not seem to correspond to any land or manmade features:

artifacts
artifacts1224×874 1.81 MB

Am I doing mistakes here? My workflow is currently:

  • Input images are CSKS1_SCS_B_HI_05_HH_RD_SF_20180125170731_20180125170739.h5 and CSKS1_SCS_B_HI_05_HH_RD_SF_20180109170731_20180109170739.h5. They cover the area of Venice and its surroundings (the image above is around half of Venice’s main island itself).
  • I coregister them with Radar -> Coregistration -> Coregistration and default settings (except CreateStack’s Resampling type, which I set to BILINEAR_INTERPOLATION).
  • Coregistration seems fine: creating an RGB image with the two coregistered images on different channels seems to give a good match.
  • Then Radar -> Interferometric -> Products -> Interferogram formation, again with default settings, except enabling topographic phase subtraction (although it doesn’t change much, there is hardly any topography here).
  • In the end, Radar -> Interferometric -> Filtering -> Goldstein phase filtering.

I have tried both released SNAP 7 and latest development master version, with essentially the same result. Am I doing anything wrong?

2

this happens when there is much dihedral scattering. Lower the Adaptive Filter Exponent to 0.3 or lower as suggested in this tutorial DEM generation with Sentinel-1 - Workflow and challenges (page 26)

grafik
grafik1446×365 222 KB

Same was discussed here: Goldstein filtering introducing artifacts around bright targets

1 Like
3

Hi, thanks, that worked, although to make some of the strongest artifact disappear I had to lower the exponent to 0.01, at which point the filter does barely any filtering at all. Is this normal? Are there other tricks to avoid artifacts while still getting some useful filtering?

4

yes, this is always a trade-off. These patterns are quite deterministic, so I don’t think there is a way to generally suppress them.
You can try to apply spectral range and azimuth filtering, but I am not sure about their effect.