Sarproz news – Sentinel and Adaptive Mask
August 14, 2015 at 5:42 am #1080
during this summer we worked at two new important updates.
The first one is Sentinel TOPS. Yuxiao developed the code for reading it, stitching bursts together, coregistering and resampling.
A beta version is now included in the pcodes, we’ll soon add it also in the compiled code.
The module works now on single subswaths (one has to choose the subswath to work on in the SLC processing window).
We keep working for managing multiple subswaths together.
There are surely some issues, the first of which is related to topography. Moreover, we know that ESA is keeping modifying things.
So, we’ll have to work at improvements and modifications.
However, this beta version may already provide some useful tools. Please try it out and send us your feedback.
The second update is related to the multi-temporal adaptive filter. Francis worked on it. A new module is now available for generating probability tables of similarity between pixels and for creating clusters out of that. Clusters identify extended objects while unclustered pixels may reveal isolated targets. The mask can already be used for filtering interferograms. Some issues are still open, like how to best use it for multitemporal InSAR processing or how to best use it for selecting “objects”. Anyway, a beta version is available and, once more, one can try it out and report experiences/comments.
August 14, 2015 at 9:42 am #1084
thank you all for a good job and for putting your effort into developing S1 TOPS processing and adaptive mask. I’ve been eagerly following updates last days and I’ve tried to produce coregistrated stack of 8 S1 images. However, in versions from last days (7-Aug, 12-Aug, 14-aug) I get half of extracted Master and Slave images black. I was wondering that the only parameter that could possibly change it is Rg OVS, so I put it to value 2, which results in error (see attached zip file).
Regarding adaptive mask, as I understood the goal is to get well clustered pixels that will be treated as similar while producing interferograms. Then, the abbreviatons KS and DS stands for which statistical tests? Setting the significance level and connectivity (4 – cross pixels, 8 – all surrounding pixels; right?) should follow the goal – obtaining groups of pixels with similar behaviour, remove isolated targets and optionally remove the biggest cluster?
Thank you very much.
August 15, 2015 at 1:26 am #1091
thanks for the case, we are now including the management of INVALID samples.
In your particular case it may turn out that the adjacent track will better cover the AOI (it’s close to the border).
About the adaptive mask, KS is Kolmogorov Smirnov and DS is Anderson Darling (it should be AD). The connectivity is not used right now, we’ll remove it. Depending on how u play with the threshold, u can get features as the sea clustered together (this would be a big cluster). It will be useful for classification purposes.
August 15, 2015 at 2:10 am #1092
also thanks for the TOPS module.
Seems to me (not critical) that the UPDATE mode in the SLC DATA IMPORT module does not work… if I press Slave Extraction, it tells me to use Master Extraction first… and the Master Extraction does not work either.
(encountered more times)
August 15, 2015 at 3:37 am #1093
the S1 images do not overlap by 100% (although the bursts should be synchronized, they are put to different images each time), but according to ESA, it can be tiled seamlessly (in the azimuth direction). My area has such a problem: in some images, it is contained as a whole, in some images, only part of it is contained.
I also tried to put more images with the same date but different time together into the folder but I get “Subscripted assignment dimension mismatch.” Could you please also implement this feature? I suppose I am not the only user to take advantage of it.
Thanks a lot, Ivana
August 15, 2015 at 3:46 am #1094
Sorry for asking this, I want to get a clearer idea about your question: are you referring to the overlapping of different subswath? Can you give a few screenshots or something so I can have a better understanding towards your question?
August 15, 2015 at 4:15 am #1095
Thank you Yuxiao,
no, I mean something else:
having two images:
they are apparently one following other in the flight direction,
(during spring, everywhere was disclosed that they can be tiled seamlessly, but maybe they have it later overlapped by a burst or two as it seems they are overlapping by two seconds)
is it possible to tile these two (if present in the SLC directory) before cropping the image?
I am cropping (the default) area of 20×20 km but I get error message for two of the scenes. And also, when downloading, for some dates I get only one image, and for some dates, I get two… however the area is not large.
Thank you, Ivana
August 15, 2015 at 8:12 am #1096
I keep on processing, despite some images do not cover the whole area. I have encountered two errors I do not understand:
the first is in the lasterror.mat and .log files,
the other is the scshot documenting one of the resampled slaves (I would expect black at the end of the image, but not a stripe…) there are two images like this, master and other images look fine.
have no idea where it comes from.
Thank you, Ivana
August 15, 2015 at 3:18 pm #1100
I’ve taken a look at the log file, and there’s a very stupid error in the last version of sarproz that causes the tops-post-processing (a second-time fine coregistration) not able to finish.
And the wrong result in interferogram is exactly the consequence of the tops-post-processing (fine-coregistration) not able to finish.
We have fixed this bug and will publish it in the next sarproz update.
Can you let me know if the interferogram looks reasonably is 20150703-20150305? What is the other one?
Meanwhile, I am looking into the data where the stripes appeared.
August 15, 2015 at 10:40 am #1098
And at last,
I am attaching what I got as interferograms. Two of the interferograms look reasonably, two of them contain the stripes noted in the last post and the rest (7) of images look like this (attached): stripes in coherence.
(in the last post, I forgot to note that 20150703 is the master image.)
Do not know whether it can be caused by the fact that my area is close to the image border.
Thank you, Ivana
August 16, 2015 at 3:07 am #1107
yes, the reasonably-looking interferograms are 20150305-20150703 and 20150317-20150703. I am now processing more interferograms (full graph) to see whether more interferograms are reasonable.
August 16, 2015 at 6:03 am #1109
one more fact (do not know if it is caused by the stupid error you mentioned or with the problems with topography as mentioned by Daniele earlier):
having now more interferograms, there are more of them “reasonable” (i.e. not striped coherence), but many of them contain phase fringes in the azimuth direction and their frequency is not Bn-dependent (but it is different in different intfs).
In addition, for most of the unstriped intfs, the coherence looks like this, but for some of them, it is uniformly grey/black (maybe very dense stripes, maybe just worse coherence…).
Thank you, Ivana
August 16, 2015 at 11:52 pm #1112
1. Update to the newest sarproz version and all the problems with bad interferogram should be gone. You might need to start from the very beginning..
2. I failed to replicate your scene where the black strips happened in image intensity, especially in slave image 20150504. Can you tell me your specific parameters in ‘SLC data processing’ so that I can replicate your black strips? A screenshot of the module ‘SLC data processing’ where you got the black strips in image 20150504 would be very helpful for me to look into the matter. And meanwhile, may I ask that does the black strips happened only in coregistered image, or does it also appeared in the original extracted slave image? (where you can check in your EXT folder for jpg image)
Thank you for the testing!
- This reply was modified 2 years ago by Yuxiao.
August 16, 2015 at 10:29 am #1111
one more note:
I have now reprocessed the area once again with more images, and again: the first two (different days) look OK, the others are striped (with the same version of sarproz).
August 17, 2015 at 7:04 am #1114
I reprocessed the dataset with the new version. One black strip is again present in (coregistered) 20150504 and 20150516, as these are the two images with earlier timing, so the area of interest is not completely covered. The extracted slaves look ok, but they are thinner (less lines). This is probably the source of the problem.
I am attaching another problem, not having happened in the previous version.
Will try to compute the interferograms if it is possible.
August 17, 2015 at 7:44 am #1116
the interferograms (resp. their coherence) is striped again with the new version.
You must be logged in to reply to this topic.