Ocean object detection with COSMO-SkyMed

Microwave Toolbox Problem Reports
#1

Hi,
I am performing an ocean object detection using COSMO-SkyMed stripmap data
Despite Sentinel-1 is processed, I cannot execute the operation over a COSMO-SkyMed stripmap image, even if image dimension is not big (less then 200 MB) and I have already converted .h5 image into .dim format.

I am using SNAP 5.0, will with a previous version (I thing SNAP 3) I could execute the processing.

I always obtain “java.lang.StackOverflowError” error, with both gpt processor using a graph, and throug SNAP GUI

Thanks for tour help

#2

which pre-processing did you apply on your data?

#3

I Abraun, thanks for your reply.

I first saved the .h5 image in BEAM-DIMAP format, then subset the image. At the end, I used the Ocean object detection tool as implemented in SNAP GUI.

But investigating the possible causes, I noticed that with smaller windows in the Adaptive Thresolding step the workflow ends with no error.

So I think that the problem is that using big windows (e.g. 800 m for background window size and so on as default options) causes the operator to build too big stacks, giving the error. With Sentinel-1 I did not obtain the error perhaps because at the same window dimension correspond a smaller number of pixels (due to the 20 m pixel size of Sentinel-1 GRD).

Could it be a possible solution to increase Java stack memory, adding a -Xss option in the VM performance options tab?
Do you have any other suggestions?

Thank you

#4

I personally would calibrate to Beta0 first, but this is probably not the error source.
The fact that larger window sizes produce a result indicates that it’s somehow the algorithm which caused the error.
Increasing the available memory is always a good choice to prevent errors but I’m not very firm with the Java variables. You’ll find some answers in here using the search.

1 Like
#5

When should Beta0 or Sigma0 be used?

#6

Beta0 doesn’t need information on the incidence angle. As you are working on flat water areas it should be sufficient.
Please anyone correct my if I’m wrong.

#7

What do you mean? is that necessary to make this conversion? how do you do it?

Many Thanks

#8

each file processed in SNAP is stored as a .dim-format by default. This handles the metadata best (instead of GeoTiff for example).
If there is nothing else to process, you can at least make a subset or calibrate your product.