Mapping and GIS Solutions Community

Hello,

I am looking for resources to help me learn how to perform an object based image analysis (segmentation) of Sentinel-1 SAR imagery in eCognition. All the learning resources I have come across to date focus on optical imagery. I am grateful for any help you can provide.

Many thanks,

Sarah

Hello Sarah,

You are right, radar data are not often used in this context but it works: If you convert you 16/32 bit data into 8 bit data, just for the segmentation, you will see that you can use the radar data in the same way like optical data. The trick is to generate temporary layers as output of the layer histogram modification so that you still have the "original" sentinel-1 data in the eCognition project (for further analysis). You can convert the input data into 8 bit using "layer arithmetics". Please take a look into this shared forlder with example projects for you:

4Sarah/All 

- Layer Arithmetrics sample Rule Set.zip --> how to perform pixel operations in eCognition

- Modify Layer Histrogram.zip --> easy way to modify layer histograms 

- Customized Segmentation.zip --> interesting approach to perform image segmentation in eCognition

In some weeks you will find tutorial, projects examples, tips&tricks, and a Community for eCognition releated questions here:

eCognition | Knowledge Base 

Cheers

Christian

Hello Christian,

Thank you very much for your reply, I am most grateful.

I wasn't able to open the files in the shared folders, but I did successfully convert my layers to 8 bit using the layer arithmetics algorithm. However, it's not clear to me how I use the 8-bit layers to perform the segmentation as I can distinguish even less with the 8-bit layers than I could with the 32-bit layers.

32 bit:

8 bit:

I have experimented with different layer combinations in the view settings, but I see a solid block of color no matter which combination of 8-bit layers I try. Where am I going wrong?

It's worth mentioning that I'm not intending to segment the whole scene. I have a mask containing one type of land cover (a tree crop) and I'm looking to distinguish different age classes and the presence/absence of flooding within that single tree crop. I'm not seeking advice on how to use a thematic layer to limit the segmentation, I know how to do this. I mention it so you are aware that the differences among my classes are much more subtle than trying to distinguish among buildings, roads, rivers, crops etc.

I appreciate any advice you might have.

Thank you very much.

Sarah

Hello Sarah,

Please take a look into eCognition | Knowledge Base to download the example projects. Here you will find also the new eCognition Community to ask eCognition related questions.

Concerning the 32/8 bit challenge: I believe there went something wrong. But first of all please use the 32 bit layers as input for the segmentation. If everything works good, fine. :-) If not, let us discuss in the eCognition Community what you could do to get "better" objects.

Regads,

Christian