rsgislib.rastergis.ShapeIndex

[kellyb...@gmail.com]

Hello,

I am using version 5.1.3 of RSGISLIB and I have a quick question regarding whether it is possible to calculate shape parameters within the RGISLIB toolset. The general documentation and the documentation for the Raster GIS module indicate that this is possible and that a data structure for rsgislib.rastergis.ShapeIndex exists, but the "calcShapeIndices" function doesn't seem to exist.

It looks like when this was previously brought up (the last instance I could find from a quick check was in 2017), the advice was to use version 2.1. Is that still the best way to implement these functions? (or I could just be being missing something obvious) 

Thanks in advance

Ben Kelly

[Peter Bunting]

Hi Ben, 

To be honest I’ve not looked at this function in a long time as I haven’t thought about using this type of analysis for my own work in a long time. It is probably unlikely I’m going to look at reinstating this function any time soon. However, I’d be happy to look at a pull request.

I think there are vector analysis tools which calculate some similar metrics in saga and Whitebox you could have a look at. 

Cheers, Pete

--
You received this message because you are subscribed to the Google Groups "RSGISLib Support" group.
To unsubscribe from this group and stop receiving emails from it, send an email to rsgislib-suppo...@googlegroups.com.
To view this discussion on the web, visit https://groups.google.com/d/msgid/rsgislib-support/24e94e84-8b9c-4b44-b04f-cc5b5e0d70d6n%40googlegroups.com.

[Nathan Thomas]

Hi Ben,

What sort of parameters were you looking for? I may have some code somewhere.

Nathan

To view this discussion on the web, visit https://groups.google.com/d/msgid/rsgislib-support/11E3780B-F283-4DE4-BB67-89D50681576A%40gmail.com.

[kellyb...@gmail.com]

Hi, thanks for the swift response both.

To give a little context, I'm currently working with high-resolution datasets, Pleiades and Pleiades Neo, and was brainstorming ways we could improve some of our neural nets. Specifically, for some classes where the target features tend to have largely regular/consistent shapes (vehicles, buildings, solar panels, etc.). So the indices that jumped out at me were things like asymmetry, compactness, rectangular fit, and roundness.

@Peter Burnting, do you mind if I ask if there's any particular reason why you stopped developing this feature for use in analysis? Or is it simply a case of your use-case (which I believe is largely global mapping of mangroves with freely available sources like Sentinel-2, Landsat, etc.) not really benefiting from these descriptors? I'll have a look at what SAGA and Whitebox have available. It looks like Ofreos Toolbox might also have some similar metrics. If I end up working with/updating the old RSGISLIB code I'd be happy to contribute that back to the libary. 

@Nathan Thomas, as mentioned above, I'm mostly looking at the basic descriptors of shape that could be used to help identify man-made objects, so again asymmetry, compactness, rectangular fit, and roundness. However, any code or direction you have would be greatly appreciated.

Ben