CANUPO fails to separate trees from the rest properly

[MaximusFuscus]

Hi, I am classifying a cloud that was acquired with aircraft in British Columbia, Canada. The area is very mountainous, it contains some steep inclines, a lake, powerlines, trees, fallen trunks + a residential area with homes, lodges, sheds, boats, fences, etc... The area is about 1.5km by 1.5km. I successfully classified the GROUND vs the rest and it worked beautifully. I selected quite a few various ground samples when training CANUPO. Then, I did a CHM with a surface raster. Then I was able to easily eliminate the lake water. After that I tried many many various forms of training in order to separate the trees (coniferous dense canopy) vs the rest. So let's say class 5 vs 6 (in this case 6 holds everything for now since CANUPO only allows 2 classes at a time). It's always removing most of the trees as I want but the bottom of some tree areas with bushes stay in class 6. Also, the powerlines mostly go to class 6 as I want it but also to class 5 (trees). I tried smaller minimum distances, large numbers of core points, whatever. Nothing seems to do the job properly. Is it possible that my samples are too large? I figured since the terrain is so varied I should have as many as possible. What do you suggest? I appreciate any answers very much as I am stuck with a project. Thank you so much.

[daniel]

Yes, there's definitely a limit to what Canupo can segment and differentiates.

Maybe it's time to step up and try 3Dmasc? I know it's more complicated, but also much more powerful.

Just my 2 cents as I'm not a user of these plugins myself ;)

[MaximusFuscus]

Thanks for the reply. I'm using CloudCompare 2.14, does 3DMASC work with any version? Any source of training materials for this? Thanks again.

[daniel]

Yes it should be included in any version above 2.13 I believe (at least on Windows).

See https://lidar.univ-rennes.fr/en/3dmasc