I have high resolution UAV-Images (4x4cm) in RGB and would like to detect individual tree canopy's. I was using Object Based Image Segmentation (OBIS) from SAGA to classify the images by homogeneous spectral regions and got that result:

enter image description here

I humanly can interpret the segmentation and be able to see individual tree canopy's. But I need a computed solution to isolate the individual canopy's. Using OS-GIS tool's.

  • I think the tool you are looking to is Object Based Image Segmentation (not only Classification). In SAGA 6.3, it can be found under Tool Chains - Imagery. – Kazuhito Jun 27 '18 at 11:59
  • I mistype. I was using OBIS instead of OBIC. The results are from the Object Based Image Segmentation. I changed that in my question. – MAP Jun 27 '18 at 15:30
  • Have you tried watershed segmentation? I cannot tell if it works for your case, though (it looks just green and black to me). Btw, you may as well fill small black dots before going to next step. – Kazuhito Jun 27 '18 at 15:48

Maybe you can try this package in R, it's ForestTools: https://cran.r-project.org/web/packages/ForestTools/ForestTools.pdf

Description Forest Tools provides functions for analyzing remotely sensed forest data. Functions like vwf and mcws are applied to rasterized canopy height models, and can detect treetops and outline their respective crowns. sp_summarise can summarize tree counts and attributes within particular areas of interest or within continuous spatial grids.


Maybe you can make a three-dimensional model of that image and use an algorithm that can select only the highest values to use as a reference for the treetops of each tree. I know that the "pix4D" software (not free) makes a three-dimensional model of very high resolution aerial mosaics. Or you could use the bands of green and red to locate the crowns illuminated by the sun, since in the green band the forest reflects a lot and in the red band its reflectance is low.

  • I already did that with Web-ODM in alternative to Pix4D. The individual tree detection by using the Canopy High Model (CHM=DSM-DEM) is working fine, but in addition I need the shape (area) of every single canopy and would like to extract it from my RGB-Image using OBIS. – MAP Jun 29 '18 at 5:16

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.