I'm working on using CNN in eCognition to identify trees in RGB aerial imagery. This part is working fine and the resulting heatmap captures the trees quite nicely. Next I'd like to use the Manual classification algorithm to manually isolate mangroves from those trees. I've tried using Manual Classification algorithm with the brush enabled and tried to use a condition where it will only classify opbects if the heatmap is greater than 0.75. When I put the condition in the algorithm and execute it I get an error:

"domain is empty (no objects availble) Process: if Mean Mangrove > 0.75 : manual classification (brush: 15) unclassified -> Mangrove"

The brush then will paint anything unclassified to Mangrove and not be limited to just the values in the heatmap. 

Is there a way to limit the application of the manual classification brush to pixels values such as those in the heat map?




  • Keith Peterson

    Hi Greg, glad to hear your are using the deep learning capabilities of eCognition and having success with the heatmaps! First, let me answer the question on the error message, this simply means that eCognition cannot find any image objects that meet the description you have set up in the domain. If I interpret your CNL code correctly, you are asking the rule set to look at a feature "mean mangrove" and if the values in that feature are >0.75 then those objects should be classified as "mangrove".
    Where I am confused is with the use of the manual classification algorithm. It sounds like setting up this condition within an assign class algorithm would be better suited and increase your automation. Or do you really want to have the manual editing/classification step in your workflow?

  • Gregory West

    Hi Keith thanks for getting back to me. I have an Assign class algorithm which will assign classes based on pixel values derived from the heat maps and that works fine, but only for broadly classifying trees. Unfortunately because the imagery is only RGB, seperating mangroves from the other trees is proving dificult to automate so I was looking at ways to manually extract these. I can use the manual classification algorithm and limit it to a tree class, but was looking at seeing if it could be limited by pixel values as well.

  • Christian Weise

    Hello Gregory, instead of using the "Mean Mangrove > 0.75", please use "Max. pixel value Layername > 0.75". This pixel based object feature could solves your issue to restrict the algorithm to objects with pixel vales higher than 0.75.

  • Gregory West

    Hi Christian. I've changed the condition as descibed and I get the same dialog pop up:

    "domain is empty (no objects availble) Process: if Max. pixel value Mangrove > 0.75 : manual classification (brush: 15) unclassified -> Mangrove"

  • Christian Weise

    Hello Gregory,

    You are right, my mistake, I thought the algorithm supports the Image Object Domain. You have only one choice: Please reclassify tree objects to potential mangroves objects if heatmap is greater than 0.75 with as assign class algorithm and restrict the manual classification on the potential mangroves objects.

    If you want, you can create a wish in the wish-list (https://support.ecognition.com/hc/en-us/community/topics/360000772240-Software-Wish-List) to improve the manual classification algorithm with a "use class description" parameter, so that the manual classification has to consider the class description for each selected object.


Recent Discussions

Need more help?

Enter Knowledge Base Contact Support