Change detection

Jonathan Mikolin

Change detection

Good afternoon eCognition users,

I'm hoping for some suggestions on figuring out how to do a change detection for two time periods in eCognition.
 
So far, I've classified one image from 2009 and another from 2018. My classes consist of hemlock trees and non-hemlock trees in each.
 
I would like to compare the two classifications and determine changes for each classified hemlock object appearing in both time periods, specifically:
  1. Size change (for a indication of growth)
  2. NDVI and GNDVI change (for a indication of health)

 

Currently, I did the classification in two separate projects and have two rulesets for each period.
 
I was considering exporting my results and attempting this in a third-party software. However, if I can complete this fully in eCognition I would like to. 
 
Thank you in advance!
Jon

Was this article helpful?

Date Votes

5 comments

  • Matthias Staengel
    • Edited

    Hi Jonathan,

    Yes! Stay in eCognition for the Change detection ;)! There are numerous approaches that you could use in eCognition to apply a change detection on your Classification results. As you already have the classification results for both time steps, my recommendation would be the following workflow:

    1. export your classification as a vector polygon from both projects
    2. create a new project with both image files and both vectors
    3. Create objects using the "vector based segmentation" including both vectors
    4. Copy the create image object level above
    5. Classify one Image object level with the information of vector 1 (using "assign class by thematic layer" algorithm)
    6. Classify the other image object level with the information of vector 2.
    7. Create yet another image object level "change Level" and classify it based on the two levels below (if Level 1 "forest" and Level 2 "urban" then classify as "change forest to urban" have a look at the approach here). Or something like that.

    You also can easily compute the NDVI changes easily if you have both raster layers loaded. Simply compute two NDVIs (for t1 and t2) and then create a customized feature to compute the difference, or use the layer arithmetics, to create a NDVI raster change layer.

    Have a look at these change detection tutorials here, they will give you additional ideas:

    1. https://support.ecognition.com/hc/en-us/articles/4406747424146-From-the-Ground-Up-Multi-temporal-NDVI-Change-Detection
    2. https://support.ecognition.com/hc/en-us/articles/4407828517778-From-the-Ground-Up-Change-Detection-using-Maps

    Hope this helps?

    Cheers,

    Matthias

  • Jonathan Mikolin

    Hi Matthias,

    Thank you for the guidance and links! I'll give this a try this weekend and see what I can come up with.

    Jon

  • Jonathan Mikolin

    Update:

    It worked! Both for the hemlock change detection and index layers between time periods.

    For the tree change detection, I followed your suggested workflow above. At step 7, I ended up using the Classification algorithm after setting up class definitions as shown in the Change Detection using Maps video.

    For creating index layers, both the index layer calculation and layer arithmetics algorithms worked great.

    Thanks again,

    Jon

  • Christian Weise

    Nice!

  • Matthias Staengel

    ヽ(´▽`)/ Looks perfect! 

     

Recent Discussions

Need more help?

Enter Knowledge Base Contact Support