I am doing my thesis research about analyzing settlements in Mali with Sentinel2 images with the use of the program eCognition (object-based). We have trouble with determining the settlements because they have almost the same spectral values as the surrounding (the houses are from loam, the same as the ground, see picture..).

A few things I already tried / analyzed for inserting conditions for classification:

  • NDBI
  • NDVI
  • Means from all the Sentinel-2 layers
  • Standard deviations from all the Sentinel-2 layers
  • Brightness

In the picture beneath, the area and two examples of settlements are visible.

I hope you could give some suggestions.

enter image description here

enter image description here

enter image description here

  • I haven't worked with eCognition, but does it have any capability for shape/texture identification? Though the settlements have similar spectral responses to the surrounding terrain, the settlements seem to have more sudden shifts in value than the open terrain. Perhaps you could do something that looks at the variance in the neighborhood of each cell? – JoshC May 20 '19 at 14:44
  • Thankyou very much for your response. It indeed has capabilities for shape, it makes segmentations on basis of scale, shape and compactness. Maybe, looking at the variance of neighborhood, is a good idea. Thanks, I will try this. – Roos May 21 '19 at 13:43

Your Answer

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

Browse other questions tagged or ask your own question.