Unsupervised change detection from satellite images

(Priit) #1

Hi all,
I would like to ask if anyone has experience working on satellite image segmentation problems.
I am thinking of a use case of detecting changes in the images by first doing semantic segmentation on two images of the same location (different times) and then a comparison to flag out any changes. For example mapping out areas of deforestation.

I am currently looking for the best dataset to work with it. So far the closest one seems to be from the Kaggle DSTL challenge from 3 years ago. Maybe anyone knows, has tried or found any other dataset?

Also would be cool to hear if anyone has tried something similar and how did it turn out.



(Antonin Sumner) #2

I had the idea to do some segmentation on satellite images of glaciers and icepack, but haven’t digged into this yet.

1 Like