I'm using the algorithms from OTB, to segment and perfom remote sensing on mangrove areas.
The idea here is to delineate mangrove cover from different satelite images and, later, compare mangrove cover between the different images.
The result isn't as accurate as I need. I make some corrections by comparing each feature to its neighbors and, when I have small areas without any neigbhors of the same attributes intersecting, I eliminate it by merging it with the largest area.
For example, here, the dark green features will be merged with the brownish green one (not all, depending on their size, which will leave some unmerged, sadly). It works, but I'm also loosing good ones, like the one circled.
In another area, I have small patches of mangroves. Those patches must be kept, but they are also removed from my operation. See picture below.
Don't pay attention to the differences in the delineation of the polygons, the pictures are not taken from the same segmentation output.
I can't use an expression to filter the result and avoid merging mangrove patches, because it would need a lot of exception and checks.
Do you have any idea on how I could resolve my issue ? I can't see any post-treatments that could help remove these errors.