Whole-site orthomosaics segmented into vegetation, water and bare ground across tiles stitched from multiple flights.


Site-scale drone orthomosaics were segmented into land cover classes across the full survey extent, including the site boundary itself.
Working at mosaic scale changes the failure modes. A single image is a consistent exposure from one moment, but a mosaic is assembled from many passes, so colour and shadow shift across the sheet and a class boundary that looks obvious in one tile is invisible two tiles over.
A continuous land cover segmentation across the full site, with class regions that hold together across stitching seams and exposure changes.
Delivered site-wide land cover polygons referenced to the survey boundary, continuous across tiles and ready for area computation.

Green cover, residue and bare soil segmented inside survey quadrats, with the quadrat frame itself located per image.
Polygon masks isolating black grass from crops and similar-looking grasses across dense agricultural field imagery.
Polygon masks tracing blueberry plants through dense foliage, overlapping branches and look-alike vegetation in open field conditions.