While such logic might apply to parts of UK, you’ll end up with negative consequences elsewhere.
There is no such thing like an easy recognition of such structures from top-down imagery.
I don’t want to have barns everywhere where there is a farmer’s house…
The image recognition is good for finding buildings and assigning correct roof color and footprint (rounded, rectangular etc), the rest should be fed by proper data, otherwise you end up fixing one location and cause wrong depiction everywhere else.