I am detecting rooftops and output them as shapley polygons in square, U and L shape. Sometimes the output is in U shape with a very thin rectangle on one side ( true shape is L shape) which needs to be cleaned. I have tried using perimeter/area ratio to detect such parts of a polygon and remove them but I am not able to implement it.
An example is shown in figure below. 
I am looking for a solution preferably with shapely.
The polygon in the example in GEOJSON can be found here.
If small changes in the shape that cannot be visually detected are not important, try this way:
from shapely.geometry import shape
test = {"type": "FeatureCollection", "features": [{"id": "0", "type": "Feature", "properties": {}, "geometry": {"type": "Polygon", "coordinates": [[[6.980710220282101, 51.243221513354044], [6.980706911073409, 51.24322119551712], [6.9806699432069745, 51.24337283218559], [6.980864165775662, 51.24339148636556], [6.980901140490866, 51.24323981899073], [6.980795641198141, 51.24322968631755], [6.980776365429034, 51.243308752807856], [6.980690951372498, 51.243300549145346], [6.980710220282101, 51.243221513354044]]]}}]}
poly = shape(test["features"][0]["geometry"])
d = 0.00001 # distance
cf = 1.3 # cofactor
p = poly.buffer(-d).buffer(d*cf).intersection(poly).simplify(d)
You may need to change d and cf values. This is not a perfect solution but it can solve your thin rectangle problem by small changes.
