The same probably happens with blurry small aircraft in the scene. It'll "upscale" (i.e. draw in) all kinds of objects with what it thinks is most likely from the context, from its training set.