This is not a definitive answer, but it is generally accepted that Ed Catmull introduced Texture Mapping in his 1974 thesis, "A SUBDIVISION ALGORITHM FOR COMPUTER DISPLAY OF CURVED SURFACES"
In that, he uses U,V to access the image data (page 42)
MAPPING
Photographs, drawings, or any picture can be mapped onto bivariate patches. This is one of the most interesting consequences of the patch splitting algorithm. It gives a method for putting texture, drawings, or photographs onto surfaces....
...If a photograph is scanned in at a resolution of x times y then every element can be referenced by u·x and v·y where 0<=u,v<=1. In general, one could think of the intensity as a function l(u,v) where I references a picture.
I believe this thesis also introduced the concept of the Z-Buffer.