I am not sure that it helps but I saw somewhere the following simple way that  is to consider a projected (in the plan perpendicular to the shaft) mesh with circles (like a "polar mesh") and to count the number of signal changes in the mode along all the circles but checking radius by radius. But I dont know if torsion of the blades may compromise this.

Lets say, if all circulars change signal along one radius then is highly probable to be a nodal diameter. Of course as frequency increases it may gets more complicated.

It is just an idea. There are more complicated techniques for object identification used in computer vision but I never needed to go so far. May be some specialist in that areas can give you a better response.