Presumably the bots would be programmed to recognise key characteristics of the particular crop, then instructed to eliminate any non-conformers.
But that still doesn't change the time it takes to anlyze morphology, which is the slow step. Whether your database for comparison has 1 entry to compare to, or thousands, won't affect processing times much. DBSCAN and K-Means are very fast and effective clustering algoriths...
It looks, from what you've all said, that were talking exclusively about visual ID. Would there not also be a detectable chemical signature for a given crop type?
Possibly, although our technological ability to do that kinds of work is pretty limited - hence why dogs are still the preferred way of doing those kinds of detections. And I doubt it would work in the cases like canola farming, where the major weeds are simply wild strains of canola.