That is amazing. I do not know anything about plants - is this approach generically applicable for all crops or are there limitations on what kinds of weeds are likely to grow in a field? For example: weeds in wheat fields are too visually similar/closely spaced/different life cycle than the cash crop for this to work?
The further you let them grow the more differentiated they become. So time could be leaned on as a predictor. But even at the very young cotlydon phase you can differentiate between certain species, even at the level of seeds. It depends how closely related things are of course but with time every plant identifies itself pretty much.