Great suggestion, that's the approach I've been looking at most recently (manual selection of crop first, then classifier for only that crop's diseases) and it

Great suggestion, that's the approach I've been looking at most recently (manual selection of crop first, then classifier for only that crop's diseases) and it does give better results. Still, many plant diseases can look quite similar - more training data once we get it (and more variety in the photos) should help accuracy