Current progress and boundaries
Crop identification and disease classification apps are available, but clear laboratory photos differ substantially from real field images.
Why it remains difficult
Lighting, variety, growth stage, and mixed infections all affect diagnosis. Recommending chemicals based on a single image can create economic and ecological risks.
Paths worth exploring
Start in one region with a few common crops and have the model ask about key conditions first. Refer low-confidence results to agricultural advisers and provide safety guidance.
How to evaluate success
Test identification accuracy, harmful recommendation rates, human referral rates, and outcomes after farmers act, in the field and across different seasons.