Most farms spray their entire field, even though weeds, pests and disease usually show up in patches. US farmers spend $21.7 billion a year on chemicals, and much of it lands on plants that never needed it.

AgriVision fixes that. We split each field into a grid of small cells, and a drone flies over and checks the health of every one using infrared imaging. It then goes back to the cells that look off and takes close-up photos. Our model looks at those photos and tells whether the problem is a weed, a disease, insect damage, or something a spray won’t fix, like dry soil. Then it decides whether treating it is actually worth the money.

The farmer gets a spray map that works with the equipment they already own, so they treat only the spots that need it.

We trained our model on more than 11,000 crop images from 12 research datasets, covering soybean, corn, cotton and sorghum, from drone shots down to single leaves. On images it had never seen, it identified the problem correctly 98% of the time, and it caught weeds more than 99% of the time.

Targeted spraying has already cut chemical use by 59% on large farms using expensive equipment. AgriVision brings that to every farm, with just a drone.