Bob Buckley Aug 13, 2026
https://myfox8.com/spotlight/buckley-report/nc-at-state-university-uses-ai-in-farming/
GREENSBORO, N.C. (WGHP) — The way we feed the world has taken leaps forward as human ingenuity pushes our abilities forward.
The combine harvester, when married to gas-powered tractors that freed it from the labor of horses, could easily replace the work of three dozen men. That made work go not only more quickly but cost less as well.
Artificial intelligence may do even more for agriculture in the long run, but it is already doing a lot. North Carolina A&T State University is leading the way in that arena, and Professor Harman Sharma is a big part of that effort.
Sharma is developing AI systems that can do some of the most tedious and back-breaking tasks. She is currently working on a model that can both tell when crops are about to develop health issues and identify weeds within a field and pull them while leaving the crop in place to thrive.
“It’s capturing the imagery,” Sharma said. “And then once we have that imagery, we are training through AI to detect it. … It’s like reinforcement learning, like we teach our kids, right? So it can go in the field by itself, and it can detect … this is the weed, this is the crop. And the next step is, once I detect the weed, it has the capability to even go and just take out that particular weed.”
And maybe the “cutest” part is that the robots they’ve developed to do this work are little dogs. But AI is software, so Sharma and her team have to teach it what to do.
“AI is only as good as your data is good,” Sharma said. “It’s called garbage in, garbage out. So our … focus is how we can develop and have … high-quality, high-resolution data, and then based on that data, we should train our dog.”
But there is a lot of work to do before the training.
“After collecting all those images, the next step is we have to annotate those images so we can train our model what is a weed and what is a crop itself,” Sharma said.
“We will pick out the type of images we will be training the model on. … The better the image quality, the better the model,” said Sharma’s graduate student research assistant, Edmond Kwekutsu. “So the next step is to employ this model into the dog itself.”
Sharma said that the work of annotating the photos is kind of like building a video game.
“That’s how that idea came to my mind. … How when we play video games, it can generate the bounding boxes around things that we want to kill or something,” she said.
So her students build those boundary boxes in the photos to teach the robot dogs what they need to take out. Large, commercial farms will use the protocol to tell the robots what to spray with pesticides. Sharma’s work is more oriented to smaller, organic farms that aren’t allowed to use pesticides. Sharma believes that in the not-too-distant future, the robot dogs will be able to not just identify the weeds but pull them as well.
She said the technology is not too much more for most farmers to use.
“They have already adopted some of the low-cost technology,” Sharma said. “They’re using some kind of sensors already, so the main issue whenever we talk with farmers is the cost.”
It will be a little pricey for a small farmer on the front end, but Sharma has an argument for why it’s a worthwhile investment.
“Think about large-term return on investment . … It might cost initially, but if you look over long-term, how much it’s going to save you in terms of water, fertilizers, herbicides, so it’s going to even out at the end,” she said.
See the robot dogs in action in this edition of The Buckley Report.