A Drone Finds the Bare Spots in a Cotton Field, and NC State Is Building a Robot to Plant Seeds Only There

Every gardener knows the disappointment of a seed row that comes up patchy. A few feet of healthy seedlings, then a bare stretch, then a few more. In a backyard bed, the fix takes five minutes and a pinch of seed. On a cotton farm covering hundreds of acres, the same problem can mean replanting an entire field.

A team at North Carolina State University is working on a middle path. In a cotton field at the university’s Sandhills Research Station, a drone maps the gaps between plants, artificial intelligence picks out the empty stretches, and a self-driving ground robot is being built to roll down the rows and drop new seed only where it is missing. NC State’s College of Agriculture and Life Sciences described the project in a Nov. 2 feature by Dee Shore.

Growers call those gaps “skips.” They appear where seeds fail to sprout or where young plants are lost to cool or wet weather, hail or other problems. Right now, project leader Frank Bai said, when there are too many gaps, growers have no way to avoid replanting. Filling the holes one by one by hand would run up labor costs and risk damaging the plants that did come up.

Why a partial replant is worth chasing

Guy Collins, an NC State Extension cotton specialist and one of the project’s investigators, put the stakes in dollars. “Growers plant into a compressed window of time, and replanting is difficult to do in a timely manner. It also comes with significant costs,” he said. “With seed costs exceeding $100 per acre, the ability to replant only where is needed and in a timely manner could be very valuable for cotton growers here in North Carolina and beyond.”

The new project builds on earlier NC State work on the same question: how many skips are too many? A two-year study funded by Cotton Incorporated and published in the Journal of Cotton Science used drones to count plants and measure skips at three sites in eastern North Carolina in 2019 and 2020. The authors noted that the suggested retail price of a bag of replanting seed had risen by more than 25%, which made the replant decision harder. Older guidance justified replanting when half or more of the planted area was taken up by skips of at least 3 feet. The updated recommendation was lower: a replant should be triggered when 30% to 40% of the planted area is in skips of 3 feet or longer.

A case study from Solvi, the drone-analysis company whose plant-counting software the researchers used, added a few practical findings. Detection worked best when cotton plants had four to six true leaves, and flying at about 30 meters gave sharper images and earlier plant detection than flying at 53 meters. The study also found no single critical skip size. Many small gaps could hurt yields as much as a few large ones.

That last point is what makes a spot-replanting robot interesting. If a field sits just under the replant threshold, a grower may live with the loss. If a machine could fill only the gaps, quickly and without trampling the healthy plants, the math changes.

How the drone and robot work together

The system starts in the air. From an office in Weaver Labs on NC State’s Raleigh campus, research associate Justin Macialek launches the drone from a weatherproof docking station about an hour and a half away at the Sandhills station in Jackson Springs. Macialek holds a rare beyond-the-line-of-sight license from the Federal Aviation Administration, so he can monitor the flight from his computer instead of standing in the field.

The drone photographs every row with high-resolution cameras. AI-aided models then identify individual cotton plants and flag skips longer than 3 feet. Rather than shipping terabytes of images to a distant cloud service, the team processes the data on a powerful computer server at the farm. The setup came out of an NC State effort called Automating Intelligence from Research Stations, or AIRS, which upgraded computing and connectivity at the university’s agricultural research stations.

To turn the images into planting instructions, the team pairs the system with a satellite positioning network that the university says is accurate to within plus or minus one inch. The result is a digital map of every missing plant. The ground robot, still in development, is meant to receive that map, drive itself down the rows and place seed only in the empty spots.

ShengDao Du, a doctoral student in biological and agricultural engineering, has built a prototype of the robot, written its software and installed its sensors. “Traditionally, the manual data collection is very tedious,” Du said. “If we can use some machine to help us, hopefully it can help some farmers in North Carolina.”

The work is funded by Cotton Inc. Other team members include associate professor Jason Ward and lecturer Tommy Stephenson from biological and agricultural engineering, and Robert Austin, a crop and soil sciences specialist in geographic information systems.

A sandy test field and a bigger research program

The Sandhills station is a tough place to grow things, which is exactly why researchers like it. “Nearly 95% of the station’s soil type is Candor sand, and it’s very porous. It drains well and it flushes our nutrients through very easily,” said superintendent Jeremy Martin. “So we’re basically a torture chamber.”

Gardeners in the low desert will recognize that description. Fast-draining soil that won’t hold water or nutrients is a daily fact of life in many East Valley yards, and it is the kind of stress that separates sturdy plant varieties from fragile ones.

Bai’s lab uses the same drones for more than cotton. Working with the U.S. Department of Agriculture’s Agricultural Research Service, the team runs frequent remote flights over soybean fields to measure evapotranspiration, the rate at which crops and soil release water vapor, by tracking small changes in leaf temperature through the day. The goal is to help breeders find drought-tolerant soybean lines years faster. Ground sensing stations beside the fields check the drone measurements. Bai has said in an interview on the AgTech360 podcast that the long-term aim is a drone that can fly out on its own each morning and send a farmer a report on what needs attention that day.

Doctoral student Mingjun Wang is now applying the water-use approach to cotton. “We want to know if we can use the same approach to understand how much water the crop will use,” he said. “That will help farmers and maybe breeders to make better decisions.”

What it means beyond North Carolina

The spot-replanting robot is a research prototype, not a product, and the university has not announced when or whether it might be available to farmers. NC State also has not published yield or cost results from the robot itself. What is established is the problem it targets and the replant thresholds that earlier drone studies helped set.

Bai framed the motivation broadly. “Today’s farmers are under enormous stress from labor shortages, rising costs for inputs like seeds, fertilizers and chemicals, and extreme weather events, resulting in razor-thin profit margins,” he said. “We believe that the technology we’re looking at will provide farmers with affordable, scalable and reliable solutions.”

For a home gardener, the lesson scales down neatly. Check new seed rows about the time seedlings show their first true leaves, while there is still time to resow. Fill short gaps rather than replanting a whole bed, and keep a few seeds from the same packet on hand for that purpose. NC State’s engineers are trying to give cotton farmers the same option, just across hundreds of acres instead of a few raised beds.