Agriculture embraces artificial intelligence

Graybeards may remember the thrill they felt when pencil-laden math calculations moved warp speed ahead into the calculator age.

These days, artificial intelligence (AI) promises to bring the same heat to agriculture that it did to math classes decades ago. Artificial intelligence is a technology that includes several subsets such as machine learning, says Rania Khalaf, Inari chief information and data officer.

READ MORE: Space for ag: NASA works to ensure food security

“Machine learning enables computers to mathematically predict outcomes or make classifications by finding patterns in large amounts of data,” she says. “It then learns to update these patterns or classifications over time as it sees new data.”

“The biggest advantage of artificial intelligence is the ability to make complex calculations at a high speed that previously required a human to perform,” adds Kent Klemme, general manager of See & Spray for Blue River Technology. “The recent improvements in GPUs [graphics processing units] have provided the computing power to make this possible. It takes a lot of data to target specific problems.”

See & Spray Ultimate technology — powered by machine learning — enables sprayers to target just weeds while spraying among crops. “We’ve taken thousands and thousands of images of different weeds in different crops under different situations such as clear skies, cloudy skies, dark skies, different soils, and varying levels of residue,” says Klemme.

Blue River and John Deere data scientists then train the See & Spray Ultimate system to recognize plants under myriad conditions. These images are sorted out through algorithms, which involve repetition of one or more mathematical operation. Algorithms are often implemented and solved on computers.

Patriot 50 series sprayers from Case IH use a form of machine learning called vision guidance.

“It’s a row guidance solution that makes a steering command based on plant location,” says Chris Dempsey, global precision technology director for Case IH.

Vision guidance uses an on-board camera that relays corn plant location to the sprayer so it stays on the row rather than run over crop plants, he adds.

It’s Growing

Artificial intelligence is a broad area that includes many subsets, such as machine learning. Basically, though, it uses reams of data to drive efficiencies, says Dempsey.

“The biggest obstacle that the whole agricultural industry faces in digital farming is taking large and complex data sets and turning them into meaningful insights,” says Ashwin Madgavkar, founder of Ceres Imaging. “AI can really help bridge that gap by synthesizing all this data into actions that a grower can take.”

Its use is increasing in crop breeding.

“We’re looking into all kinds of different new technologies, whether it’s machine learning or advanced analytical models to predict hybrid performance,” says Mike Popelka, AgReliant Genetics hybrid product breeder manager. “The [crop breeding] industry is going more toward these models.”

READ MORE: Tech companies hone in on carbon measurement

AI use is also commonly being used across many machinery lines. Case IH uses machine learning through 16 sensors that adjust its AFS Harvest Command system.

“Increasing throughput while decreasing [grain] losses is all now done automatically,” says Dempsey. “Historically, combine operators would have to make sieve adjustments if they had too much cob or foreign material in a grain sample.”

Machine learning now does this automatically.

“Sensors tell the combine it needs to shut down a lower sieve or increase fan or rotor speed,” he says. “Those adjustments are based on knowing what foreign material or bad grain quality looks like in a known database of a given crop type. It’s basically a database of images showing good [quality grain] from bad.”

John Deere also uses machine learning in its Auto Maintain feature on S700 and X9 series combines to maintain targeted grain loss and grain quality performance.

“The operator will set a target as to how the combine is to run in terms of loss levels and grain tank sample,” says Nick Howerton, product marketing manager at John Deere Harvester Works. “ActiveVision cameras take pictures of the clean grain and tailing elevators every two seconds. This data feeds into an algorithm which is used to compare foreign material and damaged grain levels with the target. If levels exceed the target, adjustments are automatically made.”

Smart Tillage
“Tillage solutions used to be simple,” says Chris Dempsey, global precision technology director for Case IH. “You put an implement in the ground and it tilled the soil.” That simplicity sometimes backfired. “If you had your machine set wrong, it would create ridges,” he says. This spurred planter hopping, which raised havoc with stand uniformity and ultimately yield. Technologies spurred by machine learning such as Case IH’s AFS Soil Command agronomic control strategy changes this, says Dempsey. It automatically adjusts the optimum tillage depth for different soil types and conditions. Dempsey sees this type of precision expanding into other implements. “Everything from a baler to a field cultivator to a combine to a sprayer has, or will have in the near future, some level of machine learning technology,” he says.

Smart Tillage

“Tillage solutions used to be simple,” says Chris Dempsey, global precision technology director for Case IH. “You put an implement in the ground and it tilled the soil.”

That simplicity sometimes backfired.

“If you had your machine set wrong, it would create ridges,” he says. This spurred planter hopping, which raised havoc with stand uniformity and ultimately yield.

Technologies spurred by machine learning such as Case IH’s AFS Soil Command agronomic control strategy changes this, says Dempsey. It automatically adjusts the optimum tillage depth for different soil types and conditions.

Dempsey sees this type of precision expanding into other implements. “Everything from a baler to a field cultivator to a combine to a sprayer has, or will have in the near future, some level of machine learning technology,” he says.

In-Season Decisions

Ceres Imaging uses a form of artificial intelligence called computer vision that detects in-season crop issues, says Madgavkar. Aerial imagery and sensor data that detect different wavelengths of light are fed into an algorithm that helps reveal in-season maladies, such as nutrient deficiencies.

“We also look at disease risk and where a fungicide might be optimally sprayed,” Madgavkar says.

Still, data gleaned by machine learning is only as good as data that is input, Madgavkar says. “Garbage in, garbage out” still applies, so it’s important to input the right quality of data, he adds.

“AI alone is no silver bullet,” he says. “It’s still important to combine computer vision-derived data with a local agronomist and or/farmer input to make a recommendation. However, AI technology can help an agronomist make better use of his or her time in targeting fields most likely to benefit from in-season treatments.”

Irrigation System Repair

AI technology is also making irrigation repair easier. Lindsay is collaborating with Microsoft Azure in using machine learning to predict irrigation component failures before they occur through Smart Pivot technology.

READ MORE: Ceres Imaging and Soiltech Wireless partner to optimize irrigation strategies

“If we can predict component failures, such as a gearbox or center motor drives, we can notify customers before they occur. This prevents even bigger [repair] issues from occurring during the growing season,” says Kurtis Charling, vice president of digital product management at Lindsay. “Being down one to three days during the peak of the growing season could be detrimental to production. Predicting component failures can help the farmer avoid downtime headaches and make them more efficient.”

Lindsay uses extensive irrigation system data gleaned over decades to machine- rain a predictive system.

“For example, a machine that failed because a gearbox went bad was a training situation for the artificial intelligence model,” says Charling. “If we have access to sensor data we collected from the time of the failure, we can give it to the AI model and basically tell it ‘If this happens again, this is what a gearbox failure looks like.’ "

Limitations

AI has its limits, though.

“There are situations where something we predict isn’t true in the field, such as a gearbox issue,” says Charling. “It’s still important to have the end user or dealer who can say yes or no. This in turn feeds back into our training models to make them more accurate.”

In the See & Spray Ultimate model, differentiating young soybeans from young velvetleaf is challenging because they look similar.

“As we take more images and retrain the model, it gets better,” says Klemme. “It’s no different than the time a human scout would take to tell the difference between velvetleaf and young soybeans. You train the [machine learning] model in the same way.”