AI meets the farmer, helping make each plant count

Agriculture is not one of the first markets people connect with fledgling technologies such as artificial intelligence (AI). But for agricultural and construction equipment manufacturer John Deere, AI is already starting to help farmers improve crop yield and maintain better control of daily farm tasks.

In a phone interview with FierceElectronics earlier this week, Sona Raziabeegum, Strategy Lead, Digital Solutions for John Deere Intelligent Solutions Group, said AI is being adopted as part of high technology solutions the company is implementing for agricultural facilities which must deal with variables such as bad weather, poor soil conditions and the constraints of a short growing season.

“There is not a lot of do-over in agriculture,” said Raziabeegum. “We want to convey the message that there is a lot of great technology in agriculture solving real problems. It is not technology looking for a solution.”

Raziabeegum gave the example of an application where machine learning was used to “train” vision cameras mounted on farm equipment to recognize bad wheat as it is being harvested into a grain bin. A neural mask processes this vision data and instructs the machine to separate the bad grain.

In addition, Raziabeegum noted that the neural mask is programmed to spot any debris that may get into the grain bin, which would also adversely affect grain quality.

The benefits are better quality of grain and a reduced need for skilled labor. “You’re doing this in real time. Farmers operate on thin margins and are open to the technology’s benefits on their yield and crop quality.”

In another application, Raziabeegum said a herbicide sprayer system, called See and Spray, employed vision cameras with a neural net that helps precisely identify where the herbicide should be sprayed. The technology is currently being demonstrated in Mississippi for a cotton field and in a midwestern location for corn, she noted.

“We believe the technology in this application can reduce herbicide use by as much as 90%, which helps achieve the goal of using less chemicals in farming.”

Because AI has not been used in agriculture previously, Raziabeegum said John Deere invested a lot of time and effort learning to apply the technology in a new use case. It is a different set of parameters than say, automotive, which Raziabeegum said agriculture is often compared to.

“In agriculture, it is not just traveling from point A to point B. Once you get there, you need to perform a complicated task. Also, because [of] the unpredictability of agriculture, your AI models have to be a lot more sophisticated in order to apply to crops.”

Because agriculture is largely an outdoor application, there’s no enclosed, central onsite facility where one can implement a server and a communications network. Edge computing, with processing onboard the machines doing the work, becomes vitally important.

“Connectivity not a given,” Raziabeegum explained. “About 95% of computing on the edge needs inference in real time. Our edge network architecture had to be custom designed.”

Because John Deere is largely an equipment manufacturer, it had to acquire some its expertise in AI and machine learning from outside the company. The firm did just that in 2017, when it acquired Blue River Technology, a company specializing in applying machine learning to agriculture, based in Sunnyvale, California. Blue River developed the See and Spray integrated computer vision and machine learning technology for herbicide spraying that John Deere is now implementing.

According to Raziabeegum, the average size of the farm John Deere is supporting is 5,000 acres. But she emphasized the company is able to scale technology solutions for facilities that are smaller as well.

For the future, Raziabeegum foresees expanding the use of AI and machine learning to develop complete, detailed plant growing and maintenance plans. She also expects AI to be leveraged more for construction applications.