- Five years after acquiring the technology, Deere's method of using machine vision and machine learning to identify individual plants will be tested on farms in summer 2021.
- The goal: Farm machines moving at up to 20 miles per hour make decisions on spraying of herbicides at the level of individual plants and weeds in seconds, reducing the need for widespread use of chemicals.
- The AI effort is one of many technology projects associated with precision agricultural taking farming into the 21st century and changing the nature of work in rural America.
If it comes as a surprise that a recent government auction of 5G broadband licenses was won by agricultural giant Deere & Co. rather than AT&T or another telecom stalwart, maybe it shouldn't. Farming — which over thousands of years evolved from humans pulling plows to chemical and most recently its genetic era — is entering its digital age. Also referred to as precision agriculture, the changes being wrought by collection and analysis of data, on life and work in rural areas, are set to accelerate.
One example from Deere that is set to debut in the farm fields next summer combines machine vision and machine learning — or, to put it in words easier to understand, think facial recognition for plants. Back in 2017, Deere acquired a company called Blue River Technology, which has been working on a way to identify individual plants and weeds. That's not an easy task when you consider that a single acre of farm can encompass thousands of plants and the heavy machine moving through the field is operating at a speeds of 10-20 mph.
AI is quickly moving into all varieties of farms and on a global basis. In China, pork farms have been using facial recognition to map and monitor pigs' faces. And from an Irish start-up to ag giants like Cargill, facial recognition of cows for dairy farms is advancing.
"Farms in these rural environments are very technologically capable, tech savvy, creating significant data each and every day during the growing season," said Jahmy Hindman, who became Deere's chief technology officer this past July, speaking at Thursday's CNBC @Work Spotlight event. "The information being created is really going into helping them be me more productive and sustainable and more precise. .... Information is really critical to making decisions in the moment, minimizing inputs farmers have to put into the business and maximizing productivity."
If the AI technology works as envisioned, the primary input that would be reduced is chemical applications to kill weeds in the fields, herbicides. Instead of widespread spraying of chemicals killing everything but genetically modified plants designed to survive the application, sprayers could target individual plants recognized as being the correct targets, which could have major implications for businesses like Bayer's Monsanto, which creates chemicals and GMO crops, the most well-known being Roundup.
Money Report
Hindman described the AI technology as training new neural network models to see weeds and spray only weeds in crop fields.
Getting more information to the grower at the individual plant level is a key goal for Deere.
"Think about corn or soy operations in the Midwest ... 40,000 plants per acre on a farm 2,000 acres large," Hindman said. "We are interested in being able to manage each plant over the course of its life, minimizing inputs and maximizing productivity. ... Being able to make decisions in real time is absolutely key to unlocking added economic value for growers and productivity in the agriculture space."
Farm facial recognition
The Blue River Technology approach, down to the level of the individual crop plant — taking pictures of plants so that while a machine is cruising it can make the decision to spray within seconds, or less — is potentially the most important technology coming to the farm, according to Rob Wertheimer, an analyst with Melius Research who covers Deere.
Between seasons, farmers spray herbicides like Monsanto's Roundup on entire fields to kill everything. Deere's intention is to launch Blue River in fallow fields as the first experiment, rather than fully planted crop rows. In spring and summer, before planting, weeds have grown in empty fields and that is not as complex a task for AI as identifying targets in fields where there already are thousands of crops planted, but it is the first step in proving the technology.
"You're taking pictures of plants and training algos that need to make spraying decisions fast, in seconds, at fast speeds, 15-20 mph, and bouncing around, the sprayer bouncing around and doing it day after day for five or 10 years with no errors. That's hard," Wertheimer said.
As in many sectors, the pace of technology change on farms is occurring much faster than the industry anticipated. Wertheimer noted that only a decade ago, former Deere CEO and chairman Sam Allen thought it would be a long time before autonomous tractors took over farms, for reasons including safety issues. But with rapid improvements in self-driving technology like Lidar, as well as AI improvements, Allen changed his view within the span of a few years.
"The farmer doesn't drive much anymore," said Stephen Volkmann, a Jefferies analyst who covers Deere and compared autonomous advances in farm operations to an aircraft pilot, where today much of the flight is automated. "The farmer needs to sit in the cab and monitor, but lets the tractor drive itself."
Volkmann said the see-and-spray AI is the "sexiest" technology coming to the farm.
"I think people believe it's real," he said. "This is exactly like an autonomous car, a camera that can recognize lots of stuff and train it with AI algos and identify lots of different plants."
The challenges to making it work are numerous: plants get stepped on and leaves get bent and there are shadows created in fields, and fields are dirty places, which means reliably performing this task all the time is a challenge, and it is a task that requires a high level of success.
"Just like self-driving, they can do it 95% of the time today, but that's not good enough. You need to get to 100% to call it success. You don't want to spray the wrong chemical on the wrong plant even 5% of the time," Volkmann said.
Ultimately, there is potential for AI to learn to recognize "good" plants versus "bad" plants using a variety of factors, as well as the best locations for planting, rather than just target the right weeds for spraying.
Today, a corn farmer may get on average 170 bushels produced from an acre, though a record level of 600 bushels per acre has proven to be possible, if weather and weeds and other factors in the field, from sunlight to insects and fungus, soil nutrient characteristics and sunlight and shadow, can be analyzed to ultimately create greater crop productivity.
"There is lots of data in millions of plants and weeds," Wertheimer said.
Deere already offers ExactEmerge and ExactApply technology which were introduced over the past decade and have turned core farm tasks such as seed planting and spraying into precision agriculture machine operations, and Deere executives said on its most recent earnings call that the uptake of these technologies by farmers is accelerating.
"Facial recognition is getting a little scary ... but there is no reason to think it can't be successful," Volkmann said. "See-and-spray is one of several advanced farming technologies that seem to be moving closer to an inflection point," he said, though he added it is still likely a few years out for full plant recognition technology to be commercialized.
Deere and 5G
Rural connectivity is tied to these technology efforts that Deere is focused on for its operations and the rural communities in which its farmers work and live. While the 5G licenses the company recently acquired are for its manufacturing operations — allowing it to operate smart factories — Hindman said there are tailwinds to bringing more broadband and 5G to rural America.
"The divide between urban and rural connectivity is an important one for us, and farmers, and also important in rural communities which they happen to work in for reasons far outside the scope of agriculture," he said.
For farmers, more investment is needed to support data flows between Deere's own cloud computing center and farms, for reasons including the ability to remotely monitor heavy machinery on farms for preventive maintenance needs (e.g. a water pump being repaired remotely rather than someone having to travel out into the field), as well as for remote operation of equipment in the future. The effort is underway through partnerships with government and private enterprises, the Deere CTO said.
Hindman said with 5G bandwidth and the latency reduction it offers, automatically controlling machines on the farm from a remote location become a possibility.
"There's a whole host of benefits that come to society when that happens. ... We're confident the winds are at our back on that," he said of federal government support for 5G rollouts in rural parts of the country.
Hindman said hiring at the company has changed, as well as training of current employees, in line with newer efforts like the plant recognition AI and other technology.
Machine learning skill sets are in high demand, and in general, Hindman said in recent years Deere's hiring has been "significantly more indexed to software skills," while there has been concurrent upskilling of existing employees to meet the needs of the latest technology.