Round Down: John Deere Brings Machine Vision to Precision Agriculture
In 2017, the world's largest tractor company spent $305M to acquire a startup that distinguishes lettuce from pigweed. As a result, US farmers are freeing themselves from Monsanto's Roundup. Is John Deere's machine learning approach the future of ag tech?
Source: James Vincent, 鈥淛ohn Deere Is Buying an AI Startup to Help Teach Its Tractors How to Farm,鈥� The Verge, September 7, 2017,
, accessed November 2018
“Take one man and put him on a tractor, and we can do the work of 8 to 10 people.” – Clint Bigham, Blue River client [1]
With global food demand slated to increase up to 98% by 2050聽[2], farming yield en route to drop 2-3% due to climate change [3], and a superweed crisis emerging in response to reliance on a narrow range of herbicides聽[4], the agriculture industry faces an unprecedented need for innovation. Recent years have seen a spate of consolidation and investment, with a record $10.1B pouring into agricultural technology (鈥渁g-tech鈥�) startups in 2017. Last year also featured a bumper crop of acquisitions, including purchases of farm software company Granular by DowDuPont for $300M and agricultural robotics startup Blue River by Deere & Company for $305M.聽[5]
These acquisitions typify two dominant trends in the farming industry: continuing progress in 鈥減recision agriculture鈥�, the use of sensors and algorithms to manage site-specific crop and soil variability聽[6], and the emerging space of agricultural machine learning. The latter industry is expected to grow 23.2% year-over-year through 2023, concentrated in crop monitoring, robotics, irrigation, and farm drones.聽[7] With data availability multiplying 1000-fold in the past two decades and software reaping concomitant performance improvements, machine learning investment over the next decade will prove not merely beneficial but indispensable to agricultural companies鈥� survival.聽[8]
Deere & Company, the world鈥檚 largest farm equipment manufacturer with $29.7B in 2017 revenues聽[9], is an established leader in precision agriculture and innovator in autonomous navigation. The company鈥檚 pioneering AutoTrac™ guidance system, together with competitors, is used in 60-70% of U.S. farm acreage聽[10]. The Blue River acquisition enhances Deere鈥檚 technological portfolio, enabling its tractors to identify specific crops and weeds using the same deep learning algorithms underlying facial recognition聽[11]. Now used in lettuce production with 10% US adoption, Blue River鈥檚 software replaces the current standard of broad-based Roundup sprays, made costlier and less effective by herbicide-resistant 鈥渟uper-pests鈥�, with micro-sprays the size of postage stamps. Technology director Willy Pell claims that the successor for cotton will reduce herbicide usage by 90%. Soybean- and corn-specific products, as well as harvesting and seed-planting products that incorporate soil type and crop size, are currently under development.聽[12]
Blue River Technology,聽鈥淪ee & Spray – Blue River Technology鈥檚 Precision Weed Control Machine,鈥� YouTube,
published August 23, 2017, , accessed November 2018.颅颅
Looking forward, ag-tech machine learning presents an especially attractive opportunity for large firms because of the scale of data required. Today鈥檚 algorithms require tens or hundreds of thousands of training images [13] to achieve competence, further multiplied by crop and geographical variation. Deere thus enjoys a rare competitive advantage by using the same field data that informs its farmers鈥� chemical regimens to enhance its product lines. This feedback loop will enable Deere to create a defensible IP moat and reinforce its autonomous navigation offerings, where costly false negatives (e.g. disregarding safety hazards) and weather variability remain serious challenges.聽[14]
Over the next decade, Deere must capitalize on its reach and culture of innovation to preserve its advantage among equipment manufacturers and the broader agriculture industry. Diversifying its current precision-agriculture lineup with further robotics investments will allow the company to offer a cohesive product suite, delivering unparalleled crop yield and cost savings. Deere has recently moved in this direction, integrating with ag-drone data platform provider Agribotix聽[15] and drone analytics startup Sentera聽[16], whose AgVault utilizes high-resolution imagery live in the field. [17] Deere also stands to reap continuing rewards from collaborations with academia, where it has sponsored and licensed past research.聽[18]
Finally, Deere must embrace Blue River CEO Jorge Heraud鈥檚 vision to move 鈥渇rom the field level to the plant level鈥using] machine learning to help smart machines detect, identify, and make management decisions about every single plant鈥�.聽[19] Generalizability remains the grand dream of machine learning; Deere鈥檚 scale and diversity present a unique opportunity to thrive by contributing to its realization. A decade from now, Deere may not need to customize software per-crop 鈥� given sufficient training and algorithmic advances, its systems could not only generalize across observed data, but also infer the spaces in between.
As rising demand, declining yield, burgeoning investment, and data proliferation make continued development inevitable, investors and companies must ask: which innovation segment will dominate? Double-digit growth looks set to continue on all fronts [20], but a single breakthrough, whether in machine vision or microbiology, could transform minor players into multibillion-dollar juggernauts. Additionally, which companies will lead the charge, and which regions will see the greatest change? Large firms have thus far maintained their advantage through acquisitions, but growth investors such as Softbank, whose globally oriented Vision Fund recently poured $160M into Zymergen, are redrawing the competitive map. In the ag-tech industry, as with the farms it promises to revitalize, the direction has long been clear, but 鈥渢he devil is in the data鈥�. [21]
(778 words)
Source: Randall Munroe, 鈥淗ere to Help鈥�, XKCD, , accessed November 2018.
[1] Blue River Technology,聽鈥淪ee & Spray – Blue River Technology鈥檚 Precision Weed Control Machine,鈥� YouTube, published August 23, 2017, , accessed November 2018.颅颅
[2] Maarten Elferink and Florian Schierhorn, 鈥淕lobal Demand for Food Is Rising. Can We Meet It?鈥澛�性视界 Business Review, April 7, 2016, , accessed November 2018.
[3] David Reid, 鈥淯N Report Shows Climate Change Effect on Farming,鈥� September 17, 2018, , accessed November 2018.
[4] Union of Concerned Scientists, 鈥淭he Rise of Superweeds鈥攁nd What to Do About It,鈥� , accessed November 2018.
[5] Reuters, 鈥淕lobal Ag Tech Startup Investments Rise 29 Percent in 2017,鈥� March 6, 2018, , accessed November 2018.
[6] Tom Meersman, 鈥淧recision Agriculture Becomes Mainstream in Minnesota,鈥� Star Tribune, , accessed November 2018.
[7] Prescient & Strategic Intelligence, 鈥淎I Market in Agriculture Size, Global Industry Report, 2023.鈥� , accessed November 2018.
[8] Erik Brynjolfsson and Andrew McAfee, 鈥淲hat鈥檚 Driving the Machine Learning Explosion?鈥澛�性视界 Business Review, July 1, 2017, , accessed November 2018.
[9] Statista, 鈥淢ajor Farm Machinery Manufacturers – Global Revenue 2017鈥�, , accessed November 2018.颅颅
[10] The John Deere Journal, 鈥淛ohn Deere Hands-Free Guidance System Continues Its Evolution,鈥� March 29, 2016, , accessed November 2018.颅颅
[11] James Vincent, 鈥淛ohn Deere Is Buying an AI Startup to Help Teach Its Tractors How to Farm,鈥� The Verge, September 7, 2017, , accessed November 2018.颅颅
[12] Tom Simonite, 鈥淲hy John Deere Just Spent $305 Million on a Lettuce-Farming Robot,鈥澛�Wired, September 7, 2017, , accessed November 2018.颅颅
[13] James Vincent, 鈥淭hese Are Three of the Biggest Problems Facing Today鈥檚 AI,鈥� The Verge, October 10, 2016, , accessed November 2018.颅颅
[14] Dave Gershgorn, 鈥淎fter Trying to Build Self-Driving Tractors for More than 20 Years, John Deere Has Learned a Hard Truth about Autonomy,鈥� Quartz, , accessed November 2018.颅颅
[15] Jason Reagan, 鈥淎gricultural Drone Provider Inks Major Deal with John Deere, 鈥澛�DRONELIFE聽(blog), January 4, 2018, , accessed November 2018.颅颅
[16] Cision, 鈥淪entera Announces Series A Funding,鈥� September 5, 2018, , accessed November 2018.颅颅
[17] Sentera, 鈥淎bout Sentera,鈥� , accessed November 2018.颅颅
[18] The John Deere Journal, 鈥淛ohn Deere Hands-Free Guidance System Continues Its Evolution,鈥� March 29, 2016, , accessed November 2018.颅颅
[19] John Deere US, 鈥淒eere Acquisition of Blue River Technology,鈥� , accessed November 2018.颅颅
[20] Prescient & Strategic Intelligence, 鈥淎I Market in Agriculture Size, Global Industry Report, 2023.鈥� , accessed November 2018.
[21] Arama Kukutai and Spencer Maughan, 鈥淗ow The AgTech Investment Boom Will Create A Wave Of Agriculture Unicorns,鈥� Forbes, , accessed November 2018.颅颅
I completely agree with you that the “devil is in the data”. In this case, the availability of training data is the key element to success. Thus, I believe John Deere is much more set up for success than any other start-ups because of its unique access to create training data through its machines. This is a prime example of training data being a barrier to entry. Deere’s machines are already used by farmers; there is no additional work to do for adoption.
In other cases, start-ups can get access to data through signing exclusive contracts from data providers, or through creating the data themselves (e.g., deploying field teams or cameras/drones to create picture data). However, for agriculture, you need data across large spans of land, and Deere is uniquely positioned to take advantage of it.
However, a question remains on whether Deere should monetize this simply by selling the rights to the data, and focus on their core competency of machines, or whether they should develop new capabilities for machine learning.