The Robot That Plundered Pigweed - Blue River Technology
At the PrecisionAg® Vision Conference last October, Bill Schmarzo, renowned expert in data strategies for business and developer of the “Big Data MBA,” encouraged the crowd to embrace the power of focus. Pursue one thing that will bring value to customers and your business, and go all in. If it doesn’t work, fail fast, collect your learnings, and move on.
Blue River Technology, a five-year-old company that’s marrying robotics, computer vision, and deep learning to create solutions to difficult farming challenges, is among a rare group of agriculture-centric companies with that level of laser-like focus.
The first product out of the box, the LettuceBot, has been a big success. Lettuce thinning — the task of walking through a lettuce bed and removing young extraneous plants to achieve the ideal spacing between rows of healthy plants — has traditionally been a time- and labor-intensive task. Employing its breakthrough computer vision technology, the Lettuce Bot can scan and thin lettuce beds automatically — a brilliant solution to a monumental efficiency killer.
Ben Chostner, Vice President of Business Development at Blue River, says the company found one of its next big missions blowing up in the farm fields of the Southeast, Mid-South, and increasingly the Midwest: Glyphosate-resistant Palmer amaranth.
Palmer amaranth, A.K.A. pigweed.
“We were looking at critical needs of crops outside the specialty crop world, and noted that cotton farmers need help battling resistant pigweed,” says Chostner. “It is a big problem, and cotton has been at the epicenter of the herbicide resistance problem. We wanted to develop a tool to help them out.”
Blue River’s second-generation computer vision technology, which it calls “See & Spray,” seemed to fit the bill. “The idea behind See & Spray is a machine that can see and identify all the plants in a given area, and only apply herbicide to the weeds,” Chostner explains.
The system utilizes a regular color camera and computer vision techniques, including deep learning, to be able to identify all the plants and classify them as either cotton or weeds. “This is similar to technology in Facebook, when you upload a photo and it knows based other photos on your site who is in the photo,” says Chostner.
Last year was heavy-lifting time, as most of the components must be customized to make the system work and ruggedized to ensure durability in extreme field conditions. The development and testing were done in cotton fields in California but will be rolling out on pilot test farms in the heart of cotton country this season.
“We’re looking for two things — first to verify performance, and second to get feedback from farmers,” he says. Then, there will be the task of developing specifications for herbicide products. “We’ve never had the ability to apply herbicides directly to weeds before, so we’ll be figuring out what products and rates work best. We’re excited to get this figured out.”
Eventually, Chostner says the tool should be “flexible enough to use on any crop that is rotated with cotton — be it peanuts, soybeans, or corn.”
If all goes according to plan, Chostner is hopeful for limited commercial release in 2018.