Zapping Weeds One at a Time Down on the Farm

Farming, which is almost as old as dirt, is undergoing a significant tech makeover. Tech ideas are flourishing, including robots, drones, the Internet of Things (IoT) and container farms that can be stacked on top of each other to make vertical farms. But for farmers whose last shot at technology was taking to the air with their crop dusters, Blue River Technology has a better solution. The firm would like to make weed control more precise—and better for the environment.

At NVIDIA’s GTC 2017, Blue River showed a model of a trailer that is loaded with NVIDIA GPUs to help it search and destroy weeds—one at a time—as it is towed behind a tractor.

Using lasers?

No, laughed the Blue River spokesperson, who has had to explain to hundreds of software geeks what a giant green John Deere tractor is doing amidst all the computer hardware on the show floor of NVIDIA’s annual programmers conference, which was held in the San Jose Convention Center, the epicenter of Silicon Valley.

It doesn’t pay to cover entire fields with herbicide. With the Blue River’s “See & Spray Machines,” you can target specific plants. You can save money not only on the amount of chemicals used, but also by growing cheaper crops with plants that are not genetically modified to be herbicide resistant.

How It Works

Figure 1. Computer vision identifies each individual plan. Machine learning decides how to treat each plant. Robotics will aim and dispense the herbicide. (Image courtesy of Blue River Technology.)

The trailer tows an array of spray heads, herbicide tanks and banks of NVIDIA graphic cards that power the synthetic vision. The system can currently only identify a few weed types that are found in cotton and lettuce farms. As the trailer moves down the field, spray heads fire at will at offending plants.

“It's a very green approach,” said the Blue River Technology representative. The amount of over spray is very small compared to what it would be if the herbicide were covering the whole field.

The computational heart of the system is the NVIDIA Jetson, a computer on a board, which uses chips originally designed for graphic processing as parallel processors for crunching numbers. NVIDIA maintains that GPUs are considerably better at handling the massive processing needs of machine learning systems than CPUs. You can imagine the processing that is necessary as the chips try to keep up with the images of plants, coming at the rate of several times per second, and match leaves against known shapes, as the trailer moves down the rows of plants—and has to do so before the spray head is out of range.

The See &Spray system is set for production later this year.

Next year, we hope to see the ability to zap bugs. But I think that would require lasers.

Figure 2. Farming the Silicon Valley. Blue River brings its synthetic vision, machine-learning, GPU-powered weed zapper to GTC 2017.

Lettuce Help

For venture capitalists, whose hands may never have held a hoe and whose idea of green is how much of it they can get into their pockets, farming may represent an almost infinite potential for improvement. No doubt bringing their fresh ideas into an industry that has not benefited from the superior intellect of the Silicon Valley is bound to produce a more bountiful harvest, right?

But for Blue River Technology, being in Sunnyvale, Calif., has been a double-edged sword. It’s great for raising capital (Blue River has received a total of $30.5 million in funding in four rounds, the most recent a $17 million round in December 2015), but to get the software developers it needs, it has had to compete with every other tech firm nearby.

This is a great place to recruit, said their spokesperson, as she handed me her business card. She is from the HR department.