You’ve probably spent a lot of time keeping track of your plants and all the minor details, like the coloration of the leaves, in order to make sure they’re healthy — but for professional growers in greenhouses, this means keeping track of thousands of plants all at once.
That can get out of hand really quickly as it could involve just walking through a greenhouse with an iPad and checking off the health of each plant, and it means that a lot of things can fall through the cracks. And that’s not really a judgment on the professional abilities of the grower, but rather just the scale of the system that those growers have to deal with — and the lack of technology to support it, iUNU CEO Adam Greenberg said. So that’s why his startup, iUNU, is introducing a new system to try and help that.
iUNU’s Luna camera network works with a rail system with automated cameras that keep track of plants and how they are changing over time. So rather than having to do a daily crop walk, which could take hours, the growers can quickly have a set of cameras run across the plants and get a visual snapshot of those plants’ health. That information then feeds into a computer vision system on the company’s back-end, which applies machine learning to detect potential problems (like leaf discoloration) and helps those growers zero in on the areas that they actually need to address.
“While we’re doing something that seems really broad or really simple, it’s a derivative of highly granular HD sensors and repurposing facial recognition for plant recognition in a way that there’s a lot of highly millimeter level accuracy detail required to do it,” Greenberg said. “If you don’t have highly granular data sets that are higher quality than pretty much everyone out there, you’re not able to add more value. What’s really important from our internal perspective is that you have to deliver it in a way that’s a simple, easy-to-use decision support tool.”
It’s definitely a complicated computer vision problem, as what a plant looks like may change on a daily basis as they grow or bloom. And growers have to keep track of minute details, like minor discoloration. While it may just seem like a “Shazam for plants,” it’s involved creating a robust data set that’s able to detect those changes without showing up false negatives that could lead to a decrease in potential yield. At the scale of an industrial grower, any loss in yield means a meaningful loss of revenue.
The system is also designed to be modular, with growers over time being able to attach new kinds of tools along the rail beyond just cameras to the system. Greenberg likened it to the train sets you might have had as a kid, where you could end up with one unit for 3D modeling, one for cameras, and so on and so forth. The goal for iUNU is to be a network that any number of services where hardware or software products, like sensors or tracking systems, can just plug in and use to tap the data that it’s collecting. iUNU’s job is just to tell you if there are problems with your crops, and where they are, without requiring you to go out and do a daily crop walk.
“We’re there to be the grower and the owner’s best friend, we’re not gonna tell them how to do their jobs, we’re gonna help them do their jobs better,” Greenberg said. “When you have problems, they know how to fix them. We can tell you where your problems are. A lot of the other companies are trying to replace the grower. I don’t think that’s a good approach for the industry, I don’t think that’s the best way to work with the industry, to tell them you’re gonna replace them. It’s… that’s one fundamental difference. We play with the control systems, the sensor companies, the ERP companies, but we’re not competing with any of them.”
There are certainly other attempts to apply machine learning and computer vision to keeping track of plant health, like Prospera, which has also raised $7 million in venture financing. Greenberg said that by trying to take the platform approach, iUNU is going to be able to offer something more robust to growers in a way that individual products might not be able.