These Scientists Are Training Computers to Help Farmers Save Their Crops

Getty

Want to know why your cucumbers are producing flowers but no fruit? It could be that your flowers are only male. Can’t for the life of you figure out what those white spots on your squash plants are? A new app called PlantVillage may be able to help.

PlantVillage is essentially a giant Q&A forum for folks around the world to pose questions about their farming, from tiny backyard gardens to sprawling large-scale agriculture. Snap a photo of your problem plant, upload it, and get an answer from either a botanist or an ordinary gardener who’s been through the same problem before.

As a supplement to human answers to farmers’ questions, PlantVillage’s founders, Penn State epidemiologists David Hughes and Marcel Salathé, are working on using machine learning—in which computers teach themselves new ideas rather than having programmers beat it into them—to eventually get the app to do things like automatically recognize a picture as a certain species of weed. It’s an ambitious project, to say the least, but according to Hughes, after almost two years the site counts 500,000 unique visitors, 35 percent of whom are from developing countries—the people who so badly need this information to not only make money but to feed themselves.

“So we have a question from Cameroon on oil palms,” Hughes said. “It’s a farmer who’s literally having to travel out of his farm for quite a distance to email us, then upload some images, then get back to his farm and figure out what his issue is. So he’s really remote.”

But while that farmer in Cameroon can get to a computer to pose his question and get an answer, 4.4 billion people on this planet still have no access to the internet. It’s in these unconnected populations that PlantVillage would have the most impact, yet until the farmers among them get cheap cellphones, they’re left wanting for information. Connectivity rates, though, have nowhere to go but up, especially with Facebook and Google gunning to bring the internet to developing countries—out of the goodness of their hearts, they claim, but also necessarily to open up new markets.

Screenshot: PlantVillage

In the meantime, Hughes foresees PlantVillage—which is funded by Penn State’s Huck Institutes of Life Sciences—creating a new type of entrepreneur in communities where maybe just a handful of people have mobile phones to access the site. He says that 40 percent of users are on handsets, but “not everybody has a phone, and not everybody necessarily needs one. But if one person has a good understanding of the web and the ability to engage with researchers around the world, he or she can ask and answer questions on behalf of others, and even charge a fee for that. So there might be go-betweens.” But such a market of information could also of course lead to exploitation of farmers desperate for cures for their crops.

Yet a similar, albeit governmental, system worked in the past. During the Irish Potato Famine, Hughes notes, the government dispatched agricultural experts into the countryside to advise farmers. The idea spread throughout Europe and eventually the U.S., where it exists today as the Cooperative Extension System, in which “people with science-based knowledge go out into the field sharing that information,” said Hughes. “That extension system was exported from the U.S. to parts of the developing world, like India and Mexico in the 1960s, ushering in the green revolution. Yet that didn’t reach Africa. But now Africa has hundreds of millions of mobile phones. Our vision is an extension of extension, imagining a mobile Green Revolution.”

Extension programs are moving from this one-way flow of information to a more egalitarian approach, where farmers don’t just get talked at, but actively share information, according to Rikin Gandhi. His non-profit, Digital Green, takes a more on-the-ground tactic, helping farmers create and share videos about their techniques. He cautions, though, against leaning too hard on the technology when disseminating information.

“PlantVillage and Digital Green are examples of the sorts of platforms that can support this,” he said. “But I’d just emphasize that there are limits that technology can do in and of itself, and that we’ve found it critical to actively engage organizations and individuals who work with farming communities to catalyze this exchange.”

To that end, Hughes says the PlantVillage team has been reaching out to various organizations, including the Global Knowledge Initiative, whom they’re working with to set up farmer cooperatives in Rwanda.

Rise of the Machines

It will be the technology, though, that could one day do much of the heavy lifting for PlantVillage. Hughes’ partner Marcel Salathé, who’s overseeing the more technical aspects of the project, hopes to have a photo-recognition feature in beta testing sometime next year. The idea is for users to upload images of their diseased plants to a database, where machine-learning algorithms will identify what’s gone wrong.

“That’s pretty futuristic,” Salathé said. “But on the other hand, I don’t think we’re that far away. I think a lot of plant ecologists would probably disagree with me, and I would have to say it also totally depends on which disease we’re talking about. Some diseases will be incredibly hard to diagnose like that. Others will be incredibly easy.”

Black rot on a cocoa pod. PlantVillage’s founders hope to one day use machine learning to automatically identify pictures of such diseases. PlantVillage

For those algorithms to work, they’ll need to be fed the knowledge in the first place—and accurate knowledge at that. Bad information can lead to a bad diagnosis that could only make matters worse for farmers. So Hughes and Salathé have been running a campaign to pull in experts from the U.S. Land Grant System—colleges that have been accredited by the feds to teach agriculture—to help supply the information.

“One of the things we want to do,” said Hughes, “is take certain crops and then have crowdsourcing—but not any old crowd—actually from crowds of researchers in universities or in laboratories, as well as growers of course and ask, ‘Is this disease X or Y?’ ‘Is that what you think it is?’ And then allow for machine-learning algorithms based on this highly specific crowd.”

They’re also looking to agricultural giants for help. According to Hughes, they’ve been reaching out to companies like Monsanto and Syngenta, who have thousands of scientists with extensive knowledge of plant diseases. He wants them to be a part of the conversation as well. “After all, we all have to feed the growing population, and in a few years there will be 2 billion more people at our table.”

So if they can get the knowledge, and they can get enough photos of plant diseases, which are often trapped behind paywalls in scientific journals and elsewhere, just maybe PlantVillage can spread the information the world needs to help feed itself.

“It’s amazing that there’s more pictures on the web of Kim Kardashian’s bum than there are of the diseases that cripple the food supply of millions of people,” Hughes said. “And that’s just horrendous.”