As global food demand continues to rise and myriad threats of climate change intensify, creating more sustainable agricultural practices has become increasingly essential. This is especially true in remote areas of the world, where advanced agricultural expertise is scarce and smallholder farmers with limited resources (both financial and material) cultivate an estimated 80% of farmland.
David Hughes, behavioral ecologist and PlantVillage team lead, realized that new technology needed to be part of the solution in closing the knowledge gap between experts and farmers working in their fields.
“Technology offers the opportunity to leapfrog that [gap] by making machines as intelligent or more intelligent than current human capacity,” he noted in his talk at the 2019 AI for Good Global Summit.
But a broad realization about new technology needed a specific direction when it came to fulfilling PlantVillage’s mission: to empower smallholder farmers in East Africa to grow more food with cheap, affordable technology and democratized access to knowledge.
Enter: mobile machine learning.
Table of contents:
Bridging the knowledge gap with on-device machine learning
In East Africa, crop disease and pest infestations have plagued smallholder farmers for decades. And while agricultural experts know how to manage and treat these plant diseases and infestations, the experts themselves are few and far between. In fact, there’s roughly 1 human expert for every 3,000 farmers in the region.
So Hughes and his team set out to use machine learning to bring this expertise and knowledge into the hands (and smartphones) of farmers themselves. The result was PlantVillage’s Android app Nuru (meaning “light” in Swahili), which serves as an expert assistant for farmers in the field, detecting diseases in Cassava, potato, and African maize — all in real-time and without internet access.
Hughes realized that embedding machine learning in the Nuru app was essential in making this vision a reality.
“We serve smallholder farmers in Africa who don’t have access to experts. So an AI system to diagnose problems is great but a non-starter if not offline.”
To enable this transformational user experience, Hughes and his team turned to Fritz for the ML expertise and developer tools necessary to create a custom on-device object detection TensorFlow Mobile model that could accurately detect very small parasites and signs of diseased leaves — both of which are difficult to see with the naked eye.
Additionally, Hughes and his team needed to understand how the model performed in the field in order to make improvements over time.
“[Fritz] was a very good system to check the performance of different TensorFlow models and highlight snags,” Hughes said. “There is such an enormous diversity of phones (flavors of Android, cameras, processors, etc.) so such a system is very useful.”
From AI expertise to human problem-solving
Since its release in 2018, Nuru has evolved into more than just an AI assistant in the field. It’s also become a resource empowering smallholder farmers to solve local challenges in innovative and unexpected ways.
One such challenge is finding “clean” seeds to plant in order to regenerate a healthy field. But with a shortage of such seeds coming from external sources, finding a workable solution required a different approach. Hughes told the story of a local farmer named Josephine who translated Nuru’s AI expertise into an innovative and unexpected solution.
“One day, in late December, she used [Nuru] 32 times in her field,” Hughes recounted in his talk at the AI for Good Summit. “Why was she using it so much?”
It turns out, Josephine wasn’t actually detecting infected Cassava plants. Instead, she was harvesting plant material that Nuru identified as healthy, which she used to replant an entire acre of Cassava — without a single infected plant.
Localized Knowledge and a Global Community
With instant access to Nuru’s real-time, expert predictions that, Hughes notes, are up to twice as reliable as human expertise, knowledge can quickly become localized and democratized. Given the success of Nuru in accomplishing this, Hughes envisions a bright future for PlantVillage.
To bring this kind of knowledge and expertise to farmers around the world, organizations will need to follow PlantVillage’s lead and increasingly rely on solutions like on-device machine learning.