Born of research in the Amazon forest, the Plantix mobile app is helping farmers on three continents quickly identify plant diseases using artificial intelligence.
For several years in the Brazilian rain forest, a team of young German researchers studied the emission and mitigation of greenhouse gases due to changing land use. The team’s analysis was yielding new knowledge, but the farmers they worked with weren’t interested in those findings. They wanted to know how to treat crops being ravaged by pathogens.
“They couldn’t understand why we can estimate the carbon stock of their soil, but we couldn’t give them an idea of how to treat damaged plants in an appropriate way,” said Robert Strey, one of the researchers.
This realization led the team, led by Strey and his wife, Simone, to shift its attention to the more pressing problem of crop health. And so PEAT Technology came to be, with Simone as CEO and Robert as CTO.
Today, farmers in Germany, Brazil and India use Plantix to upload photos of diseased crops. The images are part of a huge and growing crowdsourced database that is helping farmers to identify, treat and prevent crop diseases.
Magic Behind the App
The magic of PEAT happens once it has received the photos and runs them through its image recognition software — which grows more powerful with each new crop disease the company logs. The app has been downloaded 50,000 times in the last year, resulting in 100,000 image uploads into PEAT’s dataset.
Already, PEAT can identify more than 60 plant pests and pathogens with more than 90 percent accuracy. Those numbers figure to rise as the database grows.
PEAT also hopes to start implementing its software on drones, agricultural equipment and greenhouses next year so that farmers can automate the process and respond to crop illness quicker. For now, it’s collecting photos and learning how to accurately identify as many crop diseases as possible, with a focus on major crops like corn and wheat that are planted by large-scale farmers around the world.
In exchange for supplying photos, users of Plantix receive actionable information, including the scientific names of diseases, along with triggers, symptoms, treatment options, preventative measures and the like.
GPUs Speed Up Network Training
PEAT is using two NVIDIA TITAN X GPUs, in scalable link interface mode, in combination with the CUDA parallel computing platform and programming model to not only train the neural network models behind Plantix, but also to analyze incoming photos, a process known as inference.
PEAT can calculate a new neural network in 7-10 days on the GPUs. If it used CPUs, “we would need about 30-40 weeks and this is clearly not a timeframe suitable for a startup,” said Strey.
Strey said the team’s current training efforts are focused on identifying characteristic patterns of plant diseases — such as spots on fruit or a mosaic-like discoloration on leaves — so that it can suggest actions even when it can’t definitively identify a specific plant or disease.
Eventually, PEAT hopes to refine its technology so that it can solve plant-disease issues for all types of farmers, not just those growing mass-produced crops.