Identifying plant diseases can be a challenge, but researchers from Pennsylvania State University and the Swiss Federal Institute of Technology in Lausanne (EPFL) are developing an app for that.
The PlantVillage app is expected to be available in early 2017 and will be capable of recognizing what disease a plant has based on a mobile phone photo.
The database currently holds 150,000 photos of diseased plants. The team was able to create a computational neural network that allows the system to process the plant and determine what disease it is infected with.
The algorithm enables PlantVillage to correctly identify the disease 98.21 percent of the time when using high-quality images, according to a study conducted when the database had 50,000 images.
The accuracy does drop to 31.69 percent when working with lower quality photographs from outside websites. The researchers have made the database of images available to anyone so other teams and companies can work on developing a better algorithm.
According to EPFL digital epidemiologist Marcel Salathe, he founded PlantVillage with Penn State entomologist David Hughes when he noticed a separation of information about plants from the people who need it.
“That to us seemed incredibly backwards,” Salathe told MIT Technology Review. “That was the point of the Internet.”
Right now the database focuses on plant diseases that affect 30 of the most important crops, but the team plans to grow the database to 3 million photographs.
PlantVillage varies from other current plant disease identifier apps available. The Plant Doctor app from Purdue has users match the symptoms of their plants with other pictures. The University of Hawaii’s Plant Doctor app sends images of the sick plant to an expert, who then diagnoses the problem.