PlantVillage Nuru is a deep-learning object-detection model that runs offline on a smartphone to diagnose crop diseases in the field, built as a public good by Penn State University, the UN Food and Agriculture Organization, the International Institute of Tropical Agriculture, and partners. A 2020 evaluation in Frontiers in Plant Science tested its accuracy on the viral diseases of cassava, a staple crop across Africa.
The cassava model was trained on 2,756 annotated leaf images. When examining six leaves per plant, Nuru identified cassava mosaic disease (CMD) at 93% accuracy and cassava green mite damage at 93%, with cassava brown streak disease (CBSD) lower at 73%. Critically, Nuru outperformed the people it was meant to help: untrained agricultural extension workers scored 40 to 58 percent and farmers 18 to 31 percent on the same task, and the 2020 version’s overall accuracy across different phones exceeded that of trained extension officers.
Because the model runs entirely on the device with no internet connection, it works in rural areas with no network coverage - the exact places where plant pathologists are scarce and a misdiagnosed crop disease can threaten a family’s food supply.