Enabling rapid diagnosis of crop disease in fields based on crop images, using computational techniques in machine learning implemented on a mobile phone.
On the left we have 13 example plants, with their disease levels shown in the dark boxes. The contour lines indicate our calculation of the mean disease level. The histograms on the right show our predictions at what the distribution of the five disease levels will be at positions A, B and C. Based on this model, we can use our uncertainty at different places on the map to dynamically change the survey schedule (divert the survey teams to the most informative areas) and to update prices for data collection by extension workers (increase prices for areas where data is more useful). Ultimately such maps are used to plan the way in which limited resources can best be used to limit the spread of disease, for example by starting training programs for farmers in high risk areas, or calculating the best places to take healthy planting material to replace the crops in the most affected areas.