Crop diseases, a major cause of famine, have always been diagnosed by visual inspection, though scientists today also use microscopes and DNA sequencing. But the first line of defense is still the keen eyes of farmers around the world, many of whom do not have access to advanced diagnostics and treatment advice.
To address this problem, scientists from Penn State and EPFL, the Swiss Federal Institute of Technology in Lausanne, Switzerland, are releasing 50,000 open-access images of infected and healthy crops. These images will allow machine-learning experts to develop algorithms that automatically diagnose a crop disease. The tool then will be put into the hands of farmers — in the form of a smartphone app.
By providing all these images with open access, we are challenging the global community in two ways,” said David Hughes, assistant professor of entomology and biology, Penn State. “We are encouraging the crop-health community to share their images of diseased plants, and we are encouraging the machine-learning community to help develop accurate algorithms.
The idea behind releasing the open-source image database is to provide software developers with the “raw materials” to build machine-learning algorithms, according to Marcel Salathé, associate professor at EPFL and formerly a faculty member at Penn State. He explained that machine learning is a computational way of detecting patterns in a given data set to make inferences in another, similar data set. Fueled by breakthroughs in algorithm development, cheap computing and cheap storage of very large data sets, the results of machine learning have permeated our everyday digital experiences, from face-recognition in photos and videos, to recommender systems in online shops.
The next step,” Salathé said, “is to combine the enormous expertise in data science around the globe with our open-access data sets in the form of online competitions to develop the best algorithms to diagnose plant diseases. In the very near future, we’ll launch the first online competition based on this growing data set that we make available today.”
What Hughes and Salathé want to do is to apply these techniques developed by computer scientists to the problem of recognizing and diagnosing crop diseases. Algorithms with sufficient accuracy will be incorporated into smartphone apps, allowing farmers to snap pictures of their infected crops and get instant diagnosis and treatment advice.
For the two scientists, this is a natural evolution of their website, PlantVillage, possibly the world’s largest free library of science-based knowledge on plant diseases. The still-growing site covers 154 types of crops and more than 1,800 diseases and now houses the new image database.