at the Natural History Museum of London. Hier-Majumder’s team then runs these images through its simulation—bypassing the rock’s mineral layers and zooming deep into the inner pores—to determine the pore volume of each fragment. , which has helped the team “break the barrier of how much computational power we have in the lab,” Hier-Majumder says. “The machine we use in the lab has only 120 gigabytes of memory and 52 threads. That’s just not enough to process the massive amount of scanned images.
The grower uploads those images into Oracle Cloud Infrastructure, where AgroScout’s software uses machine-learning algorithms to identify diseases and pests. The software flags possible infestations and, after an agronomist confirms them, alerts the grower where they are in a field. The grower can then treat the problem quickly—or, conversely, avoid spraying chemicals where there isn’t a problem.
One of the hardest parts of AgroScout’s data-driven model involves training the algorithms—so that they know that a given picture is a leaf, of which kind of plant, whether it’s healthy and, if not, which pest or disease is present. Machine-learning algorithms are capable of learning those things, but the process requires showing the algorithm tens of thousands of images, which AgroScout did using Oracle Cloud Infrastructure and Oracle Cloud Native Services.
AgroScout is conducting its first real-world deployment by monitoring fields of potatoes in Washington state. The startup is also conducting a pilot test in South America in partnership with a large potato processor.