John Deere's quest to solve agricultures deep-learning problems-[Spectral Lines]

Abstract

What's the world's hardest machine-learning challenge? Autonomous vehicles? Robots that can fall over and get back up? Cancer detection? ¶ Julian Sanchez believes it's agriculture. ¶ He may be a little biased. Sanchez is the director of precision agriculture for John Deere, and he's in charge of adding intelligence to traditional farm vehicles. I met with Sanchez and Alexey Rostapshov, head of digital innovation at John Deere Labs, in San Francisco, where John Deere launched the spin-off in 2017 to take advantage of Silicon Valley's tech expertise. ¶ Sanchez believes agriculture is the biggest challenge for artificial intelligence because it's not just about driving tractors around, although autonomous driving is certainly part of the mix. According to Sanchez, the more complex problems revolve around issues such as crop classification. John Deere would like to create an AI system that allows farmers to know, for example, whether a grain being harvested is high quality or low quality. The many differences between grain types, and between grains grown under different conditions, make this a tough task for machine learning. ¶ Sanchez uses corn as an example. To build a deep-learning algorithm to analyze corn kernel quality, you'd feed it millions of pictures of kernels. Kernels harvested in central Illinois might have one color. But kernels of the same hybrid from a different farm 5 miles away might look slightly different. Now imagine solving that challenge for dozens of grain varieties-some of which, like canola, are nearly microscopic.