Artificial intelligence in all of its forms is a truly wondrous achievement of modern technology. Given the right stimuli, an AI-based machine can learn just about anything, and a man by the name of Makoto Koike found one of the coolest ways to prove that notion. In a move that may make you pause for a second, then whisper to yourself, “That’s actually pretty brilliant”, Makoto devised a machine that uses AI, with a backend rooted in Google’s TensorFlow system, to take the edge off of one of the biggest pain points of growing cucumbers; sorting the different types and qualities.
When it comes to cucumbers, visual qualities hold a surprising amount of information about the vegetable inside the shell. Cucumbers that are extremely high quality tend to be straight, long, deeply and vividly colored, and quite prickly. From there, a cucumber can have a number of lesser qualities that can determine its taste, texture, and other factors. Each of these factors has a visual tell. The issue, Makoto quickly found upon taking on work at his parents’ cucumber farm, was that sorting the cucumbers manually was extremely tedious and time-consuming. The lack of clear industry standards make things that much more confusing; every farm has their own standards of quality. That meant that if Makoto wanted to make things easier on himself and his family, he would have to take matters into his own hands.
Thanks to Google Cloud Platform, he was able to do just that. Makoto created a deep learning AI using Google’s TensorFlow. Utilizing the same technology for variable analysis used by DeepMind’s AlphaGO, the AI was given a minimum of information about the various types of cucumbers and was eventually able to sort them out with an extremely high success rate. Since sorting cucumbers is a feat that takes humans months to learn, having a scalable AI with the right know-how is quite handy. Makoto’s system is deceptively simple; while the backend requires all the normal trappings of running a TensorFlow application, the client is run through only two pieces of low-power hardware. A Raspberry Pi 3-based computer uploads cucumber data to the cloud and pulls down the cloud’s sorting advice. From there, the commands are fed to an Arduino Micro that’s set up to control a conveyor belt system that picks up the cucumbers, asks the cloud about them, then drops them into the correct bin.
