Sep 13, 2016
Japanese farmer uses Google AI to sort cucumbers
Google has an interesting article describing how a cucumber farm in Japan employed Google’s open-source Tensorflow machine learning technology to help with sorting the cucumbers, a difficult and mundane task that usually takes months to learn.
It all began about a year ago, when a former systems engineer Makoto Koike started helping out with work at his parents’ cucumber farm. He was surprised at how much time and effort it takes to sort the cucumbers by size, shape, color, freshness and other attributes — which is the work his mother does all by herself, and that can take up to eight hours per day at peak harvesting season.
Just as an idea, here’s an example of cucumbers sorted into nine different classes (with premium grade at the top, lowest quality at the bottom).
There exist automatic sorting machines but they’re limited in terms of cost and performance. Smaller farms tend to do this kind of work manually.
In short, the machine Koike built uses Raspberry Pi 3 to take pictures of the cucumbers, which are then forwarded to the TensorFlow artificial intelligence network for classification. An Arduino Micro takes that info and performs the actual sorting.
If you’re technically inclined, below is a diagram of Koike’s system but you can also read about it in more detail.
According to Koike, the system he built isn’t completely accurate (when training the sorting AI, the machine recognized the images with a 70% success rate), but it hints at the possibilities in the future.