The Ag Innovation Showcase offers a platform to innovators across the agricultural value chain. Smartvision Works is improving the sorting process, helping more crops get from farm to fork. BreAnn Washburn explains how her machine learning technique is developing vision software that will help robots tell the difference between mature and unripe fruits and vegetables. Registration is now open for the Ag Innovation Showcase 2017: visit www.agshowcase.com.

BreAnn Washburn: We use a machine learning technique that allows us to develop vision software in a very quick way, but also in a way that’s highly accurate, which allows us to go into niche agriculture markets, but also markets that are a little bit more complex.

So one of the crops that I talked about in my presentation yesterday was the date. The date is a small, niche agricultural market, but it’s also a complicated market. Dates actually grow to a certain point and then they start to shrivel as they mature. And that maturity is hard to gauge. So, traditionally it’s been done by hand: once the fruit is harvested, it’s sorted by hand. That works, obviously, but it is becoming increasingly expensive for growers to be using hand labour. So they’re switching to mechanical labour.

So they need something that will help them do vision assisted sorting, because it is a niche product and because it’s complex, they need a sophisticated system. In addition to shrivelling, as it grows, dates also rot from the inside out. It makes it really hard to tell a rotten fruit. The ladies who are doing the sorting, they can tell instantly, they can tell the difference between a rotten fruit and a good fruit. But I can’t. And if I ask them what the difference is, they can’t articulate it to me, they can’t describe it to me. Which means that I can’t write an algorithm for it, I can’t create a vision system for something I can’t articulate. But using a machine learning technique, we can.

As we feed the machine learning technique lots of images of the good fruit and the bad fruit, it actually iterates an algorithm of its own. So we were able to develop a vision system that had the same results or better than hand sorting.

So we use the same technique, whether we’re looking at avocados or garlic or dates, so we’re not reinventing the wheel every time, but we do get a highly custom system every time.