Weed AI

Weed AI

Tim Hammerich
Tim Hammerich
News Reporter
It’s time for your Farm of the Future Report. I’m Tim Hammerich.

Labor remains a challenge for many farming operations, which means automating tasks can become an appealing option. But for agricultural machines to get smarter, they need to be trained. For example, to train an algorithm what is a weed and what is a crop requires an database of images. Guy Coleman and other researchers at the University of Syndey are trying to create an open source data set for weeds called Weed AI.

Coleman… “It goes back to the difficulty of capturing weeds and all of their diversity of appearances and lighting conditions, soil backgrounds, it's a real difficult part of the detection process. The current algorithm architectures, they're all based on these large open source data sets for everything except weeds. But if we can get these large data sets of weeds together, then we can potentially start developing architectures that are specific to weeds and they might actually do better. And start exploring these more research computer science-y questions that weren't previously possible with the data that was available.”

So what does it take for this vision to become a reality? Continued research and a whole bunch of images.

Coleman… “So probably that two-pronged approach: it's about getting people to start contributing and finding weeds in a whole variety of scenarios they often present. And also about opening up these new architecture research development questions as well.”

Coleman hopes this project will lay the groundwork for future agricultural automation.

Previous ReportOpen Weed Locator - Part Two
Next ReportDairy Digesters