NDSU’s Weedbot targets green-on-green weed control at seedling stage

Automation in agriculture should solve at least two issues: replace labour hours and improve the job being done. Add saving money into the mix and the technology is sure to take off.

But all technology has to start somewhere, and that somewhere can be universities. Kelvin Heppner, field editor for RealAgriculture, traveled to the Big Iron farm show at Fargo North, Dakota, to learn about some of the cool things happening at North Dakota State University (NDSU).

Dr. Rex Sun, assistant professor at agricultural and biosystems engineering at NDSU, is part of the team that has built the Weedbot, a unit that uses artificial intelligence and robotics technologies for precision agriculture applications.

The Weedbot offers site-specific targeted weed control using cameras and sensors to identify and spray weeds when they are very small.

Sun says targeting small weeds is something farmers want, based on feedback, but just-germinated weeds can look a lot like a germinating crop, making the job more complex. This green-on-green spraying is the goal, to make the process more precise, but also to save overall herbicide used, saving money and decreasing potential impacts on the environment.

“I really believe artificial intelligence and robotic technology play an important role in the future farming business. And we are glad that we are part of that, and that we are trying to move forward step by step,” Sun says.

The Weedbot structure itself is cost-effective, Sun says — the most expensive part is the camera sensors and the most complex part is the actual AI algorithms. NDSU has been working on building the background information required to “teach” the AI to identify crop from weed plant for the last five years.

Sun says they’ve already amassed over five million images of crops and weeds to build a robust database to drive the Weedbot’s selection algorithms. In the future, Sun says the plan is to release that algorithm to farmers. Sun’s team is working to include key weeds in the database, such as ragweed,  kochia, waterhemp, red root pigweed, and Palmer amaranth — several of which can be hard to tell apart.