Imagine flying a drone equipped with sprayers across a soybean field, targeting only weeds that eluded the first round of spraying. Or imagine an implement that can kill specific weeds with laser technology.

These technologies are not outlandish, and according to Dr. Muthu Bagavathiannan of Texas A&M University, they’re already at work on farms and will be increasingly utilized in many facets of agriculture. Bagavathiannan is a professor of weed ecology and management and discussed precision weeding during a University of Vermont Extension Precision Dairy Farming: From Hooves to Harvest meeting.

Precision weeding (and precision agriculture in general) relies on teaching computers how to identify weeds and other plants, a process called machine learning. These technologies and algorithms enable systems to identify patterns, make decisions and improve themselves through experience and data.

For example, to teach a computer what a dandelion is, many images of dandelions in different variations and stages of growth must be shown to the computer with proper labeling.

“In order for the model to be very robust and accurate that it can be applied across multiple field environments, we need huge datasets that capture the variability that is present within the species,” Bagavathiannan said.

He is currently working as part of a team to generate an open-source national weed and crop image repository that researchers and commercial enterprises can access.

The next step is to take this machine learning into the field. For instance, drones can be equipped with sensors that can detect color, reflection differences or other vegetation features. Sensors such as Light Detection and Ranging (LiDAR) or depth sensors can be used to map the 3-D structure of the plant canopy, providing detailed information regarding plant height, size/biomass and other characteristics. This information can be used for making site-specific management decisions and applications.

Some kind of software must then be used to translate the information from the flyover into a map of the weeds in the field. The idea is that eventually drones or ground-driven sprayers will be able to use this technology in conjunction with GPS to target specific weeds in a crop rather than broadcast spraying an entire crop.

Dr. Muthu Bagavathiannan prepares one of the drones used in his research. Photo courtesy of Muthu Bagavathiannan

The potential also exists to accommodate varying growth stages of weeds – spraying more on larger weeds and less on small seedlings, therefore reducing overall application rates.

“You can actually select the specific management tool option on those specific groups of plants. You could even do it on an individual level. That’s the power that machine learning is going to allow us to do,” Bagavathiannan said.

It doesn’t always take a drone flyover, however, to make weed management decisions. Some farming equipment equipped with similar sensing and software can make field decisions on-the-go, such as John Deere’s trademarked See & Spray system. Currently, though, these systems – “green-on-green,” Bagavathiannan called them – can only distinguish differences between the crop and all other vegetation. The sprayer will, for instance, spray everything other than the soybean crop.

It will be a matter of time before these technologies offer species- or individual plant-level recognition and management. Nonetheless, combining drone images with ground sprayers can offer additional information for improved field-scale decision-making.

With site-specific herbicide applications, there is also the potential to reduce the herbicide use rates on an acre basis, thereby reducing the environmental load of herbicides. This will hopefully allow for more herbicides registered for field use that are otherwise not possible due to high environmental risk.

Another example of a precision weeding technology is a weed chipper/hoe being tested in Australia. It has been tested on fallow ground to kill individual weeds rather than tilling the entire field. Bagavathiannan referred to this as “brown-on-green,” meaning the implement has been designed to chop off the green weeds in the brown soil.

Similar is a precision weeding laser system (used primarily by high-value organic crop producers) invented by California-based Carbon Robotics. High resolution cameras scan the field, crops and weeds in real time. This autonomous system uses see-and-burn technology – high powered lasers target thermal energy at each weed’s meristem.

“One of the limitations with the laser weeding, at least what we’ve observed in our prototype testing in Texas, is that the laser technology is the most effective on the smallest weeds, especially for grasses. You have to get them when they are less than one inch,” Bagavathiannan said.

Another limitation is the high price tag, although Bagavathiannan predicts that as the use of this technology is further developed and competition increases, costs will go down.

This precision technology is also relevant in other aspects of agriculture. Drones and machine learning can be used to map and estimate weed seed production in a crop field. Knowledge of how much weed seed is potentially going back into the soil can alert growers to implement management that targets weed seeds.

Moreover, with the knowledge of field areas where weed seeds are produced, they may be able to make variable rate pre-emergence herbicide applications using these maps. Additionally, drones can be used to identify herbicide injury on crops in order to evaluate the potential crop yield loss. The technology can also be used to measure the amount of biomass produced by a cover crop or how much forage is available for livestock.

According to Bagavathiannan, global trends in weed issues with weed adaptation and herbicide resistance necessitate the use of this technology. He is amazed at how much the field has grown in the past five years and does not expect the rate of research, development and deployment to slow down.

Bagavathiannan does have a word of caution, however: “We shouldn’t forget that the weeds are always smart, and they will adapt over time to whatever management that we do. We have to always keep in mind that weeds can and will adapt to any management tool if we rely on it heavily without sufficient diversification.”

by Sonja Heyck-Merlin