A discussion with a neighbour, about the high cost and waste involved in spraying dock weeds in grassland, prompted a young engineer to design and build a spot sprayer to identify and treat this pernicious weed.
Colin Taylor of Taylor Technologies, based in Lanark, Scotland developed the system in his own time using all his personal savings, without any outside investment. Called the RUMEX system, it employs a combination of computer vision cameras, artificial intelligence software and individual nozzle control.
Spot spraying to save herbicide and money
“The difficulty with treating docks, my neighbour explained, is the selective herbicide is costly and he needs to cover the whole field to treat isolated weeds. The spray can also stunt grass growth,” explains Colin Taylor. “I was already interested in autonomous vehicles and had built my own drone as well as an agricultural robot. I immediately realised spot spraying, treating the individual weeds would save herbicide and money as well as increase grass yields.”
Colin started developing the system in 2017, while studying for his Agricultural Technology MSc research project at the Royal Agricultural College. At that time, computer-vision and artificial intelligence (AI), the technologies at the heart of the system, were becoming more widespread and easier to implement for other applications.
“I saw that we could take the technology that is notably used for facial recognition and self-driving cars, and re-purpose it to spot weeds,” says Colin. “Combining this with electronic control of a sprayer would mean that we could automatically spot-spray weeds, only activating the sprayer when the weeds are there.”
Identify a weed in a growing crop
While spot spraying is no longer completely new, it’s technology is still grappling with the challenge of ‘green on green’ recognition, which is required to identify a weed in a growing crop. “With docks, which are large broad leaf weeds, that’s not such a problem. They are considerably easier to spot in a field of grass, compared with looking for example, blackgrass in wheat,” Colin explains.
We needed the software to be able to recognise docks; and this involved training it using 20,000 different images of docks and grass
His computer vision system takes up to 30 pictures every second, which are analysed by the AI software – basically this sees whether it is seeing a weed or not a weed. “We needed the software to be able to recognise docks; and this involved training it using 20,000 different images of docks and grass. These included lots of different views, growth stages and in different situations, so it can always spot the weed,” he explains.
Hardi 12m wide, 600 litre sprayer
His prototype is fitted to a Hardi 12m wide, 600 litre sprayer, for which he has developed his own electronic control system for the solenoid-operated, individual nozzle control. He fits Hypro E80-30 nozzles because with the 80° these do not require an overlap to apply a full dose. With the current 30 frames/sec vision system it allows the sprayer to work comfortably at about 8km/hr, spotting and treating the docks in the field.
“I appreciate that is about half the speed of a conventional sprayer, but it’s producing enormous savings. A normal treatment, including the chemical and application, will cost about £70/ha. But if you look a grass field the docks will account for just 10% to 20% of ground cover,” he explains.
Huge savings from using less chemical
This results in huge savings from using less chemical – typically up to 75% to 85% of the cost. Also, because the grass yields increase, because it is not sprayed and does not have the weeds to compete against it. Moreover, the clover, which is particularly sensitive to herbicides, is not killed and therefore doesn’t need reseeding.
Field trials
Now, in its later prototype stages, Colin is conducting field trials, refining the system and looking to develop more ‘green on green’ AI software for other applications. He is also exploring working with an existing manufacturers and industry partners.
