Intelligent sprayer targets individual weeds

Spot spraying to target individual weeds is no longer a distant research ambition but a reality. Commercial systems are becoming available for conventional sprayers as well as a future generation of autonomous weed controlling robots.

GPS to map weed species

Much of the pioneering work in this area was carried out in the 1990s at Britain’s Silsoe Research Institute. It investigated the practicalities of using GPS to map weed species that typically grow in clumps or patches.

Research into this ‘patch spraying’ approach treating 2x4m ‘cells’ with different herbicide doses ended with the institute’s closure in 2006. But the knowledge gained was maintained through the private technology company formed by ex-Silsoe researchers Nick Tillett and Tony Hague, who today develop and manufacture computer vision technology, mainly for weed control. “Inter-row guidance of cultivators between crop rows where a farmer wishes to reduce herbicide input is the most common application for our technology,” says Dr Tillett.

Video image analysis

However, the latest application for Tillett & Hague Technology’s video image analysis system is the Robocrop Spot Sprayer from Zürn Garford. It is designed for weed control situations such as rogue potatoes growing amongst carrots, parsnips, onions and leeks.

_{Text continues underneath image}

Individual weeds are now in the sights of precision targeting systems for row-crop and broadacre sprayers – this case Romex species in grassland. – Photo: Agrifac

Unlike the image analysis used for inter-row and within-row hoeing, which detects crop rows for the former and also individual plants within the row for the latter, Spot Sprayer imaging searches for clumps of vegetation that do not conform to the crop row characteristic. These objects are tracked as they pass down through the camera field of vision, to trigger the relevant nozzles on a spray bar to deliver a measured quantity of herbicide.

Select the minimum weed clump size

Through the Robocrop user interface, operators can select the minimum weed clump size and a percentage area of the plant size as the target; together with an adjustable nozzle on/off period starting at just 30 milliseconds. A minimum target area of only 40mm (1.5 inch) is possible to minimise damage to non-target plants.

Depending on weed density, the quantity of herbicide used is reckoned to be commonly less than 2% of that delivered by an overall spray treatment. So there is considerable potential for reducing costs and minimising the environment impact.

_{Text continues underneath image}

The Bilberry technology used by Agrifac for its AiCPlus system uses RGB colour cameras to distinguish weeds in crops by plant shape, structure and contrast. – Photo: Agrifac

Specialised nozzle tips

The use of specialised nozzle tips to ensure accurate delivery is highlighted by the UK’s Horticulture Development Company, which part-funded the technology as part of the Agriculture & Horticulture Development Board’s R&D programme. These nozzles generate very large droplets to produce a directionally stable spray, at low pressure to minimise the risk of splashing upon impact.

NIR sensors

The WeedSeeker sprayer activation system, developed by Trimble subsidiary NTech Industries, uses near-infrared reflectance sensors to detect weeds for nozzles’ on/off switching. The WEEDit system developed from fundamental research in the 1990s at Wageningen University in the Netherlands does the same.

Trimble’s WeedSeeker sensors can be mounted up to 30 at a time on broadacre spray booms up to 45m (150ft) wide, with settings for up to 16km/h (10mph) or 32km/h (20mph) working speeds, operating day or night.

_{Text continues underneath image}

Kuhn-Blanchard is adopting Carbon Bee AgTech’s AQiT-Sensor hyperspectral imagery technology for weed location and identification. – Photo: Kuhn

Rometron’s WEEDit Ag system uses sensors scanning a 1m band of soil 40,000 times per second while emitting red light. This is absorbed by chlorophyll in the foliage of plants, resulting in near infrared (NIR) light being reflected back to the sensors. When a weed shows up, the system recognises an increase in the ratio of near-infrared to red light and activates a solenoid valve to deliver a pre-mixed herbicide spray. It can be fitted to new sprayers, or as a retro-fit installation on booms up to 36m (118ft) wide.

In Europe, the Rometron technology has been adopted by Amazone as the AmaSpot system. It is available on 4,200- and 5,200-litre UX trailed sprayers equipped with a purpose-made 24m (78ft) boom and Agrotop nozzles with pulse width frequency modulation (PWFM) output control. This approach enables each nozzle to deliver from 100% to 30% of the application rate – or to be switched off altogether. This suits a combined approach of weed-specific on-off spraying and variable rate blanket treatment, as appropriate for weed type and population density.

_{Text continues underneath images}

Weed species identification

An alternative optical weed location system developed by US company Blue River Technology, now a John Deere company, was inspired by advances in facial recognition. Ben Chostner of Blue River explains: “We saw an opportunity to take cameras, computers and artificial intelligence to allow ag machines to see every plant in a field and give farmers the ultimate flexible tool to spray very precisely, whether it be herbicide only to the weeds or fertiliser or fungicide directly on each crop plant.”

Deep learning algorithms

The system employs deep learning algorithms to enable recognition of not only the presence but also the identity of different weeds within crops. It does not rely on spacing or colour to identify them. Instead it recognises differences between plant shapes and structures. A second set of cameras automatically checks the machine’s work as it operates.

Pigweed (Amaranthus species) is among the first targets. This is a summer annual resistant to most common herbicides that growers aim to control in cotton and other row crops. “The first time our system saw a pigweed it didn’t know what kind of plant it was. But we taught it by showing it tens of thousands of images of pigweed – and now it’s an expert,” says Ben Chostner. “Spraying apart, there are opportunities to inform growers about how many weeds the system has seen and even what kind of weeds exist in the field, so they can tailor their herbicide programme towards those weeds.”

RGB camera detection

French tech company Bilberry has also developed a camera-based ‘intelligent’ weed detecting, identification and sprayer control system. It will treat weeds within crops. Guillaume Jourdain, co-founder and CEO, says: “The deep learning software and constant gathering of new field data from farms around the world, enables us to continuously improve the precision of our system and add new types of weed to the ever growing library.”

_{Text continues underneath image}

With Rometron weed-detection technology and pulse width frequency modulation nozzle control, the Amazone UX AmaSpot trailed sprayers can deliver 100% to 30% application rate or be switched off altogether. – Photo: Rometron

Earlier this year, Bilberry started working with sprayer maker Berthoud of France to exploit the Intelligent Spot Spraying System. But Dutch manufacturer Agrifac – another company of the Exel Industries group – has already launched a commercial version. It is called AiCPlus, for Condor self-propelled sprayers. Product manager Steven Koop explains how it functions: “We have RGB colour cameras mounted on the boom, recognising weeds and crops based on shape, structure and contrast. They detect certain weeds, such as Rumex species in a grass field, and will only apply chemical where the weed is. That way you use less chemicals, you don’t harm the grass, and you only kill the weeds. So it’s better for the crop and with less impact on the environment as well.”

With individually-controlled close-spaced nozzles, a Condor additionally equipped with Agrifac’s DynamicDosePlus technology can deliver variable rates from each nozzle in a minimum 25cm (10in) square under the control of a prescription map.

_{Text continues underneath image}

Rometron weed-detection is incorporated into the iPad operating system for SwarmFarm Robotics autonomous vehicles, now in commercial use in Australia for controlling a broad spectrum of broadleaf and grass weeds in fallow paddocks after grain or cotton crops. – Photo: Rometron

Hyperspectral imagery

Carbon Bee AgTech’s weed location and identification technology has attracted Kuhn-Blanchard, the crop protection arm of the Kuhn Group, to develop a spot sprayer using the firm’s AQiT-Sensor hyperspectral imagery system. This was developed principally for gathering crop data by drones or ground vehicles, to make informed management decisions. For this application, it incorporates internal GPS and a wifi access point and comes with a docking station that provides corrected images, interactive maps and results visualisation to highlight the presence of weeds or early disease symptoms.

Savings of up to 80%

Kuhn-Blanchard aims to adapt the artificial intelligence deep learning capability for real-time identification of different weed families within a crop, or post-harvest residues, for targeted spraying. Initial figures from trials are said to show savings of up to 80% in plant protection product usage by reducing treatments to the target plants. It also offers the prospect of improved weed resistance management, adding to the cost-savings and environmental benefits.