On this page you will find short summaries of some of the non-confidential research projects that we have undertaken over the past 20 years. Many of these projects have contributed to the technology used in our current commercial applications as well as those we are developing for the future.Autonomous Crop Treatment Vehicle (1993 - 1996)
Highly targeted individual plant scale operations was thought at that time to be too slow for manned vehicles. We therefore developed an autonomous vehicle as a platform for experimental plant scale operation. A treatment device consisted of an array of solenoid operated nozzles at 50 mm pitch. Selective treatment was achieved by switching each individual nozzle on or off as the vehicle progressed through the crop.
The planting pattern of the crop, imaged by a vehicle mounted camera, was tracked using an extended Kalman filter. To allow for reliable navigation, including headland turns, the Kalman filter also integrated odometric and inertial sensor data. In addition to providing guidance information, tracking the crop pattern allowed the discrimination of crop - which matched the planting pattern - from weeds which did not.
This project was concerned with reducing agrochemical inputs through the use of inter-row cultivation to control weeds. This practice goes back to Jethro Tull's invention of the seed drill in the 18th Century. However, the advent of herbicides all but eliminated inter-row hoeing in the UK. Our objective was to reinvigorate this environmentally friendly technique and make it commercially viable again by the application of modern robotics technology. Defra funded an investigation into the potential for use in cereals and with support from the British Beet Research Organisation we extended the research to cover sugar beet. Within this project we collaborated with ADAS, IACR, Garford Farm Machinery, Robydome Electronics and KRM Ltd.
The principle technical problem was to locate and track the crop rows such that cultivators could be brought into close proximity to crop plants, thus maximising weed kill, without reducing yield. We chose to develop a non-contact sensing system using low cost CCD video cameras and computers. This approach had the advantage of being very flexible in respect of alternative crops and growth stages. However, in agricultural fields we were faced with a number of challenges. These included naturally variable lighting, crop emergence that may be incomplete and weeds that tend to mask crop rows.
Our approach had two principle innovative steps. The first was in the initial image processing in which we exploit prior knowledge of crop row spacing to identify crop rows. The second was in the application of a tracking algorithm to track row location between images. Together these give very reliable performance under a wide range of conditions typically placing cultivators well within +/-3cm at speeds of up to10kph and beyond. This performance has proved commercially attractive. Working with Garford Farm Machinery and Robydome Electronics we brought a product to market in 2001 under the brand name Robocrop.
We used our ability to accurately locate crop rows to deduce weed density on the basis that green plant material between rows will be weed. This technique accurately detects weed density at early stages of crop growth when rows are discrete. We have used similar techniques to successfully assess crop density. The illustration above is of an automatically generated weed map in a field of winter wheat grown on a 25cm row spacing. The technique could be applied to a wide range of agricultural and horticultural crops and linked to either weed map generation or real time sprayer control.Top of page
Diminishing herbicide options, fear of groundwater contamination and
customer pressure to minimise herbicide use are all pushing the
horticultural industry away from reliance on herbicides. However,
product contamination concerns, much of which relate to weeds,
necessitate high levels of weed control and have resulted in increasing
use of unsustainable hand weeding.
Our scientific approach to the project was to develop a fast, two dimensional mathematical template matching techniques,(exploiting periodicity within the planting grid), enabling individual crop plants with some spacing variability to be located. Regular observations of plant position are passed to a tracking algorithm that can follow plant location from a moving vehicle.
A novel shallow cultivation mechanism was also developed. This is synchronised with the plant tracking algorithm enabling weeds to be removed from between crop plants leaving the crop undisturbed.
The sequence of images above shows an experimental cultivator operating above ground on an artificial crop to demonstrate the principle of operation An experimental implement was constructed based on a standard front mounted Garford vision guided inter-row steerage hoe. A camera was mounted centrally on the implement looking ahead and down such that the full width of the bed was visible over a length of approximately 2.5m. Each hydraulically driven rotating disc cultivator was attached to a depth wheel unit mounted on the implement so that cultivation depth could be consistently maintained (normally at approximately 20mm). The experimental implement had the capacity to take up to five rotating disc units.
Results showed that the machine was effective at removing weeds within the row without removing crop plants. A field trial has shown that a single pass of the machine incorporating intra and inter-row cultivation removes 80% of weeds.
This project was sponsored by Defra under the Horticulture Link Programme and was conducted jointly with Warwick HRI who provided the weed science and agronomy expertise. The project was part funded by the Horticultural Development Council and had the following additional industrial partners: Robydome Limited, Garford Farm Machinery , Edwards Brothers, Robert Montgomery Limited, Allium & Brassica Centre , AGCOWithin-row weeding machines based on this technology were first sold in 2008 and have continued to evolve with over 90 machines sold by 2016 . Click on this link to see a 25 second, 6 MB Windows Media File format video a commercial within-row weeding machine in action
Due to field variability in the process of drilling and germinating seed some high value row crops are sown at a higher density (typically x3) than that which is ultimately desired, and then post emergence, mechanically thinned to the required stand. The selective thinning process has the objective of leaving the healthiest plants at as close as possible to the optimum spacing. The intelligence required to make this selection coupled with the accurate mechanical control required has ensured that this process is almost always conducted manually using hand hoes. There is world wide interest in reducing difficult repetitive field work due to labour availability, cost and health concerns.
The aim of this project is to develop an intelligent thinning technology that uses an ability to look ahead and automatically make selection decisions to maximise crop quality and yield. The project is utilising much of the technology developed in the within row weeding project described above. Unwanted plants are removed using a selective thinning disc similar to that used for removing weeds between transplanted crops. The innovation specific to the thinning application lies mainly in detecting the very small closely spaced seedlings and developing algorithms to automatically decide which plants to remove and which to keep. There are also mechanical challenges relating to the variability in plant spacing that occurs through this process.
Results from initial experimental work funded by the Douglas Bomford Trust in 2007 were encouraging and so we have continued development.. That follow on work has included field testing in California with the assistance of Robocrop importer Solex Corporation. Thinning software has been loaded onto two within-row weeders and tested in lettuce grown on 38" and 80" bed configurations. The photograph below show the effect of running the automatic thinner on lettuce drilled at a nominal 3.5" spacing set to produce a stand at 11.5". The area to the left is unthinned and the area to the right has been thinned. The final spacing is generally regarded as at least as good as that achieved manually, though the machine sometimes has difficulty removing plants ("doubles") that have grown closer(<2.5") to each other than intended. This problem would not occur often with a well set up drill, though it is accepted that drilling sometimes has to be conducted under less than ideal conditions and so work is continuing to improve performance in this respect.
A 22 second, 5 MB Windows Media File format video of automatic thinners being trialled on commercial lettuce crops in California can be viewed by clicking here.
EU legislation (e.g.the revision of 91/414 EEC and the Water Framework Directive) is reducing herbicide availability - the limited range of herbicides remaining does not cover the weed spectrum encountered and for some weed species there is,or soon will be, no means of control. Mechanical weed control is now more widely practiced, but there are a number of circumstances when these methods are unsatisfactory, in wet weather, and for control of perennial weeds and species with a strong tap root. Flame and steam weeding are damaging to invertebrates and consume large amounts of energy. Hand labour has now become expensive and scarce. Targeted application of herbicides to weeds that are difficult to control mechanically is an attractive option potentially providing good control with minimum chemical quantities and thus a low cost and environmental impact.
Losses resulting from pesticide reviews are reducing selective herbicide options to the point that for some situations there are no products available. For example, the Essential Use of post emergence metoxuron (Dosaflo) used to kill potatoes in a wide variety of vegetable crops expired at the end of 2007. Volunteer potatoes are a particular problem as they are competitive and not very susceptible to mechanical methods. Furthermore, surviving volunteer potatoes provide a bridge for potato disease, notably blight, which is a concern for potato growers.
In the absence of selective herbicides the targeted application of total herbicides to weed potatoes is an attractive option providing good control with low cost chemicals that have a low environmental impact. The only available commercial application technique using wipers relies solely on height differential between crop and weed and has important limitations in some circumstances.
The aim of this initiative was to develop and demonstrate a computer vision based technology that uses weed detection and targeted application of minimal quantities of a total herbicide (e.g.Glyphosate)to control volunteer potatoes in a range of vegetable crops particularly in onions and carrots.
Initial Horticultural Development Council funded work conducted jointly by The Arable Group's spray unit at Silsoe and THT showed great promise and so a follow on Hort Link project was undertaken. Discrimination between live plant material and background was on the basis of colour as developed in earlier projects. The discriminators of weed from crop that were investigated included a combination of factors including plant size, shape, height (using optical flow), and position relative to crop rows. In the final implementation crop height was not measured as the relatively high steep downward camera view that allowed best determination of the other parameters was not well suited to accurate height measurement by optical flow. Spot spraying of individual plants requires novel spray solutions as the required combination of narrow fan angles, low drift and relatively low volumes are rare amongst agricultural spray nozzles. These factors coupled with the need to form a spray immediately and to stop it without dripping for pulses of down to 30ms duration provide a challenging specification. The consortium has therefore been evaluating and developing novel nozzle technologies which meet this specification. Field plot trials in 2008 were conducted in commercial onion and carrot crops. A single bed width spray boom fitted with spot spray nozzles was fitted to a toolframe tractor equipped with a camera and computing equipment for image processing, target tracking and nozzle control. Overall results showed promise with a very high proportion (95-75%) of volunteer potatoes being killed,though there was sometimes crop damage in the area immediately around target. Very low levels of crop loss may be tolerable in commercial situations.
A new experimental spot sprayer spanning three 1.8m beds was developed for the 2009 season. This was based on a disc steered self leveling front mounted toolframe built by Garford Farm Machinery. Normally this type of toolframe would be fitted with intra or inter-row cultivation blades, but in this case it was fitted with three spray bars each consisting of a 2m long extruded aluminum section with mounting faces on all four sides. This allowed mounting brackets, electronic boxes and hoses to be conveniently mounted on the bar whilst allowing unrestricted lateral adjustment of nozzles mounted along the forward face. Hypro EU Ltd provided "Alternator" nozzles especially designed for this application. These nozzles feature narrow (13°) fan angles and produce droplets well suited to the spot application of glyphosate. The machine was equipped with three cameras connected to a Core Duo PC that processed images, tracked weeds and interfaced with a network of CAN equipped microcontrollers that controlled toolframe functions such as steering and leveling as well as the on/off control of individual nozzle solenoids. An additional cab mounted PC provided a user interface and display of live images.
The three bed experimental machine performed well in trials covering several hectares through the 2009 season. The maximum speed was 5 kph and good control of volunteer potatoes was achieved with low levels of damage that were acceptable to growers. Click on this link to see a 31 second, 4 MB Windows Media File format video of the experimental spot sprayer in action
Trial crops showing automatically spot treated to the left and untreated to the right (onions left parsnips right)
This project was conducted under the Horticulture Link Programme with Defra and the Chemical Regulation Directorate as co-sponsors. The project was part funded by the Horticultural Development Council and The Potato Council and had the following additional partners: TAG, Robydome Limited, Garford Farm Machinery , Micron Sprayers, Allium & Brassica Centre, Hypro EU Ltd, Monsanto, Hunterpac Ltd, F B Parrish and Sons, A Findlay.Top of page
This project followed on from the work on volunteer potato control described above and sought to broaden the range of target weeds against which the approach can be applied and widen the range of crops in which it can operate. Specifically we operated with both selective and non-selective herbicides in leeks, onions and sugar beet. This has important implications for the weed detection algorithms and for the spray application technology. Experimental evaluation of the technology was conducted in commercial crops from a modified form of the 6m wide experimental spot sprayer built for the earlier volunteer potato control work.
The approach to detecting large weeds developed in the earlier work was refined to improve accuracy and facilitate faster operation. In the image above the detection algorithm has drawn a polygon around large broadleaf weeds within an onion crop. Weeds are tracked down through successive images their outline being continuously refined until, as they drop out of the bottom, they are placed into a treatment map for spraying. A revised "Alternator" nozzle was designed by Hypro and shown in the image above. The new cassette based design alows different spray tips to be fitted for different spray properties. For example spot application of glyphosate benefits from easily directable low dfift large droplets (~1000µm) whilst selective herbicides normally require a good coverage of smaller (<300µm) droplets.
In addition a different approach for detecting patches of smaller weeds was developed building on our earlier weed mapping work described above in which inter-row areas known to be free of crop are assessed for greeness as a measure of weed density. This was tested in both leeks and sugar beet. In the former case the weed density was recorded and used to create the weed map shown below.
This project was conducted under the Horticulture Link Programme with Defra and the Chemical Regulation Directorate as co-sponsors. The project was part funded by the Horticultural Development Council, the British Beet Research Organisation and The Potato Council and had the following additional partners:TAG/NIAB, Robydome Limited, Garford Farm Machinery , Allium & Brassica Centre, Hypro EU Ltd, Monsanto, Allpress Ltd, F B Parrish and Sons, A Findlay.