WO2022038363A1 - Agricultural machine - Google Patents

Abstract

Disclosed herein is an agricultural machine (10). The machine (10) comprises a supporting frame (11), ground contacting mobility means (12), a drive means (17), a row detector, a processor and at least one implement (18) configured for plant and/or soil engagement between crop rows, mounted to and extending from the supporting frame (11).

Description

AGRICULTURAL MACHINE

Field of the Invention

The present invention relates to an agricultural machine and method of operating an agricultural machine.

Background to the Invention

Common agricultural practice is to use a blanket approach for soil preparation, seeding, chemical application and harvesting. This is because historically it has not been financially viable to use different machines, or different sprays etc on different parts of the field. Such precision approaches are developing though, and with the development of sophisticated aerial analysis from unmanned aerial vehicles, farmers are now looking to a more direct approach in their farming practices. This is borne in part from the lack of sustainability in current practices. For example, relying solely on blanket herbicides for the control of weeds is not a sustainable model as weeds are becoming more and more resistant to applied chemicals. The use of blanket herbicides also impacts the quality and financial viability of the crop they are applied to.

Precision agriculture has developed the fastest in the vegetable crop industry. This is partly because the crop is high value and farmers can afford to implement more direct approaches. Crops are therefore planted in well-defined rows so that special implements can be used to cultivate between rows or harvest individual plants. Inter row hoeing is commonplace in the vegetable crop industry, and is an effective way of controlling the growth of weeds, but relies on the fact that the rows are already well spaced apart, and well defined, and sewn like that for the purpose of being able to hoe.

Doing the same with cereal crops is not viable as it would result in insufficient crop density to produce any decent yield. Therefore, cereal crops are grown very densely with a typical row width of only 10-12 cm and cover a very large area. So, in a very short space of time the rows lack definition due to the creation of the overlapping plant canopy. This means that any activity performed on an inter-row basis, such as spraying or hoeing or inter crop seeding within a cereal crop is in danger of damaging the crop itself rather than leading to an increase in yield. Farmers have turned to a blanket approach for cereal crops consequentially and the blanket approach used as described above. And to combat disease and herbicide resistance, farmers have turned to changes to rotations, cover crops, field drainage and equipment sterilisation to prevent weed spread and to improve yields.

There has now been devised an agricultural machine that overcomes and/or substantially mitigates the above referenced and or other disadvantages associated with the prior art.

Summary of the Invention

In an aspect of the invention there is provided an agricultural machine comprising, a supporting frame mounted on ground contacting mobility means or to a traction unit, a drive means mounted to the supporting frame or to the traction unit and configured to drive the mobility means or the traction unit, respectively, a row detector mounted to the supporting frame or traction unit and configured to determine the position of cereal crop rows, a processor configured to receive data from the row detector on the position of the crop rows, the processor also configured to navigate the machine between the rows by activation of the drive means in response to said data and at least one implement configured for plant and/or soil engagement between the rows, mounted to and extending from the supporting frame.

The machine is advantageous primarily because it enables working of the soil or plant material between the rows of a crop where the definition of the rows is not apparent, for example, where the vegetative canopy has formed and the ground or plants between the rows are not visible or consistent. Also, due to the ease with which the device operates, it can be introduced into a crop many times in the year. It has been observed during experimentation in support of the invention that weeds such as Blackgrass emerges after the Autumn planted crop emerges and when it does, it is a very feeble (in relation to the planted crop), one stemmed, small plant. This is true for many other types of weeds. Therefore, utilising the machine of the invention at this early stage, weeds can be effectively attacked. Thus, several passes with the device in the early growing season will be effective. Furthermore, the machine operates in a similar manner to a steerage hoe (although it is not restricted to performing the function of a hoe) and is therefore reactive to changes in direction of the rows. Fields are rarely square and drilled crops are rarely drilled in perfectly straight lines, so the ability to work plants and/or the soil on an inter-row basis and on a steerage basis, is beneficial to the farmer.

A cereal crop is a term used in the art and its conventional meaning is taken here. Cereal crops therefore include plants which are members of the grass family, perhaps grown for their edible starchy seeds. Typical cereal crops include but are not limited to wheat, barley, oats, rye and corn, triticale and grain millet, rice, or sorghum.

The implement is preferably a plant destruction implement configured to destroy plants between the crop rows. This is because inter row plant destruction is the most efficient and effective use of the machine. In this instance the implement is not selective of the type of plant it can destroy between the row. Thus, it will destroy weeds as well as any other plants such as crop plants sewn, or which have developed in error out of the normal row pattern. The implement may act on the soil or the plant directly in order to destroy the plant.

The plant destruction implement preferably comprises one or more of a hoe, tine, disc, strimmer, flail, shear, knife, or roller. These are all devices that can destroy plants. Other suitable plant destructions implements include but are not limited to electrocution units, thermal or flame emitters, solar directors or cutting knife.

The implement may be a seeding implement, chemical applicator, harvester or observation/measurement implement. Examples of suitable observa- tion/measurement implements include but are not limited to cameras or sen- sors. Plant and/or soil engagement between the rows is therefore not limited to requiring actual contact with the plants or soil between the rows and is not limited to plant destruction.

When the implement is a plant destruction implement, the machine is further advantageous because it selects out only the plants growing in the row. Thus, only the plants seeded or developed in the row, survive, albeit with a small population of weeds that exist between plants within the same row. But even then, with repeated usage of the machine from season to season the existence of weeds is eliminated. All this is done without the use of herbicides which boosts the efficiency and financial viability of the farm.

It is recognised that the device and method of the invention (described later) are particularly effective against Blackgrass (Alopecurus myosuroides), which is now resistant to most common agricultural sprays and exists in the soil from year to year.

The ground contacting mobility means may comprise a form of robotic mobility system, such as a form of legged device. Preferably though, the ground contacting mobility means comprises wheels or tracks. These are easiest to manufacture to the correct dimensions to run between the crop rows and provide the necessary traction for the machine in so doing.

Suitable traction units include but are not limited to tractors or caterpillars. Traction units are known to have their own ground contacting mobility means for pulling themselves and any attached devices along with them.

The drive means is preferably an electric motor and is powered by electricity stored in an on-board battery. This is the most environmentally friendly form of drive. The battery power source may also be used to power the processor and other associated parts of the machine requiring electrical power. The power source is preferably supplemented or recharged by solar energy and to do this there may be a photovoltaic cell or equivalent means mounted to the support frame. Therefore, the machine preferably further comprises an energy harvesting device and/or an energy storage device. Alternatively, the forms of power could come from thermal means, but this is less desirable from an environmental perspective. For example, there could be an on-board internal com- bustion engine generator to generate electricity to power the electric motors to drive the drive means and the remainder of the machine. Alternatively, the drive means may be an internal combustion engine directly driving the mobility means. In this case the internal combustion engine may secondarily generate electrical power to power the remainder of the machine. The drive means may be a hydraulic motor powered from a pump driven by an internal combustion engine or electric motor. As with the above, the internal combustion engine may secondarily generate electric power to power the remainder of the machine.

The row detector preferably comprises an image capture device, such as a camera or video device. However other methods of detection might be utilised such as motion sensor, temperature sensor, light sensor, pressure sensor, proximity sensor, accelerometer, gyroscope and/or moisture sensor. The row detector also comprises a CPU, configured to process the data collected by the row detector and calculate the location of the crop rows. The CPU is also configured to send said data and row location information to the processor of the machine.

The row detector is preferably configured for plant identification and/or identifies plants in use. This enables the machine to determine which plants are between the crop rows.

Preferably the support frame, mobility means, and associated components are constructed from lightweight materials so as to cause minimum impact on the ground upon which the machine operates.

The row detector may determine the location and/or space between the crop rows within an established crop. Preferably the row detector determines the position of crop rows within an established crop. This means that the machine can be used in an established crop and not just one where the plants have just emerged. An established crop might be one where there is a crop canopy and visibility of the soil or ground beneath the canopy is not consistent or complete.

The row detector performs a first step of eliminating the line features generated from horizontal leaf edges (i.e. those running left to right or right to left across a plan image) and selecting those forward-facing leaves/line seg- ments of crop rows. The row detector then performs a second step of calculating the centroid of each forward-facing line segment, feeding that into a meanshift clustering algorithm and then calculating the lane cluster centres, wherein the lane cluster centres are indicative of the presence of a cereal crop row.

To further improve the accuracy of the row detector, the row detector may perform any number of the following processes, including but not limited to overall Image Pre-Processing, Region of interest section, Colour Transformation and Threshold Image Determination, Direction Feature Extraction, Adaptive Crop Row Formation and Kalman Filter Tracking.

There may be a plurality of implements and at least one of the implements is movable along the width of the support frame and the lateral position of the at least one of the implements is controlled using a second drive means mounted to the supporting frame. This enables the separation of the implements to be controlled. In some instances, it may be beneficial to pre-set the separation along the support frame of the implements (or set it in use), for example if the machine is used in different crops that have a different row spacing. And rather than do this manually this can be done using the second drive means as described above. The control of the second drive means is preferably through the processor which may be remotely controlled for this purpose by the user of the machine.

To be movable along the width of the support frame the implements may be slidable along the support frame but held in position by a hydraulic ram. Alternatively, the implements may be mounted with a rack and pinion type mechanism on the support frame and lateral movement is affected by a motor actuator. Alternatively, the implements may be supported from a swing arm which is pivotally mounted to the support from. Movement of the swing arm maybe brought about by actuation of a connected hydraulic ram or other actuator, such as an electric motor.

In this example, preferably the processor is configured to, in use, control the lateral position of at least one of the implements in response to the crop row data from the row detector by activation of the second drive means. This means that the separation of the implements on the support frame is reactive to the position of the detected rows. Therefore, if the row separation changes or the rows bend then the position of the implements can adapt accordingly. This reduces the incidence of damage to the crop and improves the accuracy of the machine.

When the implements are not moveable along the support frame, preferably, there are a plurality of plant destruction implements spaced apart from one another on the support frame by a predetermined distance equal to the distance between the rows at seeding. This means that implements are in the right position and separation to fall between the crop rows on a pre-set basis. Therefore, when the processor navigates the machine between the rows by activation of the drive means in response to the row data the implements automatically fall between the rows and effect the soil and/or the plants therein.

The support frame may comprise a carriage which is suspended from the support frame and which carries the implements. Thus, the parts of the description above that refer to the different ways of mounting the implements to the support frame apply equally as if they were mounted to the carriage. The carriage may be moveable upwards or downwards (i.e. away from the soil/crop of closer to the soil/crop). Such movability may be affected by hydraulic rams, actuators, or other suitable mechanical means and controlled by the processor. Therefore, in use if the machine enters a part of the field where no interaction with the soil or plants between the rows is required, then the processor can control the implements to lift them out of the way. And vice versa when interaction is required.

The processor is preferably configured to receive data from a remote source to instruct it in its navigation. The data may directional data to instruct the processor to navigate the machine to particular area. This may be a particular area in a field where the machine is required to work, or in a greenhouse for example. The data may be functional data to instruct the processor to operate the implements or not operate the implements, for example if the machine is required to stop or start working. Or if one or more of the implements are required to be lifted out of engagement with the plant or soil between the rows. The remote source may be an unmanned aerial vehicle, satellite, other agricultural machine, or user interface. For example a UAV may take an aerial survey of a field and identify a patch of weeds. The UAV then sends directional data to the machine to tell it to navigate and move to that particular area. The machine then performs its work on that area under the continued instruction from the processor as directed from the UAV.

The processor is preferably configured to transmit data to a remote station. Said data in this instance may include but is not limited to information on the plant types identified, battery status, fuel status, and/or position.

In another aspect of the invention there is provided a method of operating an agricultural machine comprising, providing an agricultural machine having a supporting frame mounted on ground contacting mobility means or to a traction unit, a drive means mounted to the supporting frame or to the traction unit, a row detector mounted to the supporting frame or traction unit, a processor and at least one implement configured for plant and/or soil engagement between the rows, mounted to and extending from the supporting frame, determining the position of cereal crop rows by the row detector, the processor receiving data from the row detector on the position of the crop rows and the processor navigating the machine between the rows by activation of the drive means in response to said data to drive the mobility means or the traction unit, wherein at least one implement of the machine engages the soil and/or plants between the rows.

In the method, the agricultural machine may be the agricultural machine substantially as described above.

The machine delivers an efficient low labour and low carbon footprint solution through use of a propelled roving vehicle benefiting from imaging technology, either supplied by external mapping or determined on-board. The vehicle is configured to cope with precision farming requirements and aims to minimise crop damage. As a relatively low cost system it targets sustaining or improving on current yields whilst dramatically reducing input costs and working capital exposure for farmers at a time when farming subsidy payments are profiled to significantly reduce. The invention can not only reduce chemical usage, but the benefits to the farmer are greater as they drive yield by gaining control of weeds such as blackgrass. While the initial development market for the invention is the UK, substantial overseas markets are probable, not only for Blackgrass control but any arable row crop. Globally all row crop farmers (cotton, soya, etc.) are facing immense issues from resistance of weeds to glyphosate. The approach of the present invention is far simpler and more robust than the prior art systems.

Furthermore, there is now widespread recognition that modern heavy mass farm machinery is damaging soil. By contrast, the present invention involves a light weight vehicle that has low mass. This increases the sustainability of agriculture on soils, and the electric power can be provided from low carbon sources (solar PV / wind) reducing carbon dioxide emissions.

The invention demonstrates an effective robotic system which is scalable to a fleet control system. Preferably the machine is therefore self-propelled. The invention also demonstrates a device trailed from a traction unit, such as a tractor or caterpillar with relevant aspects of the machine integrated into or onto a second agricultural device such as a tractor unit. The machine may be operated autonomously. In such instances, the work program may be operated on a base station and transmitted to the machine to instruct it in its operation and navigational route. The autonomous path may be controlled by GPS.

Brief Description of the Drawings

An embodiment of the device will now be described with reference to the figures in which

Figure 1 shows a perspective view of an embodiment of the machine,

Figure 2 shows a rear view of the machine shown in figure 1 ,

Figure 3 shows a side view of the machine shown in figure 1 ,

Figure 4 shows a perspective view of a second embodiment of the machine,

Figure 5, shows a perspective view of a third embodiment of the machine,

Figure 6, shows a perspective view of a fourth embodiment of the machine and Figure 7 shows an example of the imaging results obtained from the row detector,

Figure 8 shows a perspective view of a fifth embodiment of the machine and

Figure 9 shows a perspective view of a sixth embodiment of the machine.

Detailed Description of the Illustrated Embodiment

The following description presents exemplary embodiments and, together with the drawings, serves to explain principles of the invention. However, the scope of the invention is not intended to be limited to the precise details of the embodiments, since variations will be apparent to a skilled person and are deemed also to be covered by the description. Terms for components used herein should be given a broad interpretation that also encompasses equivalent functions and features. Descriptive terms should also be given the broadest possible interpretation; e.g. the term "comprising" as used in this specification means "consisting at least in part of" such that interpreting each statement in this specification that includes the term "comprising", features other than that or those prefaced by the term may also be present. Related terms such as "comprise" and "comprises" are to be interpreted in the same manner. Directional terms such as “vertical”, “horizontal”, “up”, “down”, “upper” and “lower” are used for convenience of explanation usually with reference to the illustrations and are not intended to be ultimately limiting if an equivalent function can be achieved with an alternative dimension and/or direction.

The description herein refers to embodiments with particular combinations of features, however, it is envisaged that further combinations and crosscombinations of compatible features between embodiments will be possible. Indeed, isolated features may function independently as an invention from other features and not necessarily require implementation as a complete combination.

An example of the machine is shown in Figure 1. The machine shown in Figure 1 is a weeder and is generally designated 10. It will be appreciated however that this is only one example of the invention, and in other examples the machine is not a weeder but may be directed towards other tasks such as chemical spraying, spreading, seeding, ground working, harvesting or other common agricultural processes. To illustrate this example of the machine as a chemical spreader and as a seeder are provided below.

The weeder 10 is associated with development of an environmentally sensitive mechanical weed control system for use across arable farms. The motivation for the invention is to address the widespread issue of weed resistance, particularly the Blackgrass weed, within arable cropping land, with a view to providing an alternative method of husbandry to spraying with herbicides. The objectives are to optimise crop production and longer term soil quality through the reduced use of herbicides across arable farm enterprises.

The weeder 10 includes a main supporting frame 1 1 , primary ground contacting wheels 12 and leading/trailing stabilising wheel pairs 13 and 14 respectively. A direction of travel is denoted D. Stabilising wheels 13 and 14 are mounted for rotation at the end of legs 15 extending downwardly toward a ground surface from main frame 1 1 , while the much larger primary wheels 12 are mounted from an axle (or pair of axles) 16 that differentially drives said wheels 12 from a gearbox/electric motor unit 17 (see Figure 2). The illustrated embodiment resembles a simple trolley construction, although other configurations are possible.

The trolley can be simple because its main purpose is to provide a convenient platform from which to mount implements 18 that will perform mechanical destruction of weeds between crop rows, in use. In the form illustrated by Figures 1 to 3, implements 18 can be generally termed “hoes”, with a sharp end/head 19 for breaking earth and cutting the stems/roots of weed plants. However, alternative descriptions may be possible.

Each head 19 is mounted from a rod-like extension 20 extending downwardly toward a ground surface from the main frame 1 1 , the rods 20 being configured for movement as the wheels 12 drive the weeder 10 forward in the direction of arrow D. Reciprocating movement may be possible by a number of suitable means, such as turning a supporting bar 21 , along with a spring construction 22 at the end of the rods 20 distal from the heads 19. The rods 20 may be actuated simultaneously, individually or in groups depending on the effective- ness of each option. Alternatively, the hoes may be simply dragged along the ground, using momentum/power of the platform to drive the hoes into the ground.

The plurality of hoes 18 are spaced apart at a distance convenient to working the type of crop to be patrolled. For example, the distance may be I Q- 30 cm between rods. There must also be a clearance between respective heads 19 in order for a continuous row of crop plants to pass there through undamaged, while any unwanted plant in the substantive space between respective crop plants/rows is disturbed/unearthed/broken by the hoe and destroyed.

In another example of the machine according to the invention the machine has all the features substantially as described above, but at least one of the rods 20 holding the implements 18 is movable along the width of the support 21 and the lateral position of the at least one of the rods and therefore the implements is controlled using a second drive means (not shown) mounted to the supporting frame. This enables the separation of the implements to be controlled. Typically, all of the rods are moveable and controlled in this manner. To be movable along the width of the support frame the implements are slidable along the support frame, but held in position by individual electric motor powered actuators.

Navigation and control equipment is located in a housing 23 upon or within the main frame 11 . Suitable sub-compartments can house computer processors, batteries, cameras and other sensors but, it will be appreciated, the aesthetic considerations are not paramount to effectiveness of the device.

The weeder 10 is powered by electricity stored in an on-board battery (not explicitly illustrated although intended to be integrated with the motor/power plant 17, within housing 23 or elsewhere upon/within frame 11) and/or supple- mented/recharged by solar energy, e.g. from a photovoltaic or equivalent means.

The weeder 10 comprises all necessary components to provide an autonomous unit that could be externally/remotely activated and deployed.

Figures 4 to 6 illustrate alternative embodiments of the machine according to the invention. The machine is substantially as described above but with variations to the effective width (dictating how many rows of crop can be covered in one pass) and/or mechanical implement carried.

Figure 4 shows a machine 10 which is substantially as described with reference to Figures 1 -3, but where the total width is at least twice that of the embodiment from Figures 1 to 3. This simply achieves a greater coverage. Also, the destructive head 30 of the rod 20 is of a different, flat, form. Figure 5 shows a machine 10 which is substantially as described with reference to Figures 1 -3, but which is narrower than the machine 10 shown in Figure 1 , including a narrower form of hoe/tine 18, but otherwise analogous in operation. The simplest form of implement is in fact a tine 40 that is simply a ground contacting extension of a rod 20. Such tines 40 can be spring-loaded or more rigidly fixed and dragged behind the platform, disturbing the earth and any root systems for newly growing weeds. Figure 6 shows a machine 10 which is substantially as described with reference to Figures 1 -3 but of intermediate width and a yet further type of mechanical implement for performing the weeding function. The implement includes a cutting blade 50 and individual roller 55.

Further examples of the machine 10 are provided herein (not illustrated) where the machine is substantially as described with reference to Figures 1 -3 but where the plant contacting/destroying device include any of hoes, tines, sprung hoes, sprung tines, strimmers or mini-strimmers with a vertical or horizontal rotating head, mowers, flail mowers or shear. The implement type may be interchangeable into the main frame of the machine 10 and, indeed, any combination of such devices may be possible.

All variants of the machine are moved in a similar manner, namely driven and steered by differentially powering wheels 12. In the known way, a greater RPM of one wheel compared to the other will cause a change in course in the opposite direction, e.g. as the left wheel (facing the direction of travel) speeds up, or the right wheel slows down, a turn to the right will be executed. Wheels 12 are large so as to effectively steer the weeder and navigate through the uneven terrain typical of a planted field and between the rows. An alternative form of weeder may include tracks. The weeder 10 has a processor (not shown) housed within the housing 23. In other examples the processor is positioned elsewhere on the machine 10. The processor is in electrical communication with the power unit 17 which drives the wheels 12 and instructs the power unit 17 accordingly as described above when the machine 10 is required to turn left or right or drive straight on or back.

The processor is also in electrical communication with a row detector (not shown). In this case the row detector is housed within the housing 23, but may be located anywhere on the machine 10.

Navigational control of the machine 10 is achieved by the interaction between the row detector and the processor. The row detector detects the position of the crop rows as the machine 10 moves in direction D and sends data on the position of the rows to the processor. In turn the processor commands the motor drive unit 17 so that the wheels 12 fall between the rows.

Where the rods 20 holding the implements 18 are mounted at fixed positions on the support bar 21 , and spaced apart from one another on the support frame by a predetermined distance equal to the distance between the rows at seeding, the implements 18 are in the right position and separation to fall between the crop rows. Therefore, when the processor navigates the machine between the rows by activation of the motor/drive means in response to said data the implements automatically fall between the rows and effect the soil and/or the plants therein.

Where the rods 20 holding the implements 18 are movable along the support bar 21 or moveable laterally in some other mechanism, the motors which drive the position/deflection of the respective rods is controlled by the processor. In response to data from the row detector, the processor then makes sure that implements fall between the rows by operating the motors which move the implements laterally across the support and effect the soil and/or the plants therein. This embodiment of the invention is particularly useful if the row spacing alters across the width of the machine, for example if the machine 10 operates over a tramline, or where two drill passes meet and the spacing between the rows is either extended or decreased. Thus, the position of the implements is reactive to the position of the crop rows.

Navigational control of the machine 10 is also achieved by the interaction between the processor and a remote source. To do this the processor has a receiver and is able to receive data sent to it from a remote source to instruct the processor on its navigation course. The remote source may be a base station, or a UAV. The processor also has a transmitter and transmits positional data back to the source to confirm its position. The processor may also transmit other data such as plant type or battery status for example, back to the source.

Field trials can be employed to test the most effective methods of weed destruction and precision digital mapping of fields will be combined with engineered vehicles as described to provide a sustainable alternative to herbicidal use bringing financial and environmental improvements to the farming economy in the UK.

In an example of the invention the machine is integrated with an aerial surveying device, such as an unmanned aerial vehicle (UAV). Available unmanned aerial vehicles are employed to survey a field. Resultant image data is analysed (potentially in real time) for pinpointing the most problem areas for weeds such as Blackgrass. Alternatively, the UAV might identify other problem areas that require site specific use of the machine, for example site specific fertiliser application, or slug pellet application. Once analysed, the UAV sends location data to the processor and the processor then navigates the machine to the locality of the problem area within the field. A machine is thus deployed and directed to specific problem areas to automatically commence weed destruction with an appropriate implement. A implement can be chosen dependent on the image analysis and seriousness/age of weed infestation.

Where the machine is a weeder it is particularly effective against Blackgrass. The Blackgrass weed (Alopecurus myosuroides, also known as slender meadow, foxtail, black-grass, twitch grass, and black twitch) is an annual grass causing a widespread problem for arable farmers on cultivated land, being particularly abundant in winter crops. The weed is observed to produce a large amount of seed which is shed before the crop is cut and has developed re- sistance to a range of herbicides used to control it. Blackgrass can occur at very high densities, competing with the crop and seriously reducing the yield of crops such as wheat and barley if not controlled. The threat is purported to cost UK agriculture up to £0.5 billion per annum. There is presently no alternative to herbicides that kill the blackgrass weed in the current crop that can be deployed cheaply at field scales.

Blackgrass is one of the most economically costly weeds in Western Europe (Moss et aL, 5 2007 Weed Tech. 21 , 300-309) and the major weed of winter wheat, one of the UK's most valuable crops. Severe blackgrass infestations can reduce yields by up to 12% and cost up to £300 per ha in lost yield and extra control costs (Hicks, 2018, Nature, Ecol and Evol, 2, 529-26), and UK losses to the weed can reach up to £500m per annum if effective controls cannot be found. To date control of Blackgrass has relied on the intensive use of herbicides. However, 80% of fields surveyed across England were resistant to all three major herbicide groups used to target Blackgrass (Hicks, 2018), making herbicide centred control non-viable going forward. There are also growing concerns over evolving glyphosate resistance, a broad-spectrum herbicide (kills weeds and crops) of last resort in this system. The invention represents a vital alternative control to the current range of herbicides. The machine offers an alternative that can control blackgrass cheaply and at scale, preserving an “organic” status.

It is noteworthy that it takes most weeds less than a week to germinate and about another week to start developing true leaves. The easiest time to kill such weeds is after they have cotyledons have just appeared, and before they have true leaves to start putting energy back into the roots. Similarly, with perennial weeds like grasses, the new, tender shoots are much easier to hoe back than established stems. Knocking back perennial weeds while they are young helps deplete the roots of energy. Those skilled in the art will appreciate the most critical opportunity for mechanical destruction, e.g. hoeing. This may be the week after planting of the crop.

Thus, the machine and method of the invention proposes a more targeted approach by implementation of a fully autonomous, camera guided, robotic steerage hoe (or other form of mechanical destruction) to control blackgrass. Autonomy reduces operational costs and the robot work plan will be guided by prior data on weeds maps provided by drones. Additionally, implementation of the invention enables assessment for how agronomic changes can be integrated with hoe operations, changes in drilled row widths can enhance performance; slightly wider drill width may increase hoe speed and accuracy whilst minimising crop damage.

In another example of the machine, the machine comprises substantially all of the features described above, except that instead of a hoe 19 as one of the implements 18, the ends of each of the rods 20 have suspended from them a spreader plate. A hopper is present on the top of the machine above the housing 23, and pipes run from the hopper to each of the spreader plates. Each of the spreader plates are rotatory and powered electrically by motors mounted to the end of the rods 20. Each motor is in electrical communication with the battery on the machine. The processor controls the operation of each of the motors to operate the spreaders individually. The hopper is filled with chemical agents such as slug pellets, fertiliser or herbicide. As the machine moves forward, the agents fall down the respective pipes and contact the respective spreaders. The spreaders are operated and the this spreads the agent around the base of the crop between the crop rows. The benefit of such a system is that the chemical agents do not contact the leaves of the crop which might otherwise damage the crop and allows for a much more targeted precision application of the agent. Ultimately this means that less of the agent is used than in conventional applications.

In another example of the machine, the machine comprises substantially all of the features described above, except that instead of a hoe 19 as one of the implements 18 or a spreader, the ends of each of the rods 20 have suspended from them a sprayer nozzle. A tank is present on the top of the machine above the housing 23, and pipes run from the tank to each of the nozzles. The tank is filled with chemical agents such as herbicide or insecticide or fertiliser. As the machine moves forward, the agents are pumped along the respective pipes by an on board pump to the respective nozzles. The nozzles are operated and the this sprays the agent around the base of the crop between the crop rows. The benefit of such a system is that the chemical agents do not contact the leaves of the crop which might otherwise damage the crop and allows for a much more targeted precision application of the agent. Ultimately this means that less of the agent is used than in conventional applications.

The following is an example of the row detector as described above.

The row detector comprises stereo video cameras mounted to the frame of the machine 10 and mounted to look forwardly in order to detect the crop in the direction of arrow “D”. In this example two cameras are mounted to each side of the frame at the front. The cameras take continuous images as the machine 10 moves forward. The two cameras are linked to a row detector CPU which in use operates a number of calculation processes in order to define the crop rows and provide data to the main processor of the machine 10.

The method followed by the row detector starts with colour space segmentation of foreground green either in HSV space or a single channel depth map. The first step is a set of image pre-processing operations to separate green pixels (crops and weeds) from the rest of the image (soil, stones, and others). To overcome the problem of canopy overlapping between rows, the inventors surprisingly found that they could determine if a plant leaf (line segment) and therefore a plant was a lane candidate by searching for a number of line segments, obtaining the maximum accumulation of segmented pixels along vertical alignments, and applying an angle (e.g. 60 - 120 degrees) constraint as line segments selection. The inventors also surprisingly found that by using mean-shift clustering, an unsupervised machine learning method, they could provide an adaptive lane separation strategy, with neighbouring line segments associated grouping with corresponding ownership of index (dictionary). Linear regression is then utilized for the line fitting to form the designation of the separated crop rows. Data association and Kalman filtering is then used to perform row following by associating tracking the multiple detected lanes across frames.

Details of each step in the method are given in the following subsections. A. Image Pre-processing Image pre-processing is first used to reduce inherent noise of the camera sensor and blurring from its motions, and environmental effects including variable illumination due to weather conditions and time of day. In-field wheat row video is captured in RGB colour space with the vehicle-mounted cameras, together with depth map images from stereo cameras (Zedcam, www.stereolabs.com). The task is to detect and track the rows as the vehicle drives forwards. This is then used to control the vehicle to keep its wheels between rows. Individual image frames are extracted from the decoded video streams for sequential single frame processing. A parsing step converts raw data into various format, visible RGB and a 3D depth map. The original image contains not only the lane information, but also a lot of redundant information and non-lane noise in the surrounding area, which can have a negative impact on image segmentation and line detection. This redundant information is cropped out using an adaptive area of interest (ROI) on the image. Image warping is used to translate the images into a ‘top-down’ view. It is conducted through perspective transformation in 2D space. In the default camera perspective, objects further away from the camera appear smaller and the row lines appear to converge the further they are from the camera, which isn’t a true representation of the real world. The perspective transforms the perspective of the image as vertical. Image warping brings distortion effects into the image quality. The warped image becomes more ’blurry’ afterwards. Image sharpening and filtering is performed to recover some of this lost detail. As the video is captured in RGB colour space it is sensitive to light intensity which may change due to time of year, clouds and time of day. The RGB images are converted into the HSV colour space, defined by Hue (colour or tint), Saturation (amount of grey) and brightness value (V) (HSB refers to the same model where B represents brightness here). In this case crop rows are dominated by the colour green. The HSV colour space separates hue (including the amount of green) from brightness and intensity. Green pixels are extracted from the image with a colour mask in the HSV space. The values of green colours are very distinctive from the background in the V-component compared to other colours and are easily extracted, providing a good basis for the colour based segmentation. Experi- ments show that the colour processing performed in the HSV space is more robust to detecting specific targets. The green colour space is split from HSV image, with a bitwise mask operation. This image presents a good explanation on our colour transformation strategy here. The binary feature image was further processed using the threshold method based on the brightness values in HSV colour space. Although in normal cases the colour transform can benefit the lane detection, it is not robust and has limited ability to deal with shadows and illumination variation, etc. So the other image processing techniques are essential to enhance the image features according to different task, e.g., enhancement of sharpness and contrast, etc. The depth map is the pseudo image converted from 3D cloud points by built in functionality of the ZED camera. For the depth map image, one channel was extracted for the composed three dimensional pseudo image. Next contrast enhancement is performed. Contrast enhancement is a process that makes the image features stand out more clearly by making optimal use of the colours available in the image. A histogram equalization method is used for this which achieves a global contrast improvement. Histogram equalization spreads out the most frequent intensity values, i.e. stretching out the intensity range of the image. This allows for areas of lower local contrast to gain a higher contrast on HSV colour space image. The next step is binarization. In conventional computer vision, binarization of the image is the key to extract target objects. The large number of frames in the video are captured from various environments and weather conditions. Each single image frame is individually colour scaled and blurred. In order to cater for different lighting and contrast conditions, an adaptive threshold is utilized during the preprocessing phase. An OTSU is used made adaptive with mean and Gaussian kernel (OpenCV). The OTSU involves iterating through all the possible threshold values and calculating a measure of spread for the pixel levels each side of the threshold. An adaptive threshold varies depending on the characteristics of its neighbourhood pixels. So different thresholds are obtained for different regions of the same image which gives better results for images with varying illumination for colour based object detection. OpenCV library provides a built-in method of adaptive thresholding where the threshold value is calculated by the mean of the neighbourhood or Gaussian-weighted sum of the neighbourhood. With the thresholded binary image, the possible undesirable regions are removed such as black or white spots by morphological operations, e.g. erosion and dilation. Those spots in the image are side effects obtained during the binary conversion or the remaining noise in previous steps. The next step is noise filtering/edge keeping. This operation is carried out on binary image for noise removal. It replaces the intensity of each pixel with a weighted average of neighbourhood pixel values, which removes unwanted noise but preserves features such as edges. The last standard image pre-processing based on morphological operation is region filling.

B. Feature Detection

A major problem to overcome in the image analysis is how to analyse the image when the leaves are crossing over. This happens most often when a canopy has formed in the crop. It is important to be able to do this to identify individual crop lanes/rows from each other. Cereal crop leaves are long, narrowed, ellipsoidal shape. The solution is to eliminate the line features generated from horizontal leaf edges (i.e. those running left to right or right to left across the image), and only select those forward facing leaves of crop rows. This therefore overcomes the ambiguity in multiple row separation.

First of all, a feature extraction module is added before the lane formulating, i.e. in colour space, a directional feature extraction is conducted using as ’X-gradient’ feature by Sobel filter (a discrete differentiation operator by Irwin Sobel and Gary Feldman, 1968) before further applying the edge feature detection with a canny edge detector. This is possible because the image from the field presents very rich ominous features. In the spatial domain, Hough transformation is used to produce the line segments which although omni-bearing are mostly forward facing (i.e. in the direction of travel of the machine/vertical). The directional line segments selection is then applied, where each line segment’s slope is calculated and only restricted directional slopes are selected, so the favourable line segments (i.e. pure forward facing) can be obtained. The purpose of feature extraction is to help keep any features that may belong to the lane feature in forward facing direction and filter out features that may be exag- gerated in horizontal lane crossover buried. A Sobel filter is used to carry out directional feature extraction when the image is highly blurry with rich and various features. The consecutive operation of edge feature detection is applied to provide the edge image for Hough line extraction.

C. Edges extraction

The Canny filter also calculates the gradient (derivative), using nonmaximum suppression to find the local maxima in order to obtain the “thinned edge image”. This is a key element for the line extraction. Generally, the Canny edge detector is utilized to extract the ‘borders’ of the image in the spatial domain and provide the preliminary foundation for Hough line detection later. Normally, it offers a good signal-noise ratio, compared to other border extractors. It depends on how the blurry crop rows are. The Sobel filter has a natural directional gradient calculation and is used to obtain gradients along vertical or horizontal image. The Sobel operator here acts as selected feature filter before edge detection is applied. Based on those narrow down gradients, the extracted edges features can be seen from selected directional gradients, but all Hough line segments extracted from edge features are not perfectly forward facing. This means that the extracted Hough line based on edge images are not essentially strict vertical or horizontal. It is worth pointing out that, in the ideal case, the edges extracted can be narrowed down from more specific directional gradients, but the Hough line formed from edge features won’t be directional restricted and this will add unavoidable noise. Nevertheless, the above operation still helps the later line refining and reduce the computation cost of detecting horizontal wheat/grass canopy and occluded crossover between neighbouring rows.

D. Line segments

The search for pixels that belong to row markings consists of looking for green features in the vertical direction. Edge features detection is undertaken prior to line segment extraction by Hough transformation. To obtain the vertical line segments and ignore the horizontal line segments, we apply the Sobel operator on the pre-processed image to calculate the gradient magnitude and direction. We then apply a Canny edge detector to the vertical gradient features only, to detect and connect edges with thinning lines. For straight line segment extraction we use the Hough transformation, a method for extracting parametric curves from images. The Hough transformation is used to detect any shape that can be expressed in a mathematical formula, even if the shape is broken or distorted. It is assumed that crop rows will be straight lines at the scale of the images (a few m2 is size) and so use the equation for a straight line. In practice, to make the Hough transformation work in real time a probabilistic Hough transform is commonly used.

E. Line segments Filtering/Restriction

Based on the line detection module in the previous steps, the most Hough lines obtained from the edge image are lying in forward-facing direction. These carry the potential information to designate an individual lane, therefore those directional line segments have to be selected for the lane forming in the forward facing fashion. To achieve this each line segment slope is calculated, and only lines with angles ranging between 60 degree and 120 degrees are selected as lane line candidates.

F. Line segments Clustering

After the crop row aligned (forward facing) line segments are selected the centroid of each line segment is calculated. The line segment centroids are then clustered so that each line segment can be assigned to a crop row. Mean shift clustering is used. An algorithm scans line candidates, searching for groups of line segments which are spaced within a range of crop row widths. This tuning parameter can be varied based on prior information on crop row spacing. The row width is considered to be between fifty and two hundred in pixels in image coordinates.

G. Lane centroids

The above mean shift operation provides clusters centroids as one of the outputs. Those cluster centroids are used as lane centroids, which can be regarded as potential lane separation points.

H. Line Segments Sorting /Refining

Mean Shift provides a good solution on basic clustering, but the pre-set or estimated standard variance of clustering data sets cannot be perfectly accu- rate. Together with noise data, they can have lot impact on both the number of centroids estimated and their accuracy in position. It will later affect the ultimate clustering expectation. The solution on this is then to use a recursive data sorting routine combined with bandwidth, to eliminate close neighbour’s ambiguity.

I. Line Segments Association

When the lane centroid pattern is found, the centroid point is labelled as a lane marking. Using those cluster centroids, the nearest neighbour association algorithm is applied to retreat all possible line segment centroids associated with corresponding lane/rows. Each individual lane line cluster is grouped separately and stored in dictionary below.

J. Lines regression

Lines regression is simply multiple line regression obtained by each individual line data fitting. It is achieved here using the formula y = mx + b in terms of linear regression, where m is the slope of the line and b is the y-intercept. The solution is based on the condition of the smallest overall distance from the line to the points. By recalling the lane line segments (centroids) coordinates stored in the database using dictionary key, a line regression is applied for the data fitting, which provides the estimation of slope and intercept from the straight Hough line segments. The individual lane is therefore able to be identified with drawing accordingly.

K. Lane Forming

With lines regression, lane forming is therefore naturally obtained for displaying, which consists of lane centroids, intercept, etc. This represents the hypothesis of the lane estimation as position and trajectory relative to the natural crop row in the farmland.

L. Multiple lane measurement (dictionary)

The clusters are stored in a dictionary data structure for convenient search when needed.

M. Tracking with Kalman Filter

To estimate lanes continuously in a video stream a Kalman filter was applied. The crop rows are not always measurable due to the breaks in the crop rows which can be caused by sudden changes in soil conditions, mechanical inconsistency in the crop drilling machine and pest damage. Kalman filters are used as a method to track the centroid of each row that is robust to these environmentally imposed inconsistencies. The Kalman filter estimates the lane centroids based on the current row label and assigned row ID in previous video frames. With the lane information recorded in the dictionary, the centroid of the row is calculated and used as measurement for the purpose of data fusing through linear Kalman filter. Although, there is different way to choose the state as measurement, like intercept, slope, or box coordinates, etc. here for the simplicity and lower computing, the lane centroid is used due to the image is always approximately rectified as vertical warped in each frame, so the tracking state can be simplified for fast computing and avoiding infinity slop in practice. We implemented Kalman filter in dynamic fashion, and each individual crop row lives as an instance in memory. The Centroid horizontal pixel value is used as tracking ID which is being updated in each processing cycle. This is convenient in controlling and updating, see the Fig.21 below. The method is robust to work under the above-mentioned image sequence.

An example of the results obtained by the row detector is shown in Figure 7, which shows the rows of wheat identified in an established wheat crop. The size of the ROI was 1060x1000 pixels which was cropped from video frame in 1080x1920 pixels. An offset of 50 pixels was used for top-down view warping. The row width metric in each frame was temporary set as 75 pixels which was adjustable. Figure 7 shows the typical testing results performed in real-time from the video cameras, where the black rows are the row detection, while white as tracks for both estimation and prediction when the natural row is broken. The results depicted here are based on the algorithms described working on both RGB and depth map image data sets. The time cost for each from was 0.05 seconds running on intel i7 3.1 GHz 64bit CPU in average. The method described here provides a solution for row detection and following in cereal crops. The results show that the method is able to identify cereal crop rows under circumstances of highly occluded leaf crossover between adjacent rows. It can detect any number of crop lanes in the farm fields. It works with either high or low image resolutions. It is able to operate in real time on relatively low cost hardware. The machine and method are useful in tasks such as precision weed control, crop care and monitoring, and harvesting.

Another example of the machine according to the invention is shown in Figure 8. The machine is designated 100, but is substantially as described above apart from a few differences. The machine 100 has four wheels 120 (only 3 shown). Each wheel 120 has an axle which is rotationally connected to an axle support. A first pivot arm 122 connects onto the lower part of a frame extension 110 at either the front or back of the machine 100 and is further connected to the lower part of the axle support. The upper part of the axle support has connected to it a second pivot arm 124. The opposite end of the second pivot arm 124 is connected to the frame extension 110 also but above the position where the first pivot arm connected. In this manner the wheels 120 can travel upwards and downwards, to enable travel of the machine over rocky or rough terrain, but the wheels themselves remain vertically tracked true. An electric motor is attached to each of the axle supports and is functionally connected to the axle so that in use the motor is able to drive each of the wheels. In this way, the speed of each of the four wheels is independently controlled. The machine 100 is a two wheel steer device. The front two wheels (the front being the end with the bumper 150) are steerable. To achieve this each axle support has connected to it a steering arm 160, which is mounted at its opposite end to a centrally mounted drive motor. Operation of the drive motor moves the steering arms forward or backwards, and therefore affects movement of the front wheels left or right. A carriage 170 is provided for an array of spring tines 18. The carriage is suspended from the frame 110. The array comprises three rows of tines 18 with each row being laterally off set from one another, therefore the spacing between the tines if viewed from the front of the machine 100 is equal to the row spacing of the crop and is pre set by the farmer in use. The carriage 170 is mounted between the front wheels and back wheels, so is effectively centrally mounted. This gives it the greatest downward pressure to effect deep soil/plant interaction. Another example of the machine according to the invention is shown in Figure 9, which is substantially as described with reference to the machine shown in figure 8, but the steering mechanism and the lift mechanisms for the carriage 170 is different. The carriage 170 is suspended from two control arms 172 which are pivotally connected to lift arms 174. The lift arms 174 are fixedly mounted to the ends of a rotatable actuator tube 176 which is rotated by a hydraulic ram 178. When the ram 178 is extended the tube 176 is turned and the arms 174 are lifted thereby lifting the arms 172 which in turn lift the carriage 170 holding the tines 18 out of the soil or away from the plants. The tines are lowered into contact with the soil/plants when the reverse operation happens. At the front end of the carriage there is mounted two downwardly extending ground rollers 179 (only one shown). These engage with the ground as the machine moves forward and prevent the tines 18 from embedding too deeply into the soil. They also control the depth the tines are engaged with the soil or plants. At the steering end of the machine 100 there is a downward extension 110. An axle 112 is mounted at its middle to the extension 110, then at each end of the axle is a downwardly extending axle extension 114 which each carry a wheel. The extension 114 have attached to them a steering arm 16 which is actuated by a hydraulic ram 162. Actuation of the ram 162 causes the extensions to rotate with respect to the axle 112 and thereby turn the wheels either both left or both right.

It will be recognised that with the machines shown in figures 8 or 9, the tines may be replaced by any other agricultural implement in order to engage the soil or plants between the cereal crop rows.

In another example of the invention there is a tractor provided as a traction unit (not shown). The machine is substantially as described above, but without the wheels or architecture on the machine for steering etc. The support frame 11 or carriage 170 holding the implements 18 is suspended from the front or rear of the tractor. This may be typically from a standard three point linkage well known in the art. The row detector is mounted to the frame or carriage and detects the crop rows as the machine moves forward, substantially as described above. The processor controls the steering mechanism on the tractor to navi- gate the machine between the rows by activation of the drive means in the tractor in response to the data received from the row detector. This makes sure that the wheels of the tractor fall between the rows in use, and the implements are engaged between the crop rows in use. Where the implements are moveable along the bar 21 , frame 11 or carriage 170, the processor also controls the movement of the implements there, to further maintain the tools in engagement between the crop rows.

Claims

- 29 -CLAIMS

1 . An agricultural machine comprising, a supporting frame mounted on ground contacting mobility means or to a traction unit, a drive means mounted to the supporting frame or to the traction unit and configured to drive the mobility means or the traction unit, respectively, a row detector mounted to the supporting frame or traction unit and configured to determine the position of cereal crop rows, a processor configured to receive data from the row detector on the position of the crop rows, the processor also configured to navigate the machine between the rows by activation of the drive means in response to said data and at least one implement configured for plant and/or soil engagement between the rows, mounted to and extending from the supporting frame.

2. A machine according to claim 1 , wherein the implement is a plant destruction implement configured to destroy plants between the crop rows.

3. A machine according to claim 2, wherein the plant destruction implement comprises one or more of a hoe, tine, disc, strimmer, flail, knife, shear, or roller.

4. A machine according to claim 1 wherein the ground contacting mobility means comprises wheels or tracks.

5. A machine according to claim 1 or claim 2, further comprising an energy harvesting device and/or an energy storage device.

6. A machine according to any preceding claim, wherein the row detector comprises an image capture device.

7. A machine according to any preceding claim, wherein the row detector is configured for plant identification and/or identifies plants in use.

8. A machine according to any preceding claim, wherein the support frame, mobility means and associated components are constructed from lightweight materials so as to cause minimum impact on the ground upon which the machine operates.

SUBSTITUTE SHEET (RULE 26) - 30 -

9. A machine according to any preceding claim, wherein the row detector determines the position of crop rows within an established crop.

10. A machine according to any preceding claim, wherein there are a plurality of implements and at least one of the implements is movable along the width of the support frame and the lateral positon of the at least one of the implements is controlled using a second drive means mounted to the supporting frame.

11. A machine according to claim 10, wherein the processor is configured to, in use, control the lateral position of at least one of the implements in response to the crop row data from the row detector by activation of the second drive means.

12. A machine according to any preceding claim, wherein there are a plurality of implements spaced apart from one another on the support frame by a predetermined distance equal to the distance between the rows at seeding.

13. A machine according to any preceding claim, wherein the processor is configured to receive data from a remote source to instruct the processor in its navigation.

14. A machine according to any preceding claim, wherein the processor is configured to transmit data to a remote station.

15. A machine according to any preceding claim, wherein the row detector is configured to perform a first step of eliminating the line features generated from horizontal leaf edges (i.e. those running left to right or right to left across a plan image of the crop) and selecting those forward facing leaves/line segments of crop rows.

16. A machine according to any preceding claim, wherein the row detector is configured to perform a second step of calculating the centroid of each forward facing line segment, feed that into a meanshift clustering algorithm and then calculate the lane cluster centres, wherein the lane cluster centres are indicative of the presence of a cereal crop row.

17. A method of operating an agricultural machine comprising, providing an agricultural machine having a supporting frame mounted on ground contacting mobility means or to a traction unit, a drive means mounted

SUBSTITUTE SHEET (RULE 26) to the supporting frame or to the traction unit, a row detector mounted to the supporting frame or traction unit, a processor and at least one implement configured for plant and/or soil engagement between the rows, mounted to and extending from the supporting frame, determining the position of cereal crop rows by the row detector, the processor receiving data from the row detector on the position of the crop rows and the processor navigating the machine between the rows by activation of the drive means in response to said data to drive the mobility means or the traction unit, wherein at least one implement of the machine engages the soil and/or plants between the rows.

18. A method according to claim 17, wherein the machine is substantially as described with reference to any of claims 1 to 14.

19. A method according to claim 17, or claim 18, wherein the implement is a plant destruction implement configured to destroy plants between the crop rows.

20. A method according to claim 19, wherein the plant destruction implement comprises one or more of a hoe, tine, disc, strimmer, flail, knife, shear, or roller.

21. A method according to any of claims 17 to 20, wherein the machine comprises ground contacting mobility means comprising wheels or tracks.

22. A method according to any of claims 17 to 21 , further comprising an energy harvesting device and/or an energy storage device.

23. A method according to any of claims 17 to 22, wherein the row detector comprises an image capture device.

24. A method according to any of claims 17 to 23, wherein the row detector is configured for plant identification and/or identifies plants in use.

25. A method according to any of claims 17 to 24, wherein the agricultural machine comprises a support frame, mobility means and associated components that are constructed from lightweight materials so as to cause minimum impact on the ground upon which the machine operates.

26. A method according to any of claims 17 to 25, wherein the row detector determines the position of crop rows within an established crop.

SUBSTITUTE SHEET (RULE 26)

27. A method according to any of claims 17 to 26, wherein there are a plurality of implements and at least one of the implements is movable along the width of a support frame and the lateral positon of the at least one of the implements is controlled using a second drive means mounted to the supporting frame.

28. A method according to claim 27, wherein a processor is configured to, in use, control the lateral position of at least one of the implements in response to the crop row data from the row detector by activation of the second drive means.

29. A method machine according to any of claims 17 to 28, wherein there are a plurality of implements spaced apart from one another on the support frame by a predetermined distance equal to the distance between the rows at seeding.

30. A method according to any of claims 17 to 29, wherein the processor receives data from a remote source to instruct the processor in its navigation.

31. A method according to any of claims 17 to 30, wherein the processor transmits data to a remote station.

32. A method according to any of claims 17 to 31 , wherein the row detector is configured to perform a first step of eliminating the line features generated from horizontal leaf edges (i.e. those running left to right or right to left across a plan image of the crop) and selecting those forward facing leaves/line segments of crop rows.

33. A method according to any of claims 17 to 32, wherein the row detector is configured to perform a second step of calculating the centroid of each forward facing line segment, feed that into a meanshift clustering algorithm and then calculate the lane cluster centres, wherein the lane cluster centres are indicative of the presence of a cereal crop row.

SUBSTITUTE SHEET (RULE 26)