METHOD FOR AUTOMATED AGRICULTURAL LAND MANAGEMENT AND ASSOCIATED SYSTEM

Description

TECHNICAL FIELD OF THE INVENTION

[0001] The field of invention is agriculture.

[0002] More specifically, the invention relates to an automated weeding process for agricultural land and an associated autonomous system.

[0003] The invention finds applications in particular in the context of large-scale farming, especially for combating secondary plants, called adventitious plants or more commonly known as "weeds", growing in place of the main plants cultivated on agricultural land. STATE OF THE ART

[0004] Weed control is crucial for achieving a good yield in a field, as weeds can become dominant, hindering the growth of the main crop. It's important to note that weed control is generally most critical in the early weeks of cultivation, before the main crop has matured enough to become the dominant weed.

[0005] Thus, prior art is known from the techniques of maintaining agricultural land by weeding.

[0006] These weed control techniques generally use herbicides such as glyphosate, spread over large areas to combat weeds.

[0007] The major drawback of these techniques is that the products used, although effective, are generally harmful to the surrounding ecosystem.

[0008] As legislation becomes increasingly restrictive regarding the use of such plant protection products, limiting both treatment periods and the range of active ingredients in order to reduce residues in plant production and diffuse pollution in the natural environment, other weeding techniques, such as mechanical ones, are increasingly preferred.

[0009] These techniques, which are generally tedious, require regular monitoring of agricultural land to carry out weeding operations in a timely manner in order to prevent weeds from becoming predominant in the crop field.

[0010] To address these drawbacks, techniques have been developed to automate these weeding tasks by automatically recognizing weeds on which a weeding action is carried out.

[0011] However, these techniques are generally based on regular passage over agricultural land, in order to be able to carry out the weeding operation as early as possible, which has the disadvantage of being time-consuming and very costly from an energy point of view.

[0012] With automation in mind, techniques using drones for terrain mapping have also been developed. Examples of such techniques are described in US patent application US2017 / 0127606 for guiding non-autonomous agricultural machinery based on terrain features, and in Russian patent RU2710163 for identifying an area of ​​abnormal cropping and guiding agricultural machinery to treat it. The Russian patent describes, in particular, outlining such a plot in an image by defining a contour using high contrast, without specifically identifying the origin of the crop anomaly observed in the acquired images.

[0013] None of the current systems can simultaneously meet all the required needs, namely to offer an autonomous weeding technique for agricultural land that is effective against the weeds growing on the agricultural land, without damaging the existing crops, and is energy-optimized. DESCRIPTION OF THE INVENTION Goals

[0014] The main objective of the invention is to propose an effective weed control technique for agricultural land, enabling timely control of weeds growing on cultivated land.

[0015] Another important objective of the invention is to ensure the destruction of weeds without using plant protection products that are harmful to the environment.

[0016] Another notable objective of the invention is to offer a technique that is autonomous, particularly in the planning, optimization and execution of agricultural land treatment tasks.

[0017] Another objective of the invention is to offer a technique that is competitive with phytosanitary control methods.

[0018] Another objective of the invention is to offer a technique that is optimized from an energy point of view, while being as environmentally friendly as possible.

[0019] Another objective of the invention is to offer a technique that is robust and suitable for large-scale farming. Detailed presentation

[0020] Accordingly, the invention relates, in a first aspect, to a method of maintaining agricultural land as described in claim 1.

[0021] Thus, the autonomous vehicle moves within the agricultural land in an optimized manner, taking into account the relative position of the different crop areas as well as the orientation of the rows of each crop area to be treated.

[0022] It is important to note that each cultivation zone comprises parallel rows. However, the rows in different cultivation zones are rarely parallel to each other. The orientation of the crop rows generally corresponds to the movements of agricultural machinery that sowed the crop established in the cultivation zone.

[0023] It should also be noted that agricultural land can be vast, particularly in the context of large-scale farming. Large-scale farming refers to an operation comprising a plurality of large agricultural plots, generally discontinuous, with a great variability in soils, terrain, or natural or man-made obstacles (stones, power lines, ditches, hedges, rivers, etc.).

[0024] Therefore, the order in which cultivated areas are treated is prioritized based on the assessed agronomic pressure on each area. In other words, the cultivated areas to be treated first are those with higher agronomic pressure than others, reflecting the presence of weeds at a greater density and / or in a more advanced stage of development.

[0025] The assessment of agronomic pressure thus makes it possible to have suitable information to determine which crop areas should be treated as a priority and consequently to optimize the actions of the autonomous vehicle on areas which have an urgent need for intervention in order to avoid slowing down the development of the chosen plant in cultivation in the area.

[0026] Furthermore, autonomous vehicles typically move slowly: around 1 km / h during maintenance and around 5 km / h during relocation. Therefore, the autonomous vehicle's intervention is optimally planned based on an assessment of agricultural pressure. Areas experiencing the highest agricultural pressure at any given time can thus be prioritized by the autonomous vehicle.

[0027] It is of course important to emphasize that the relative position of the crop areas, in relation to the position of the autonomous vehicle, and between them, is generally also taken into account in the ordering of the list of crop areas to be treated in order to optimize the routes of the autonomous vehicle while prioritizing crop areas with high agronomic pressure from weeds.

[0028] In particular embodiments of the invention, the step of generating a diagnosis of agronomic pressure includes a substep of automatic recognition of the nature of a weed present in an image by means of a machine learning method previously trained on a database of images of identified weeds.

[0029] The recognized weed is generally a weed that is predominantly present in the image.

[0030] The machine learning method, such as a neural network, allows for the recognition of a weed's presence based on machine learning parameters generated during a training phase. This training is typically conducted in several stages to refine the machine learning parameters and thus better identify weeds in images using a database of previously identified weed images.

[0031] Generally, the viewing angle is advantageously similar between the acquired images and the images in the database. For example, when the acquired images are taken from above, training is performed on a database of images taken mostly from a top view, that is, from a vertical viewing angle.

[0032] In particular embodiments of the invention, the step of generating a diagnosis of agronomic pressure includes a substep of predicting the evolution of the development state of a weed at a given time as a function of at least one measured and / or predicted meteorological data.

[0033] Thus, the assessment of agronomic pressure is more precise because it depends on at least one meteorological data point, such as humidity, instantaneous temperature, or wind direction. It is important to note that meteorological data can be measured at the time of calculating the prediction of the evolution of the developmental stage at a given point in time, generally in the future. The prediction can therefore be made, for example, one day, three days, five days, one week, ten days, fifteen days, one month, or two months in advance.

[0034] A meteorological parameter used for predicting the state of development can also be predicted at the same given time or at a distinct given time.

[0035] In particular embodiments of the invention, when the maintenance process uses a plurality of autonomous vehicles, the list of cropping areas allocated to an autonomous vehicle takes into account the order of the cropping areas to be treated according to the agronomic pressure associated with each cropping area and the position of each autonomous vehicle at a given time.

[0036] In particular embodiments of the invention, the list of cultivation areas allocated to an autonomous vehicle is drawn up taking into account the autonomy of said vehicle at a given time in relation to a charging station for said vehicle.

[0037] Thus, the vehicle can return to the station, after carrying out its maintenance operations on the cultivation areas allocated to it, in order to recharge an energy storage device, such as a battery or a hydrogen tank, included inside the autonomous vehicle.

[0038] In particular embodiments of the invention, the maintenance process also includes a step of recognizing the direction of the rows in the growing area and determining a spacing between two rows from at least one image acquired covering the growing area and / or from a recording of a plurality of positions of a seeding tool used during a prior sowing phase, the seeding tool including a geolocation device.

[0039] This reconnaissance step is generally carried out concurrently with the stage of developing the map of the agricultural land in order in particular to be able to delimit on the map the parallel crop rows within a cultivation area.

[0040] In particular embodiments of the invention, the method includes a step of automatically delimiting a cultivation area by analyzing the direction of previously recognized rows.

[0041] In other words, the maintenance process also includes a step of automatically delimiting a crop area on the agricultural land by means of an automatic recognition of the direction of the rows of a crop area in at least one image acquired covering said crop area and / or from a recording of a plurality of positions of a seeding tool used during a prior seeding phase, the seeding tool including a geolocation device.

[0042] In particular embodiments of the invention, the maintenance process also includes a preliminary step of acquiring images of the agricultural land by at least one autonomous image acquisition device.

[0043] Advantageously, the autonomous image acquisition device is an autonomous ground robot or an autonomous aircraft. The autonomous image acquisition device generally includes a motorized means of locomotion such as a wheel or a propeller, as well as a camera.

[0044] The images acquired by the autonomous aircraft are generally aerial images taken directly above the farmland to minimize perspective effects. However, it is possible for the aircraft to capture images of the farmland from a different angle.

[0045] An autonomous aerodyne can be, for example, a drone generally equipped with at least one propeller capable of providing lift to the drone, or a flying wing without a propeller.

[0046] In particular embodiments of the invention, the overflight of an aerodyne is carried out at a maximum altitude of around one hundred meters above the surface of the agricultural land for mapping, and / or at an altitude of around three meters above the ground for the development of the agronomic pressure of weeds.

[0047] The flight used to assess agronomic pressure in an area of ​​farmland is advantageously conducted at a lower altitude than the one used for mapping in order to obtain a higher-resolution image of the soil, thus allowing for better detection of the type of weed present in the image. This flight is generally referred to as a very low-altitude flight.

[0048] In some embodiments of the invention, the maintenance process includes a step of determining at least one no-go zone for autonomous vehicle movement on the map.

[0049] This restricted area may correspond to the presence of an obstacle to the free movement of an autonomous vehicle on agricultural land. The obstacle can be of any type, such as a pole, a tree, a river, etc.

[0050] The establishment of a restricted area can be done automatically by the computer system or manually by an operator.

[0051] In some embodiments of the invention, the maintenance process also includes a step of associating a type of crop with at least one cultivation area of ​​the agricultural land.

[0052] Thus, it is possible to refine the value of agronomic pressure according to the evolution of the type of crop in the cultivation area.

[0053] Furthermore, given that each crop has particular characteristics, it is possible to better assess the movements of an autonomous vehicle by taking into account, for example, the typical row width adapted to the crop.

[0054] The invention also relates, according to a second aspect, to a system for maintaining agricultural land as described by claim 13.

[0055] In particular embodiments of the invention, the maintenance system also includes at least one autonomous image acquisition device.

[0056] The autonomous image acquisition device can be, for example, a mobile ground robot, which can be called by the English term " rover, or an aerodyne.

[0057] In particular embodiments of the invention, the maintenance system also includes a charging station for an autonomous vehicle and / or an autonomous image acquisition device. BRIEF DESCRIPTION OF THE FIGURES

[0058] Other advantages, purposes and particular features of the present invention will become apparent from the following non-limiting description of at least one particular embodiment of the devices and methods of the present invention, with reference to the accompanying drawings, in which: there figure 1 is a schematic view of an example of a particular embodiment of a maintenance system according to the invention; the figure 2 is a synoptic view of an example of an implementation of a maintenance process according to the invention implementing the maintenance system of the figure 1 ; there figure 3is an example of an image of the agricultural land being maintained during a stage of the maintenance process of the figure 2 ; there figure 3A is an example of a track obtained by a seeding tool performing a sowing operation on a plot of agricultural land. figure 1 ; there figure 3B is an example of trace processing of the figure 3A carried out during a step of the process of the figure 2 ; there figure 4 is a synoptic view of a step in the maintenance process of the figure 2 ; there figure 5 is a perspective view of the autonomous vehicle of the figure 1 ; there figure 6 is a perspective view from the drone of the figure 1 . DETAILED DESCRIPTION OF THE INVENTION

[0059] The present description is given by way of non-limiting attribution, each feature of an embodiment being able to be advantageously combined with any other feature of any other embodiment.

[0060] It should be noted from the outset that the figures are not to scale. Example of a particular embodiment

[0061] There figure 1 illustrates a maintenance system 100 according to the invention used to carry out weeding operations on an agricultural plot 110 comprising a plurality of cropping zones 120, each presenting a plurality of parallel crop rows 130.

[0062] In this non-limiting example of the invention, the agricultural land 110 comprises three contiguous cultivation zones 120, forming a single plot. However, the invention can also be applied to discontinuous agricultural land, for example, plots located far apart from one another, each plot comprising at least one cultivation zone.

[0063] In addition, the plant, such as for example barley, wheat, maize or rapeseed, chosen for cultivation in the three 120 cultivation zones is identical here but could very well be distinct.

[0064] The maintenance system 100 here includes at least one autonomous vehicle 140 configured to perform maintenance operations on the agricultural land 110, at least one drone 150, which is an aerodyne, configured to acquire at least one aerial image of the agricultural land 110 by flying over the agricultural land 110, at least one charging station 145 for the autonomous vehicle, at least one charging station 155 for the drone 150 and a computer device 160, in this case a server, configured to manage the maintenance system 100.

[0065] It should be emphasized that the drone 150 is an autonomous image acquisition device. Alternatively or in addition to the drone 150, another motorized autonomous image acquisition device, such as a second drone or a ground-based mobile robot 151 acquiring partial images of the terrain during its movements, can be included in the maintenance system 100.

[0066] The computer device 160, comprising a processor 161 and a computer memory 162, processes instructions for a maintenance process 200 illustrated in figure 2 in the form of a synoptic diagram, stored in computer memory 162.

[0067] The maintenance procedure 200 includes a first step 210 of image acquisition using the drone 150 flying over the agricultural land 110 at a maximum altitude of approximately one hundred meters above the surface of the agricultural land directly below the drone 150. Geographic coordinates, obtained for example via a geolocation device integrated into the drone 150, as well as the image orientation relative to a cardinal direction, are advantageously associated with each acquired image. The images acquired by the drone 150 are also generally time-stamped.

[0068] An example of a 300 image of the 110 agricultural field acquired by the 150 drone is shown in figure 3 .

[0069] From the acquired images, the computer device 160 develops, during a second step 220 of the process 200, a map of the agricultural land 110 in which the different zones 120 of cultivation and the rows 130 of cultivation are automatically delimited.

[0070] To this end, an image of the agricultural field 110 is first reconstructed from a plurality of images acquired using the drone 150 during a first sub-step 221 of step 220. The reconstruction is carried out taking into account the geographic coordinates and orientation associated with each image. The position of each image can be adjusted by superimposing at least two elements common to each image, for example by calculating a sharpness criterion at the image superposition level.

[0071] The 130 crop rows are then recognized in the reconstructed image of the agricultural field 110 during a second sub-step 222 of step 220 using, for example, an image segmentation or edge detection method.

[0072] Alternatively or in addition, crop row recognition is based on recording the positions of a seeding tool used during a preliminary seed sowing phase on the agricultural field 110. The seeding tool includes for this purpose a geolocation device, for example of the GPS type (acronym for the English term " Global Positioning System " ,providing a position at regular intervals. The recordings over a period of time form a trace on a map. By analyzing this trace, it is possible to identify different crop rows by identifying parts of the trace that are substantially straight over a predetermined length, for example, at least five or ten meters. An example of a seed drill trace traversing a plot of agricultural land is shown in figure 3A The track is advantageously processed by removing the track sections corresponding to intermediate paths representing the movement of the seeding tool from one row to another. The remaining track sections correspond to the position of the 130 crop rows, as illustrated in figure 3B .

[0073] When the position of the 130 rows is detected, their direction is analyzed to automatically delineate the 120 cropping zones, each defined by substantially parallel cropping rows, in the reconstructed image of the agricultural field 110 during a third substep 223 of step 220. It should be noted that two cropping rows are generally considered to be in the same cropping zone when they form an angle of less than five degrees. For example, three 120 cropping zones are determined on the 300 image of the figure 3 As for the figure 3B , plot 320 was delimited into five 120 cultivation zones.

[0074] It should be noted that an operator may intervene during the process to make a correction or addition to a boundary of a 120-square-meter crop area delimited on the map.

[0075] An average spacing between two rows 130 can then be determined for each cultivation zone 120 during a fourth sub-step 224 of step 220. The average spacing is calculated from straight lines, each representing one row.

[0076] An association of a crop type with at least one crop zone 120 can also be made during a possible fifth sub-step 225. This association can be made automatically by recognizing the crop type on an image or manually by an operator.

[0077] In a third step 230 of the process 200, an automatic analysis of the map, on which the initial and final positions of the autonomous vehicle 140 are determined, is performed by the computer device 160 to allocate to the autonomous vehicle 140 a list of crop areas to be treated, taking into account its range for moving between the initial and final positions. The initial position corresponds to the position of the autonomous vehicle 140 at a given moment; this position can be instantaneous or predicted. The final position corresponds, for example, to the position of the charging station 145 for the autonomous vehicle 140.

[0078] It should be noted that the autonomous vehicle 140 generally moves slowly, at a speed of approximately 0.5 km / h to 2 km / h when maintaining a 120-square-meter crop area, and at a speed of approximately 4 to 6 km / h during a relocation journey, i.e., outside of maintenance periods. This relocation can be within or outside a 120-square-meter crop area.

[0079] A movement plan for the autonomous vehicle 140 is then determined during a fourth step 240 of the process 200. This movement plan includes an optimized trajectory for the autonomous vehicle 140 allowing the best use of the energy stored in an energy storage device included in the autonomous vehicle 140, while respecting the movement constraints according to the crop rows of each crop area treated by the autonomous vehicle 140.

[0080] During a fifth step 250 of the process 200, the autonomous vehicle 140 moves following the previously established movement plan.

[0081] To improve the allocation of crop areas to be treated by the autonomous vehicle 140, the process 200 may include, prior to step 230, a step 260 in which the computer device 160 generates a diagnosis of the agronomic pressure of weeds within all or part of the crop areas 120. The agronomic pressure due to a weed is assessed based on the type of weed, its density in a crop area, and / or its stage of development. The agronomic pressure of a weed reflects its ability to become dominant in the analyzed crop area, consequently slowing the growth of the cultivated plant in that area.

[0082] Thus, if weed pressure is low, the chosen crop can grow normally. Conversely, if weed pressure is high, for example above a predetermined threshold, the development of the chosen crop is significantly reduced because weeds become dominant in the cultivated area and divert essential resources, such as water and sunlight, to their own benefit, to the detriment of the chosen crop's development.

[0083] As an illustration, table 1 shows the number of plants sufficient per m2 to reduce the yield of cereal crops by 5%. [Table 1] Bedstraw 1.8 Wild oats 5.3 Poppy 22.0 Matriculate 22.0 Ryegrass 25.0 Foxtail 26.0 Stellar 26.0 Veronica of Persia 26.0 Véronique F de L 44.0 Deadnettle 44.0 Myosotis 66.0 Thought 133.0 Lady's mantle 133.0

[0084] In order to assess the agronomic pressure related to a weed in a cultivated area, a weed identification is generally carried out during a first sub-step 261 of step 260 illustrated by a specific synoptic diagram presented in figure 4 , on at least one image of all or part of the corresponding culture area acquired previously.

[0085] Recognition is performed, for example, using a machine learning method such as a neural network, previously trained on a database of images of weeds whose identification has already been carried out. More specifically, the neural network can indicate the probability of the presence of a weed in an image provided to the network.

[0086] It is important to note that the images used for reconnaissance are generally different from those used for mapping. For accurate reconnaissance, it is preferable to use high-resolution images with a narrower field of view than those used for mapping. To this end, the 150 drone typically performs a flyover of the agricultural land at a lower altitude, around three meters, to acquire more precise ground images during an optional preliminary sub-step 269. The 150 drone can also be equipped with several image acquisition devices with distinct objectives: one dedicated to acquiring images for mapping with a wide field of view (i.e., a short focal length), and another dedicated to acquiring images for weed recognition with a narrower field of view (i.e., a longer focal length).

[0087] In the case of two autonomous image acquisition devices, it may be possible to dedicate one to image acquisition for mapping and the other to image acquisition for weed recognition.

[0088] An assessment of the number of weeds of the same species in an image is then carried out during a second sub-step 262 of step 260. It should be emphasized that the assessment of the number of weeds on a surface is equivalent to the assessment of a weed density for example per square meter.

[0089] The number of weeds can be assessed automatically, for example, by counting the weeds identified for a given species. This assessment can be performed on all images covering a 120-square-meter crop area, or on a sample of images associated with a 120-square-meter crop area.

[0090] The developmental stage associated with a weed species detected in a 120-square-meter crop area is assessed in a third sub-step 263 of step 260 using the acquired images. This assessment involves estimating, for example, a characteristic value related to weed development, such as weed size, number of leaves, leaf density, leaf size, etc. This characteristic value is then compared to a characteristic weed development scale to evaluate the associated developmental stage.

[0091] In one variant of this particular embodiment of the invention, the developmental stage is obtained concurrently with the identification of the weed by the machine learning method. In this case, the database comprises weed images, each associated with a developmental stage of the weed.

[0092] A growth model adapted to the weed can be used to predict the evolution at a given time of the developmental state of the weed, during an optional fourth sub-step 264 of step 260. Such growth models are described for example in the book "Architecture and growth of plants, Modeling and applications" by P. de Reffye et al.

[0093] It is worth noting that the growth model used can be enhanced with current and / or predicted meteorological data, such as temperature, wind direction, humidity, sunshine, or rainfall, to improve the prediction of the weed's development. This prediction can be made for one day, three days, five days, or even one or two months.

[0094] Based on the current and / or future developmental stage of the weed, an assessment of the agronomic pressure associated with each weed detected in a crop area is carried out during the fifth sub-step 265 of step 260. For example, a weed with an advanced developmental stage has a greater agronomic pressure than a weed with a lower developmental stage.

[0095] It should be emphasized that the assessment of this agronomic pressure can also take into account a comparison of the development of the weed with the development of the cultivated plant at a given time, whether current or future.

[0096] As an illustration, Table 2 presents an example of the evolution of agronomic pressure from different types of weeds, taking into account a simplified model of plant growth as a function of sunlight duration, temperature, and humidity. The parameters used here are 14 hours of sunlight per day, 70% humidity, and a temperature of 10°C. [Table 2] Day Area Evolution index J0 J0+5 J0+10 J0+15 J0+20 PAA 1 0,3 12 14 17 21 27 PAA 2 0,2 25 26 28 32 37 PAA 3 1,4 2 9 23 45 73 PAA 4 1,1 57 62 73 89 110 PAA 5 2,1 5 16 37 68 111

[0097] In Table 2, it can be seen that zone PAA 5 could be treated as a priority instead of zone PAA 2 because the weed, although presenting a lower agronomic pressure at time 0, will tend to grow very quickly as shown by the value of the agronomic pressure after 20 days, which is three times greater than the agronomic pressure of zone PAA 2. The order of the 120 crop zones to be treated can therefore be modified taking into account the agronomic pressures of the weeds expected at a given deadline.

[0098] A representative value of the agronomic pressure of weeds is calculated during a sixth step 266 of step 260 for each crop zone 120 by combining the agronomic pressures of each weed detected in each crop zone 120. The representative value may, for example, correspond to a maximum value, an average value of the agronomic pressures, or a sum of the values ​​of the agronomic pressures associated with each weed detected in the crop zone 120.

[0099] It is important to note that agronomic pressure, or a representative value of agronomic pressure in a growing area, can be weighted according to the type of crop in that area, in order to account for the relationship between crop development and weed pressure. Furthermore, this weighting allows for a more accurate comparison of weed pressures between different growing areas where different crops are cultivated.

[0100] A ranking of the 120 crop areas to be treated as a priority is then carried out during a seventh sub-step 267 of step 260.

[0101] The list allocated to the autonomous vehicle 140 during step 230 can then take into account the order of the crop zones 120 to be treated as a priority. A crop zone 120 with high weed pressure can thus be treated as a priority by the autonomous vehicle 140. It should be emphasized that this crop zone 120 to be treated as a priority is not necessarily the crop zone 120 closest to the autonomous vehicle 140, which will therefore expend energy to move to this crop zone 120 to be treated as a priority.

[0102] When the maintenance system 100 includes a plurality of autonomous vehicles 140, the distribution of the crop areas 120 to be treated among the different autonomous vehicles is optimized by taking into account the order of the crop areas 120 to be treated as a priority and the position of each autonomous vehicle 140 at a given time. This optimization generally also takes into account the position of one or more charging station(s) 145 by managing the charging times of each autonomous vehicle 140.

[0103] It should indeed be emphasized that the number of charging stations may be less than the number of autonomous vehicles. In which case, since the charging of all the autonomous vehicles cannot be carried out simultaneously, a clever management of the charging periods of each autonomous vehicle can be put in place in order to avoid unnecessarily immobilizing an autonomous vehicle.

[0104] To prevent an autonomous vehicle 140 from being hindered by an obstacle during its movement or from moving outside a designated area, the method 200 may advantageously include, before step 240 of determining a movement, a step 270 of determining at least one no-go zone for an autonomous vehicle 140 on the map. This no-go zone can be automatically identified beforehand on an image by recognizing, for example, an obstacle such as a pole, a tree, a river, a ditch, a significant elevation change, a roadside verge, etc. This no-go zone can also be added to the map by an operator.

[0105] The autonomous vehicle 140 can thus maintain the agricultural land 110 while respecting the order of the zones 120 to be treated which are allocated to it.

[0106] To that end, as illustrated in more detail on the figure 5The autonomous vehicle 140 is equipped with a set of specific mechanical tools 410, generally of the weeding type, oriented towards the agricultural terrain 110. The tools 410 are located here below the chassis 420 of the autonomous vehicle 140.

[0107] The chassis 420 houses an electronic device 460 equipped with a processor 461 and a computer memory 462 enabling in particular the control of the movement of the autonomous vehicle 400 based on the previously established movement plan.

[0108] To follow the travel plan, the autonomous vehicle 140 can advantageously include a geolocation device 470, for example of the satellite geo-positioning type such as the GPS system (acronym for the English term " Global Positioning System " or Galileo.

[0109] It should be noted that the travel plan was generally communicated to the autonomous vehicle 140 via a wireless communication device 475 included in the autonomous vehicle 400, and connected to the electronic device 460.

[0110] To detect the presence of a weed locally, the autonomous vehicle 140 is equipped with four cameras 490, located at the front of the vehicle 140, whose optical axis is advantageously oriented towards the ground on which the autonomous vehicle 140 is moving. Images acquired at regular intervals by the cameras are analyzed by the electronic device 460, which can then control the action of a mechanical tool 410 based on the weed identified in the image. Weed recognition is performed here using a machine learning method based on a neural network, previously trained on a database of images of identified weeds.

[0111] The images acquired by the 490 cameras can also be transmitted to the computer 160 of the maintenance system 100 to improve the mapping of the agricultural land and the determination of weed pressure. A correlation can thus be established between the weed pressure assessed in the area 120 of crops being treated by the autonomous vehicle 400 and the weed pressure observed during the treatment. This correlation improves the assessment of weed pressure based on the images acquired by the drone 150.

[0112] There figure 6 is a perspective view of drone 150 of maintenance system 100 implemented by maintenance process 200.

[0113] The drone 150 comprises a chassis 510 equipped with four arms extending in a main plane of the chassis 510. At each end of the arms 515, there is a motorized propeller 520, each rotating in a plane parallel to the main plane 511 of the chassis 510. The four propellers 520, arranged opposite a face called the upper face of the drone 150, thus provide, when their operation, a lift force for the drone 150.

[0114] The drone 150 has on a face opposite to the upper face, an image acquisition device 530, in this case a camera, whose optical axis 531 is here perpendicular to the main plane 511 of the chassis 510. The optical axis 531 is directed in such a way as to allow the camera 530 to acquire aerial images of the terrain 110 taken substantially vertically.

[0115] The drone 150 also includes a 540 battery for electrical energy storage as well as a 545 induction charging device.

[0116] It should be noted that the drone 150 can move automatically or be piloted by an operator.

Claims

1. Method (200) of maintaining an agricultural field (110), the method comprising the steps of: - generating (220), by a computing device (160) comprising a processor (161) and a computer memory (162), a map of the agricultural field from images previously acquired of the agricultural field and / or from geolocation data of an agricultural tool that has previously traversed the agricultural field, a plurality of crop zones (120) each comprising a plurality of parallel crop rows (130) being delimited in the generated map; - allocating (230) to an autonomous vehicle (140) a list of crop zones to be treated based on an initial position and a final position of the autonomous vehicle and the position of the crop zones to be treated; - determining (240) a movement plan for the autonomous vehicle from the list of crop zones allocated to said vehicle, respecting the orientation of the rows of the crop zones to be treated; - moving (250) the autonomous vehicle according to the previously established movement plan; the method also comprising a step (260) of generating, by the computing device, a diagnosis of the agronomic pressure within all or part of the crop zones from images previously acquired of the agricultural field covering each crop zone, the method being characterized in that the diagnosis of the agronomic pressure is a function of the nature of the weed(s) present in the crop zone, as well as their number and / or their development stage, the list of crop zones to be treated by the autonomous vehicle being ordered according to the agronomic pressure associated with each crop zone.

2. Method according to the preceding claim, wherein the step of generating a diagnosis of the agronomic pressure comprises a sub-step (261) of automatically recognizing the nature of a weed present in an image by means of a machine learning method previously trained on a database of images of identified weeds.

3. Method according to any one of the preceding claims, wherein the diagnosis of the agronomic pressure in a crop zone is also a function of a comparison between the growth of a detected weed and the growth of a cultivated plant in said crop zone.

4. Method according to any one of the preceding claims, wherein the step of generating a diagnosis of the agronomic pressure comprises a sub-step (264) of predicting an evolution of the development stage of a weed at a given point in time based on at least one measured and / or predicted meteorological data item.

5. Method according to any one of the preceding claims, implementing a plurality of autonomous vehicles, wherein the list of crop zones allocated to an autonomous vehicle takes into account the order of the crop zones to be treated according to the agronomic pressure associated with each crop zone and the position of each autonomous vehicle at a given point in time.

6. Method according to any one of the preceding claims, wherein the list of crop zones allocated to an autonomous vehicle is drawn up taking into account the autonomy of said vehicle at a given point in time relative to a recharging station (145) of said vehicle.

7. Method according to any one of the preceding claims, also comprising a step (222) of recognizing the direction of the rows of a crop zone and of determining (224) a spacing between two rows from at least one acquired image covering said crop zone and / or from a recording of a plurality of positions of a seeding tool used during a prior seeding phase, the seeding tool comprising a geolocation device.

8. Method according to the preceding claim, also comprising a step of automatically delimiting a crop zone by analyzing the direction of the previously recognized rows.

9. Method according to any one of the preceding claims, also comprising a prior step (210) of acquiring images of the agricultural field by at least one autonomous image acquisition device (150, 151).

10. Method according to the preceding claim, wherein the autonomous image acquisition device is an aerodyne (150) flying over the agricultural field at a maximum altitude of approximately one hundred meters above the surface of the agricultural field, for generating the map of the agricultural field, and / or at a maximum altitude of approximately three meters for generating the agronomic pressure of the weeds.

11. Method according to any one of the preceding claims, comprising a step (270) of determining at least one zone on the map in which movement of an autonomous vehicle is prohibited.

12. Maintenance method according to any one of the preceding claims, also comprising a step of associating a crop type with at least one crop zone of the agricultural field.

13. A system for maintaining an agricultural field, characterized in that it comprises a computing device comprising a computer memory storing instructions of a maintenance method according to any one of the preceding claims, and at least one autonomous vehicle provided with a set of mechanical tools configured to perform maintenance operations on the agricultural field.

14. Maintenance system according to the preceding claim, also comprising at least one autonomous image acquisition device.

15. Maintenance system according to any one of claims 13 to 14, also comprising a recharging station for an autonomous vehicle and / or an autonomous image acquisition device.