Accurate application of liquids to a target object in an agricultural field

Abstract

The present invention relates to the technical field of precision agriculture. The invention relates to a system and a method for accurate application of a liquid to a target object in an agricultural field.

Description

The present disclosure relates to the technical field of precision farming. The subjects of the present invention are a device and a method for the precise application of a liquid to a target object in an agricultural field. Precision farming is the term used to describe methods for the spatially differentiated and targeted management of agricultural land areas. One task in precision farming is to recognize objects such as plants, pests or accompanying plants in a digitally recorded image, and then make a diagnosis and subsequently carry out an action (for example treatment). One example of such a task is ultra-high-precision spraying (UHPS), in which a tractor is driven over a planted field and successive recordings of the agricultural field are taken using the attached spraying device. Image processing algorithms are used for example to identify accompanying plants in the recordings, and these accompanying plants are then sprayed with a herbicide in a treatment step by way of a spray bar. This type of task means having to track target objects, such as accompanying plants, once they have been recognized, from the current position in the recording recorded during the journey until the spray bar is located just in front of a target object to be treated as a result of the tractor movement, in order then to open one or more nozzles at the correct time, for example to spray a herbicide and to close said one or more nozzles again after the required opening time. One challenge with UHPS is that of scheduling and activating the nozzles at the correct time and at the correct location, because it is necessary to ensure that the precision of the application of the liquid to the target objects on the agricultural field is aligned so as to hit the target object as accurately as possible. For a UHPS with an accuracy of up to 4 cm x 4 cm, highly precise nozzle activation schedules are necessary because, if a tractor is driving for example at a speed of 10 km / h, it already covers a distance (per unit of time) of approximately 278 cm / second = approximately 2.8 mm / millisecond. If it is desired to hit small target objects with diameters in the cm range, the opening times must therefore be calculated accurately to the millisecond. These and other challenges are solved by the subjects of the independent claims. Preferred embodiments are found in the dependent claims, the present description and the drawings. A first subject of the present invention is a device for applying a liquid to at least one part of a target object in an agricultural field, comprising - a computing and control unit, - at least one storage container for holding the liquid, - at least one nozzle, - means for conveying the liquid from the at least one storage container in the direction of the at least one nozzle, - at least one sensor unit that is able to be aligned with a target object in the direction of the agricultural field, and is configured to generate at least one recording, - at least one position determination unit, wherein the computing and control unit is configured (a) to cause the at least one sensor unit, at the time t1, to generate at least one recording and to receive this at least one recording, wherein the computing and control unit is configured (b) to cause the at least one position determination unit, at the time t1, to determine the position of the at least one sensor unit and to receive these position data, wherein the computing and control unit is configured (c) to determine at least one part of a target object - if present - in the at least one recording by carrying out image processing and to georeference at least one part of this target object using the position data of the at least one sensor unit from the time t1 in order to determine the distance S1 from the at least one nozzle to the at least one part of the target object in the recording from the time t1, wherein the computing and control unit is configured (d) to cause the at least one position determination unit, at the time t2, to determine the position of the at least one nozzle and to receive these position data, wherein the time t2 is a time later than t1, wherein the computing and control unit is configured (e) to determine the distance S2 from the at least one nozzle to the at least one part of the target object based on the position data of the at least one nozzle from the time t2 and the position data of the at least one georeferenced part of the target object in the recording from the time t1, wherein the computing and control unit is configured (f) to determine the speed v2 and direction r2 of the at least one nozzle from the time t2 at least partially on the basis of the position data of the at least one nozzle from the time t1 and from the time t2, wherein the computing and control unit is configured (g) to schedule the activation time of the spraying operation for the at least one nozzle onto at least one part of the target object at least partially on the basis of the distance S2 from the at least one nozzle to the at least one part of the target object from the time t2 and the speed v2 and direction r2 of the at least one nozzle from the time t2. Another subject of the present disclosure is a method for applying a liquid to at least one part of a target object in an agricultural field, comprising steps (1) to (3): in step (1), moving a device for applying a liquid to at least one part of a target object in an agricultural field, in step (2), conveying a liquid from at least one storage container in the direction of at least one nozzle during the movement, in step (3a), causing at least one sensor unit, at the time t1, to generate at least one recording and forward this at least one recording to the computing and control unit, in step (3b), causing at least one position determination unit, at the time t1, to determine the position of the at least one sensor unit and forward these position data to the computing and control unit, in step (3c), determining at least one part of a target object - if present - in the at least one recording by carrying out image processing and georeferencing the at least one part of the target object using the position data of the at least one sensor unit from the time t1 in order to determine the distance S1 from the at least one nozzle to the at least one part of the target object in the recording from the time t1, in step (3d), causing the at least one position determination unit, at the time t2, to determine the position of the at least one nozzle and forward these position data to the computing and control unit, wherein the time t2 is a time later than t1, in step (3e), determining the distance S2 from the at least one nozzle to the at least one part of the target object based on the position data of the at least one nozzle from the time t2 and the position data of the at least one georeferenced part of the target object in the recording from the time t1, in step (3f), determining the speed v2 and direction r2 of the at least one nozzle from the time t2 at least partially on the basis of the position data of the at least one nozzle from the time t1 and from the time t2, in step (3g), scheduling the activation time of the spraying operation for the at least one nozzle onto at least one part of the target object at least partially on the basis of the distance S2 from the at least one nozzle to the at least one part of the target object from the time t2 and the speed v2 and direction r2 of the at least one nozzle from the time t2. The invention will be explained in more detail below without distinguishing between the subjects of the invention (device, method). The following explanations are intended rather to apply analogously to all subjects of the invention, regardless of the context (device, method) in which they are provided. The present invention discloses a device for applying a liquid to at least one part of a target object in an agricultural field. In other words, the liquid may be dispensed onto at least one part of a target object in an agricultural field. As an alternative, the liquid may be dispensed substantially onto the entire target object. It is also possible for the liquid to be applied to multiple target objects. The liquid may be water or an aqueous solution or suspension. The aqueous solution or suspension may contain one or more nutrients and / or one or more plant protection agents and / or one or more agents for treating seeds. The term “nutrients” is understood to mean those inorganic and organic compounds from which plants are able to extract the elements from which their bodies are formed. These elements themselves are often also referred to as nutrients. These are mostly simple inorganic compounds such as nitrate (NO3-), phosphate (PO43-) and potassium (K+). In addition to the core elements of organic matter (C, O, H, N and P), K, S, Ca, Mg, Mo, Cu, Zn, Fe, B, Mn, Cl in higher plants, Co, Ni are also vital. Different compounds may be present for the individual nutrients; for example, nitrogen may be supplied as nitrate, ammonium or amino acid. The term “plant protection agent” is understood to mean an agent used to protect plants or plant products from harmful organisms or to prevent the effect thereof, to destroy unwanted plants or parts of plants, to inhibit unwanted growth of plants or to prevent such growth, and / or to influence the life processes of plants in a way different from nutrients (for example growth regulators). Examples of plant protection agents are herbicides, fungicides and other pesticides (for example insecticides). Growth regulators are used for example to increase stability in cereals by shortening the culm length (internode shorteners), improve the root development of seedlings, reduce the plant height by stunting in horticulture, or prevent the germination of potatoes. Growth regulators may be for example phytohormones or their synthetic analogs. The term “target object” describes one or more plants, one or more regions of a field, pests or other objects. Preferably, the term “target object” describes one or more plants. In another preferred embodiment, the target object is located on the ground or close to the ground. The term “at least one part of a target object” preferably describes a part (for example a leaf, leaf portion, stem, stem portion, etc.) of one or more plants, or one or more pests. In one preferred embodiment, the target objects are individual crop plants or (individual) parts of individual crop plants or individual groups of crop plants. In another preferred embodiment, the target objects are individual seeds or groups of seeds that are sown and / or have been sown in a field for crop plants. In another preferred embodiment, the target objects are one or more accompanying plants or (individual) parts of individual accompanying plants or individual groups of accompanying plants. In another embodiment, the target objects are plant components that are infested with pests. Such pests may be animal pests, fungi, viruses or bacteria. The term “crop plant” is understood to mean a plant that is grown specifically as a useful plant or ornamental plant by human intervention. The term “accompanying plants” (often also referred to as weed / weeds) is understood to mean plants of the spontaneous accompanying vegetation (segetal flora) in stands of crop plants, on grassland, or in gardens that are not being specifically grown there and develop, for example, from the seed potential of the soil or from being blown in. The term “agricultural field” is understood to mean a spatially delimitable region of the surface of the Earth being used for agriculture, in that such a field is planted with crop plants that are possibly supplied with nutrients and harvested. The term “computing and control unit” means components of a computer or processor for carrying out arithmetic and logic operations and for controlling sequences of instructions and operations. In modern computers, the computing unit and the control unit are often closely interconnected and together form what is known as the “CPU” (central processing unit). These two units work together in order to enable instructions to be processed and executed in a computer. It is also possible for at least certain functions to be carried out via cloud-based computing. The liquid is in a storage container before application. There may be multiple storage containers. Multiple (different) liquids may be applied. The liquid is applied to at least one part of the target object via one or more nozzles. In one embodiment, the at least one nozzle is a component of an inkjet print head. In the embodiment, therefore, the technology used in inkjet printers is used to apply the liquid to at least one part of the target object. The use of inkjet printer technology in agriculture is described for example in M.- Idbella et al.: Structure, Functionality, Compatibility with Pesticides and Beneficial Microbes, and Potential Applications of a New Delivery System Based on Ink-Jet Technology, Sensors 2023, 23(6), 3053. In one preferred embodiment, use is made of multiple nozzles that are preferably part of a bar or a strip (referred to hereinafter as “spray bar”). In another preferred variant, the device according to the invention comprises a multiplicity of nozzles. The term “multiplicity” preferably means more than ten. The nozzles are preferably arranged such that each nozzle applies liquid in a region with a maximum lateral extent of less than 20 cm, preferably less than 10 cm, most preferably less than 5 cm. The multiplicity of nozzles may be arranged for example side by side along a spray bar that extends transversely (for example at an angle of 90°) to the direction of movement of the device. At least one sensor unit (for example a camera) may be assigned to each nozzle. In such an embodiment, there is the possibility of a target object leaving the field of view of the at least one sensor unit during cornering, and the adjacent sensor unit being activated as a result. For application, the liquid is conveyed with conveying means from the at least one storage container in the direction of the at least one nozzle. For example, a pump may be used to convey the liquid. The device according to the invention comprises at least one sensor unit. A sensor unit comprises at least one sensor. A “sensor” is a technical component that is able to capture certain physical and / or chemical properties and / or the material nature of its surroundings in a qualitative manner, or in a quantitative manner as a measured variable. These variables are captured by way of physical or chemical effects and transformed into a signal that is able to be processed further, usually an electrical or optical signal. A sensor unit may contain means for processing signals provided by the at least one sensor. A sensor unit may comprise means for transmitting and / or forwarding signals and / or information (for example to the control unit). The one or more sensor units may be a component of the system and / or be connected thereto via a communication connection (for example via radio). The one or more sensor units may be configured to transmit one or more signals to the computing and control unit continuously or at defined time intervals or upon the occurrence of defined events, on the basis of which signals the computing and control unit carries out the further steps, for example for scheduling the activation time of the spraying operation for the at least one nozzle. The at least one sensor unit, which is able to be aligned with a target object in the direction of the agricultural field, preferably comprises one or more cameras. In other words, the term “recording” preferably describes an “image recording”. The camera axis of the at least one camera, as a component of the device according to the invention, may preferably be aligned substantially perpendicular (vertical) to the ground (that is to say the at least one camera points “down” or “straight ahead” in relation to the ground). The at least one camera is preferably located in front of the at least one spray unit (in the direction of travel). This distance is preferably between 30 cm and 100 cm, even more preferably between 40 and 60 cm. The camera used according to the invention may comprise an image sensor and optical elements. The image sensor is a device for recording twodimensional images from light by electrical means. This usually involves semiconductor-based image sensors, such as for example CCD (CCD = charge-coupled device) or CMOS (CMOS = complementary metal-oxide semiconductor) sensors. The optical elements (lenses, stops and the like) serve for maximum sharpness of imaging of an object on the image sensor. The computing and control unit may cause the camera, at defined time intervals or continuously, to generate image recordings of at least one part of a target object (list point (a) for the device according to the invention and step (3a) in the method according to the invention). The at least one part of a target object may be for example a crop plant, an accompanying plant and / or a pest. The computing and control unit may in this case be the computing and control unit of the device according to the invention or a separate computing and control unit. The computing and control unit of the device according to the invention is preferably used for this purpose. Each recording is marked with a timestamp in order to document the exact time of recording of the image. It may also be the case that no part of a target object is present in a single recording and, for example, only the ground or the agricultural field, without a part of a target object, is captured. In addition to the above-described at least one sensor unit, the configuration according to the invention of the device also comprises at least one position determination unit. The at least one position determination unit may also comprise one or more sensors. The at least one position determination unit of the device is configured such that, at the respective time at which the at least one sensor unit generates an image recording, the position data of the at least one sensor unit (of the camera and preferably the position data of the camera lens center) are captured, and this information is forwarded to the computing and control unit of the device (list point (b) for the device and step (3b) in the method). The at least one position determination unit may comprise for example a receiver of a satellite navigation system (GNSS, Global Navigation Satellite System), colloquially also referred to as a GPS receiver. The Global Positioning System (abbreviated to: GPS), officially NAVSTAR GPS, is an example of a global satellite navigation system for position determination; other examples are GLONASS, Galileo and Beidou. The satellites of such a satellite navigation system communicate their exact position and time via radio codes. To determine position, a receiver (the “GPS receiver”) must receive the signals from at least four satellites at the same time. In the receiver, the pseudosignal propagation times are measured and used to determine the current position. In one preferred embodiment, at least two GNSS receivers are used. Preferably, the position of the at least one nozzle of the device is determined using a moving baseline differential GNSS installation (DGPS), which enables a position determination accuracy of less than 10 mm and achieves update rates of between 8 and 100 Hz. In the moving baseline DGPS installation, two GNSS receivers with connected GNSS sensors are installed on the device according to the invention at a fixed distance. One of them takes on the role of a “moving baseline”, and the other takes on the role of the “rover”. By way of example, these two components may be installed on the opposite ends of the spray bar. This technology may be used to generate DGPS correction data while traveling, which enable the accuracy discussed above. Since the distance from the at least one sensor unit (preferably the camera lenses) and the at least one nozzle to the GNSS receivers on the device according to the invention is fixed, their positions are able to be determined very quickly on the basis of the position of the GNSS receivers with only a few computing operations. In other words, as soon as the position determination unit has determined the position of the at least one sensor unit, the position of the at least one nozzle is also known. Analogously, when determining the position of the at least one nozzle at a specific time, the position of the at least one sensor unit at the same time is likewise known. The computing and control unit of the device (or of the method) receives the at least one recording from the time t1 from the sensor unit and the position data from the time t1 from the at least one position determination unit. The computing and control unit is configured to process the recordings by carrying out image processing in order to recognize at least one part of a target object - if present - in the recordings, for example a defined plant (for example an accompanying plant), a part of a defined plant (for example a leaf), a pest, a pest-infested plant and / or another object (list point (c) for the device and step (3c) in the method). Methods and devices for recognizing objects in image recordings are described in a variety of ways in the prior art (see for example: WO2020120802A1, WO2020120804A1, WO2020229585A1). Various methods may be used with regard to image processing for the purpose of object recognition, such as for example feature extraction using deep learning models such as neural networks, support vector machines (SVM), decision trees, etc. Further possible methods that may be used are color-assisted recognition, texture analysis, shapebased recognition, spectral analysis, color histogram analysis, the histogram of oriented gradients (HOG) method, morphological operations or a combination of these methods. In one preferred embodiment, the computing and control unit is configured to determine the at least one part of the target object in the recording using machine learning. The models used in this image processing are preferably based on deep learning techniques, such as neural networks (for example a convolutional neural network (CNN)). Such a model is trained with annotated images in order to learn how to distinguish between different classes, such as for example an accompanying plant and other plants. After training, such a model may be used with new recordings. A recording received from the at least one sensor unit may be preprocessed for use with the model. Preprocessing comprises common methods such as normalization or scaling steps, noise removal, changing contrasts or other methods for improving data quality, and converting the recording to the input format of the model. The model assigns each pixel in the image to a specific class, for example whether the pixel belongs to a target object, a non-target object or to the background. This method is also referred to as semantic segmentation. Semantic segmentation with regard to the identification of plants in a field is described in the prior art (see for example WO19226869 A1). As part of the postprocessing of the classified data at the pixel level of the model, it is possible to carry out transformations or analysis operations such as the generation of pixel masks (for example binary representation of pixels that belong or do not belong to the recognized target object), the generation of bounding boxes (rectangular frame around a recognized target object), the determination of the area centroid (center of the area of the recognized target object, which may be used as a representative point for the recognized target object), or the generation of an outline polygon (a series of points describing the contour of the recognized target object). In addition to semantic segmentation, however, it is also possible to use object detection algorithms such as for example Yolo, SSD, Faster-RCNN, etc. With these algorithms, it is possible - instead of the intermediate path via the pixel masks - to directly generate “bounding boxes” for the recognized target object. Preferably, the entire target object is recognized by the image analysis. However, it may be the case that only a part of a target object is present in the recording, and only this part is able to be recognized. The abovementioned image processing methods, including the preprocessing and postprocessing steps, may also be carried out if only a part of a target object has been recognized in the recording. After the image processing, at least one part of the recognized target object is georeferenced with the position data of the at least one sensor unit from the time t1 (list point (c) for the device and step (3c) in the method). In one embodiment of the device according to the invention, at least one pixel coordinate of the at least one part of the recognized target object is converted into position data on the agricultural field. In another preferred embodiment, this one pixel coordinate is the area centroid of the recognized target object. In another preferred embodiment, multiple pixel coordinates of the recognized target object are georeferenced, such as for example all points of the pixel mask that have been assigned to the recognized target object or all points within the bounding box or the outline polygon of the recognized target object. Based on the one or more georeferenced points, and on the knowledge of optical parameters of the camera (for example the aperture angle of the camera, the vertical and horizontal resolution of the camera) and on the knowledge of the distance between the camera (preferably the camera lens or the camera lens center) and the ground of the agricultural field, it is possible to determine the distance S1 from the at least one nozzle to the at least one part of the target object in the recording from the time t1. In one preferred variant of the invention, this calculation is carried out using the following formulas: s = —^-»-  s = — d y   2Htan^ , *x   2Htan^ , (1)                (2) wherein, in formula (1), Sy is the number of pixels per unit of height on the image sensor of the camera (vertical resolution per unit height), y is the vertical resolution of the camera, H is the distance between the camera lens and the ground, and ay is the vertical aperture angle of the camera, wherein, in formula (2), Sx is the number of pixels per unit of width on the image sensor of the camera (horizontal resolution per unit width), x is the horizontal resolution of the camera, H is the distance between the camera lens and the ground, and ax is the horizontal aperture angle of the camera, wherein, proceeding from the pixel coordinate of the at least one recording, the position data of the at least one pixel coordinate of the at least one part of the target object on the agricultural field are calculated via the respective reciprocal of Sy and Sx. The computing and control unit is furthermore configured to cause the at least one position determination unit, at the time t2, to determine the position of the at least one nozzle and to receive these position data, wherein the time t2 is a time later than t1 (list point (d) for the device and step (3d) in the method). ( In one preferred embodiment, the device is in motion in this case, meaning that the position data of the at least one nozzle at the time t1 and time t2 are not the same. In another preferred variant, the device is moving toward the target object. The computing and control unit is furthermore configured to determine the distance S2 from the at least one nozzle to the at least one part of the target object based on the position data of the at least one nozzle from the time t2 and the position data of the at least one georeferenced part of the target object in the recording from the time t1 (list point (e) for the device and step (3d) in the method). The computing and control unit is furthermore configured to determine the speed v2 and direction r2 of the at least one nozzle from the time t2 at least partially on the basis of the position data of the at least one nozzle from the time t1 and from the time t2 (list point (f) for the device and step (3f) in the method). It should be taken into account here that, when multiple nozzles are present, each of the nozzles may have a different speed when the device is cornering. The speed v2 of the at least one nozzle may be determined based on the existing knowledge of the distance covered by the device between t1 and t2. The direction r2 of the at least one nozzle from the time t2 may be determined based on the knowledge of the two different position data of the at least one nozzle from the time t1 and from the time t2. The direction between two coordinate points is usually expressed as an azimuth or course. The azimuth is the angle in degrees, measured clockwise from the North. The computing and control unit is furthermore configured to schedule the activation time of the spraying operation for the at least one nozzle onto at least one part of the target object at least partially on the basis of the distance S2 from the at least one nozzle to the at least one part of the target object from the time t2 and the speed v2 and direction r2 of the at least one nozzle from the time t2 (list point (g) for the device and step (3g) in the method). In one preferred embodiment, the computing and control unit is furthermore configured to activate the at least one nozzle at least partially on the basis of the scheduled activation time of the spraying operation and to apply the liquid to at least one part of a target object (list point (h) for the device and step (3h) in the method). In another preferred embodiment, the computing and control unit is configured to carry out the steps mentioned in list points (a) to (f) (steps 3a to 3f of the method) and the calculation of the respective parameters Si, ti, vi and ri iteratively (i). The activation time of the spraying operation for the at least one nozzle onto at least one part of a target object in the step mentioned in list point (g) may possibly be adapted by each iterative step. Preferably, in the case of an iterative implementation, the target object is recognized completely by the image analysis. It is thereby possible to schedule the activation time of the spraying operation for the complete target object and also to carry out the spraying operation accordingly for the complete target object. In one preferred embodiment, measured data (recordings and - if present - measured position determination data) from multiple iteration cycles are used for the calculations (list point (c), (e), (f) and (g) and (c’), (e’), (f’) and (g’) for the device and step (3c), (3e), (3f) and (3g), and (3c’), (3e’), (3f’) and (3g’) in the method). Preferably, measured data from between 2 and 100, more preferably between 3 and 50 and even more preferably between 3 and 10 iteration cycles are used. The larger database makes it possible to stabilize the measured data and analysis data. At the same time, limiting the iteration cycles prevents data that are too old and have become irrelevant in the meantime from being used for the calculation steps. In another preferred variant of the present device, at least two position determination units are used. By way of example, it is possible to use GNSS (Global Navigation Satellite System) receivers, position determination units based on visual odometry technology, radar, trilateration, inertial navigation systems (for example gyroscopes, accelerometers), magnetometers, radio navigation systems etc. as position determination units. The first position determination unit preferably comprises at least one GNSS (Global Navigation Satellite System) receiver, and the second position determination unit is based on visual odometry technology and / or radar. Generally speaking, position determination accuracy and reliability is able to be improved by sensor fusion, that is to say by using multiple position determination units. Combining multiple sensors makes it possible to compensate for errors and inaccuracies of individual sensors. By way of example, GPS, visual odometry and / or radar may be used together simultaneously to achieve more precise position determination. The sensor fusion may be carried out using algorithms and techniques such as the Kalman filter or other fused estimation methods in order to combine the data from the various sensors and obtain an optimum position estimate. In another preferred embodiment, the computing and control unit is configured to check, before and / or during position data determination using the GNSS receiver, whether the signal strength of the GNSS receiver reaches or exceeds a predefined threshold value level. If the predefined threshold value level is not reached, the position determination is determined by the position determination unit comprising a visual odometry technology. The definition of a suitable threshold value level for a GNSS receiver depends on various factors, such as for example the sensitivity of the GNSS receiver. A possible threshold value could be for example in the range from -130 dBm (decibel milliwatts) to -150 dBm. In one preferred variant, the signal strength is measured each time before the steps described in list point (b) and (d) for the device (step (3b) and (3d) in the method) are carried out, and a check is carried out to determine whether the signal strength of the GNSS receiver reaches or exceeds the predefined threshold value level. This is preferably carried out only each time before the step described in list point (b) for the device (step (3b) in the method) is carried out. If the position data determination was carried out using visual odometry, for example at the time ti-1,1, and it is identified, at the time ti,1, that the signal strength of the GNSS receiver has now reached the threshold value level, the position data determination for the time ti,1 is carried out by the GNSS receiver. In the next iteration ti+1,1, another check is carried out to determine whether the signal strength is sufficient to be able to use the GNSS receiver, or whether it is necessary to change back to visual odometry. The changeover between the two position determination units therefore preferably takes place as seamlessly as possible in order to be able to ensure uninterrupted position determination. In this case, an attempt is preferably first made to carry out the position determination via the at least one GNSS receiver. Instead of using visual odometry, such a method may also be carried out using another position determination unit technology, such as for example radar (see above). The method will be explained in more detail below with reference to visual odometry. Visual odometry is based on tracking the movement and position of an object (preferably at least one part of a target object) in the field based on the recordings from the at least one camera. By carrying out image analysis on successive recordings, the computing and control unit identifies visual features that are stable and unambiguously identifiable. These are preferably characteristic points of at least one part of a target object. The image analysis tracks the movement of the identified features between successive recordings. This may be achieved using methods such as optical flow or feature matching algorithms. The movement of the object (preferably at least one part of a target object) may be estimated based on the change in the position of the identified features. This estimate may comprise translation (movement in the x, y and z direction) and rotation (rotation about the axes). It is also possible to integrate simultaneous localization and mapping (SLAM), in which a relative coordinate system (map of the environment) is also created in addition to the direct position determination. Visual odometry technology may comprise further sensor units in order to improve position determination accuracy using this technology by way of sensor fusion. Such further sensor units comprise for example inertial sensors such as gyroscopes and accelerometers. Precise calibration of the at least one camera and possible further sensor units is necessary in order to ensure accurate position determination. In one embodiment (when using visual odometry for position determination), the computing and control unit is configured to cause the at least one sensor unit to generate at least one recording at each of the times ti-1,1 and ti,1, wherein the time interval between ti-1,1 and ti,1 is chosen to be so short that the recordings overlap (step (a’)), wherein the computing and control unit is configured, at the time ti,2 - and as a departure from the steps described in list points (c) and (e) to (g) - in the alternative steps (c’), (e’) and (f’) - to determine at least one of the data selected from the group of distance Si,2 from the at least one nozzle to the at least one part of the target object from the time ti,2, speed vi,1 of the at least one nozzle from the time ti,1, direction ri,1 of the at least one nozzle from the time ti,1 at least partially on the basis of these recordings. The computing and control unit is furthermore configured (in alternative step (g’)) to schedule the activation time of the spraying operation for the at least one nozzle onto at least one part of the target object at least partially on the basis of the distance Si,2 from the at least one nozzle to the at least one part of the target object from the time ti,2 and the speed vi,1 and direction ri,1 of the at least one nozzle from the time ti,1. In optional step (h’), the computing and control unit is configured to activate the at least one nozzle at least partially on the basis of the scheduled activation time of the spraying operation in step (g’) and to apply the liquid to the at least one part of the target object. In one preferred embodiment of the device, the computing and control unit is configured (in alternative step (c’)) to recognize matches between the recordings from the times ti-1,1 and ti,1 and (in alternative step (e’)) to calculate the relative image movement and the angle of rotation on the basis of these data. In another preferred variant, the computing and control unit is configured to recognize at least one matching pixel movement (preferably at least one pixel of at least one part of a target object) of the current (i) image in comparison with the image from the previous iteration (i-1). In computer vision, this procedure is called image registration. The relative movement and the rotation of this at least one matching pixel may be determined for example using a transformation matrix. In alternative step (f’), the computing and control unit is configured to determine at least one of the data selected from the group of: the distance Si,2 from the at least one nozzle to the at least one part of the target object from the time ti,2, speed vi,1 of the at least one nozzle from the time ti,1, direction ri,1 of the at least one nozzle from the time ti,1, on the basis of the determined relative movement and the angle of rotation. The distance Si,2 at the time ti,2 may be determined for example via the intermediate step using formula (3): Sc = Vi,1 • (ti,2 - ti,1) (13) where Sc is the position of the at least one nozzle at the time ti,2, and vi,1 is the speed of the at least one nozzle at the time ti,1. The time ti,1 describes the time of recording by the camera (list point (a’i) for the device and step (3a’i) in the method) in iteration i. The time ti,2 is the time after the image processing (after the steps according to list point (c’i), (e’i) for the device and (3c’i), (3e’i) in the method) in iteration i (that is to say a time in step f’ in iteration i). The distance Si,2 may be determined via Sc and the calculated angle of rotation. The speed vi,1 from the time ti,1 may be determined for example using formula (4): 8x,y,z -1.1 (14) where vu is the speed of the at least one nozzle at the time ti,i and dx,y,z is the relative image movement (determined by image analysis). The denominator in formula (3) contains the time difference between the two recordings from the time ti,1 and ti-1,1. The time ti,1 has the same meaning as described in formula (3). The time ti-1,1 describes the time of recording by the camera (list point (a’i-1) for the device and step (3a’i-1) in the method) in iteration i-1. The time interval between ti-1,1 and ti,1 is chosen to be so short that the recordings overlap. The direction ri,1 may be determined for example using a transformation matrix. The times t1 and t2 described above concern the same iteration. In the method, the above-described alternative steps (a’), (c’), (e’), (f’), (g’), (h’) in the third step are referred to as (3a’), (3c’), (3e’), (3f’), (3g’), (3h’). Another embodiment of the invention relates to the device according to the invention, comprising at least one further sensor unit able to be aligned with a target object in the direction of the agricultural field. The computing and control unit is configured to cause this at least one further sensor unit to capture a 3D point cloud at the time t1, and the computing and control unit is furthermore configured to receive this 3D point cloud. Suitable sensor units for capturing a 3D point cloud are for example LIDAR sensors and 3D cameras. If multiple 2D cameras are present, a 3D point cloud may be created at the time t1 through photogrammetric analyses (analyses of recordings from different perspectives). Techniques such as for example structure from motion (SfM) and triangulation may be used for this purpose. LIDAR sensors emit laser beams and measure how long it takes for the reflected light to return. This makes it possible to create 3D point clouds comprising accurate distance data. In addition to pixel data, 3D cameras (for example Kinect, RealSense or others) also acquire depth information, from which a 3D point cloud is able to be generated by conversion into 3D coordinates. A 3D point cloud is a data structure representing a collection of three-dimensional points in space. Each point in the point cloud is described by x, y and z coordinates, and may also contain additional information such as color or intensity. In one preferred embodiment of the invention, the control and computing unit is configured (in (step (c) or (c’) of the device and step (3c) or (3c’) of the method) to project the at least one pixel coordinate of at least one part of a target object on the agricultural field onto the 3D point cloud data. Preferably, multiple pixels of the at least one part of the target object (for example corner pixels of the bounding boxes or of an outline polygon, etc.) are projected onto the 3D point cloud data. In another preferred embodiment of the invention, the control and computing unit is configured to determine the height of the target object on the agricultural field using the 3D point cloud, wherein the distance between the vertically lowest point of the target object and the vertically highest point of the target object in the 3D point cloud is calculated. In this case, image analysis of at least one complete target object is preferred. In another preferred embodiment, the computing and control unit is configured (list point (g) or (g’) for the device and step (3g) or (3g’) in the method) to determine the activation time of the spraying operation for the at least one nozzle onto at least one part of the target object at least partially additionally also on the basis of the data of the 3D point cloud at the time t1. Preferably, the calculated height of the target object on the agricultural field is also taken into account here (since the time of flight of the liquid is shortened or lengthened depending of the height of the target object). In another embodiment of the invention, the computing and control unit is configured (list point (g) or (g’) for the device and step (3g) or (3g’) in the method) to at least partially take into account the latency of the at least one nozzle and the time of flight of the liquid from the at least one nozzle until it hits at least one part of the target object in order to determine the activation time of the spraying operation for the at least one nozzle onto at least one part of the target object. With the information concerning the distance S2 from the at least one nozzle to the at least one part of the target object from the time t2 and the speed v2 and direction r2 of the at least one nozzle from the time t2, it is possible to determine the nozzle opening time tA (that is to say the activation time for the nozzle opening, which, in this disclosure, has the same meaning as the activation time of the spraying operation). In one preferred embodiment of the invention, this nozzle opening time tA is also corrected with respect to the latency of the at least one nozzle LD (that is to say the nozzle opening latency) and the time of flight of the liquid from the at least one nozzle until it hits the target object tF. In other words, the scheduled nozzle opening time tP is able to be calculated using the following formula (5): tP = tA - LD - tF (5) where tP is the scheduled nozzle opening time, LD is the nozzle opening latency of the at least one nozzle, and tF is the time of flight of the liquid from the at least one nozzle until it hits the target object. LD results from various factors, such as for example nozzle type and nozzle design, liquid pressure, nature of the computing and control unit, nozzle size, viscosity of the liquid, maintenance state of the at least one nozzle and ambient temperature. tF results from the droplet speed of the liquid, which in turn depends in particular on liquid pressure and mechanical nozzle nature (and may be determined in the laboratory). Preferably, the calculated height of the target object on the agricultural field is also taken into account in order to determine tF. In another preferred variant of the invention, the computing and control unit is configured (in list point (g) and (g’) for the device or (3g) and (3g’) in the method) to schedule the duration of the nozzle opening. In other words, this is the duration of the open state of the at least one nozzle starting from the nozzle opening time. The nozzle opening duration preferably depends on the target object size: Depending on the vertical extent of the target object, the nozzle speed is used as a basis to calculate how long a nozzle has to be opened in order to completely treat the entire target object. Preferably, the calculated height of the target object on the agricultural field is therefore also taken into account in order to determine the nozzle opening duration. In a further preferred variant, the nozzle closing latency (which is often greater than the nozzle opening latency owing to pressure) is also taken into account when scheduling the nozzle opening duration. The device according to the invention may be part of a sprayer, for example an agricultural machine, a robot, or an aircraft (for example a drone) or be able to be connected thereto. Such a sprayer may move autonomously in or over a field or be controlled by a person. Embodiments of the present invention are: Figure 1 to figure 3 show, by way of example and schematically, one embodiment of the device according to the invention and of the method according to the invention. Figure 1 shows the device (10) for applying a liquid (F) to at least one part of a target object (P1) in an agricultural field (LF). The device (10) comprises a computing and control unit (11), at least one storage container (12) for holding the liquid (F), at least one nozzle (13), at least one position determination unit (14), means (15) for conveying the liquid (F) from the at least one storage container (12) in the direction of the at least one nozzle (13), and at least one sensor unit (16) that is aligned with a target object (P1) in the direction of the agricultural field (F). In figure 1, the device (10) is moving toward the target object (P1) (that is to say to the left), and the at least one sensor unit (16) is located in front of the at least one nozzle (13) in the direction of movement. The direction of the nozzle (13) is illustrated obliquely relative to the ground surface. However, the direction of the nozzle (13) may also be directed straight down relative to the ground surface. Figure 2 schematically shows steps (a) to (g) and optionally (h), coordinated by the computing and control unit (11). This corresponds to steps 3a to 3g and optionally 3h in the method according to the invention. In (a), the computing and control unit (11) is configured to cause the at least one sensor unit (16), at the time t1, to generate at least one recording and to receive this at least one recording (in figure 2, the recording is illustrated as a rectangle containing a plant). In (b), the computing and control unit (11) is configured to cause the at least one position determination unit (14), at the time t1, to determine the position of the at least one nozzle (13) and to receive these position data (in figure 2, this is illustrated schematically with a smaller version of the device (10) from figure 1 and a small crosshair for the acquisition of the position data at the time t1). In (c), the computing and control unit (11) is configured to determine at least one part of a target object (P1) in the at least one recording by carrying out image processing and to georeference this at least one target object (P1) using the position data of the at least one sensor unit (16) from the time t1 (in figure 2, this is again illustrated by the small crosshair on the recording) in order to determine the distance S1 from the at least one nozzle (13) to the at least one part of the target object (P1) in the recording from the time t1. In (d), the computing and control unit (11) is configured to cause the at least one position determination unit (14) to determine, at the time t2, the position of the at least one nozzle (13) and to receive these position data, wherein the time t2 is a time later than t1 (in figure 2, this is illustrated schematically in the same way as in step (b), but at the time t2). In (e), the computing and control unit (11) is configured to determine the distance S2 from the at least one nozzle (13) to the at least one part of the target object based on the position data of the at least one nozzle (13) from the time t2 and the position data of the at least one georeferenced part of the target object in the recording from the time t1. In (f), the computing and control unit (11) is configured to determine the speed v2 and direction r2 of the at least one nozzle (13) from the time t2 at least partially on the basis of the position data of the at least one nozzle (13) from the time t1 and from the time t2 (figure 2 schematically depicts a smaller version of the device (10), but the position of the at least one nozzle (13) is relevant). In (g), the computing and control unit (11) is configured to schedule the activation time of the spraying operation for the at least one nozzle (13) onto at least one part of the target object (P1) at least partially on the basis of the distance S2 from the at least one nozzle (13) to the at least one part of the target object (P1) from the time t2 and the speed v2 and direction r2 of the at least one nozzle (13) from the time t2 (figure 2 schematically depicts a schedule for this operation). In optional step (h), the computing and control unit (11) is configured to activate the at least one nozzle (13) at least partially on the basis of the scheduled activation time of the spraying operation and to apply the liquid (F) to at least one part of the target object (P1). Figure 3 schematically shows steps (a) to (g) and optionally (h), or (a’), (b), (c’), (d), (e’), (f’) and (g’) and optionally (h’), coordinated by the computing and control unit (11), in three successive iterative cycles i-1, i and i+1. This corresponds to steps 3a to 3g and optionally 3h, or (3a’), (3b), (3c’), (3d), (3e’), (3f’) and (3g’) and optionally (3h’) in the method according to the invention. In 5        iteration cycle i-1, steps (ai-1) to (gi-1) and optionally (hi-1) are carried out. The signal strength of the GNSS receiver as measured in step (bi-1) and (di-1) (when the steps start being carried out) has reached the predefined threshold value level, meaning that the position determination is able to be carried out using the GNSS receiver. In iteration cycle i, the signal strength of the GNSS receiver as measured in step (bi) (and optionally also in step (di)) does not reach the predefined threshold value 10         level. Therefore, steps (a’i), (c’i), (e’i), (f’i), (g’i) and optionally (h’i) are carried out in addition to these steps. In iteration cycle i+1, the signal strength of the GNSS receiver as measured in step (bi+1) and (di+1) reaches the threshold value level again. Steps (ai+1) to (gi+1) and optionally (hi+1) are therefore carried out.

Claims

1. A device (10) for applying a liquid (F) to at least one part of a target object (P1) in an agriculturalfield (LF), comprising- a computing and control unit (11),- at least one storage container (12) for holding the liquid (F),- at least one nozzle (13),- means (15) for conveying the liquid (F) from the at least one storage container (12) in the direction of the at least one nozzle (13),- at least one sensor unit (16) that is able to be aligned with a target object (P1) in the direction of the agricultural field (F), and is configured to generate at least one recording,- at least one position determination unit (14),wherein the computing and control unit (11) is configured (a) to cause the at least one sensor unit (16), at the time t1, to generate at least one recording and to receive this at least one recording,wherein the computing and control unit (11) is configured (b) to cause the at least one position determination unit (14), at the time t1, to determine the position of the at least one sensor unit (16) and to receive these position data,wherein the computing and control unit (11) is configured (c) to determine at least one part of a target object (P1) - if present - in the at least one recording by carrying out image processing and to georeference at least one part of this target object (P1) using the position data of the at least one sensor unit (16) from the time t1 in order to determine the distance S1 from the at least one nozzle (13) to the at least one part of the target object (P1) in the recording from the time t1,wherein the computing and control unit (11) is configured (d) to cause the at least one position determination unit (14), at the time t2, to determine the position of the at least one nozzle (13) and to receive these position data, wherein the time t2 is a time later than t1,wherein the computing and control unit (11) is configured (e) to determine the distance S2 from the at least one nozzle (13) to the at least one part of the target object based on the position data of the at least one nozzle (13) from the time t2 and the position data of the at least one georeferenced part of the target object in the recording from the time t1,wherein the computing and control unit (11) is configured (f) to determine the speed v2 and direction r2 of the at least one nozzle (13) from the time t2 at least partially on the basis of the position data of the at least one nozzle (13) from the time t1 and from the time t2,wherein the computing and control unit (11) is configured (g) to schedule the activation time of the spraying operation for the at least one nozzle (13) onto at least one part of the target object (P1) at least partially on the basis of the distance S2 from the at least one nozzle (13) to the at least one part of the target object (P1) from the time t2 and the speed v2 and direction r2 of the at least one nozzle (13) from the time t2.

2. The device (10) as claimed in the preceding claim, wherein the computing and control unit (11)is configured (h) to activate the at least one nozzle (13) at least partially on the basis of the scheduled activation time of the spraying operation and to apply the liquid (F) to the at least one part of the target object (P1).

3. The device (10) as claimed in either of the preceding claims, wherein the at least one sensor unit(16) comprises at least one camera and the camera axis of the at least one camera is preferably able to be aligned substantially perpendicular to the ground.

4. The device (10) as claimed in one of the preceding claims, wherein the computing and controlunit (11) is configured, in step (c), to determine the at least one part of a target object (P1) in the recording using machine learning.

5. The device (10) as claimed in claim 4, wherein the computing and control unit (11) isconfigured, in step (c), to determine the distance S1 from the at least one nozzle (13) to the at least one part of the target object (P1) in the recording from the time t1 such that at least one pixel coordinate of the at least one part of the target object is converted into position data on the agricultural field (LF).

6. The device (10) as claimed in claim 5, wherein the following formulas are used to calculate theposition data of the at least one pixel coordinate of the target object (P1) on the agricultural field (LF): s = —^-»-  s = —d y   2Htan^  , *x   2Htan^ ,(1)                (2)wherein, in formula (1), Sy is the number of pixels per unit of height on the image sensor of the camera (vertical resolution per unit height), y is the vertical resolution of the camera, H is the distance between the camera lens and the ground, and ay is the vertical aperture angle of the camera,wherein, in formula (2), Sx is the number of pixels per unit of width on the image sensor of the camera (horizontal resolution per unit width), x is the horizontal resolution of the camera, H is the distance between the camera lens and the ground, and ax is the horizontal aperture angle of the camera,wherein, proceeding from the pixel coordinate of the at least one recording, the position data of the at least one pixel coordinate of the at least one part of the target object (P1) on the agricultural field (LF) are calculated via the respective reciprocal of Sy and Sx.

7. The device (10) as claimed in one of the preceding claims, wherein the computing and controlunit (11) is configured to carry out the steps mentioned in list points (a) to (f) and the calculation of the respective parameters Si, ti, vi and ri iteratively (i), and the activation time of the spraying operation for the at least one nozzle (13) onto at least one part of the target object (P1) in the step mentioned in list point (g) may be adapted by each iterative step.

8. The device (10) as claimed in one of the preceding claims, wherein the at least one positiondetermination unit (14) comprises at least one GNSS (Global Navigation Satellite System) receiver, preferably at least two GNSS receivers, and even more preferably the at least one GNSS receiver comprises a moving baseline differential GPS installation.

9. The device (10) as claimed in either of claims 7 and 8, wherein at least two positiondetermination units (14) are preferably selected from the group of GNSS (Global Navigation Satellite System) receiver technology, position determination units based on visual odometry technology, radar, trilateration, inertial navigation systems, magnetometers and radio navigation systems and preferably, of these, one position determination unit comprises a GNSS (Global Navigation Satellite System)receiver, and the second position determination unit comprises a visual odometry technology and / or radar.

10. The device (10) as claimed in claim 9, wherein the computing and control unit (11) is configured to check, before and / or during position data determination using the GNSS receiver (14), whether the signal strength of the GNSS receiver reaches or exceeds a predefined threshold value level, and - if the predefined threshold value level is not reached - to determine the position determination from the position determination unit comprising a visual odometry technology.

11. The device (10) as claimed in claim 10, wherein, when determining position data on the basis of visual odometry technology, the computing and control unit (11) is configured to cause the at least one sensor unit (13) to generate at least one recording at each of the times ti-1,1 and ti,1, and the time interval between ti-1,1 and ti,1 is chosen to be so short that the recordings overlap (a’), wherein the computing and control unit (11) is configured, at the time ti,2 - and as a departure from the steps described in list points (c) and (e) to (g) - in (c’, e’ and f’) - to determine at least one of the data selected from the group of distance Si,2 from the at least one nozzle (13) to the at least one part of the target object (P1) at the time ti,2, speed vi,1 of the at least one nozzle (13) from the time ti,1, direction ri,1 of the at least one nozzle (13) from the time ti,1 at least partially on the basis of these recordings, and wherein the computing and control unit is configured (g’) to schedule the activation time of the spraying operation for the at least one nozzle onto at least one part of the target object at least partially on the basis of the distance Si,2 from the at least one nozzle to the at least one part of the target object from the time ti,2 and the speed vi,1 and direction ri,1 of the at least one nozzle from the time ti,1.

12. The device (10) as claimed in one of the preceding claims, wherein the device (10) comprises at least one further sensor unit (16b) able to be aligned with a target object (P1) in the direction of the agricultural field (LF), wherein the computing and control unit (11) is configured to cause the at least one sensor unit (16b), at the time t1, to capture a 3D point cloud and to receive this 3D point cloud.

13. The device (10) as claimed in one of the preceding claims, wherein the computing and control unit (11) is configured to at least partially take into account the latency of the at least one nozzle (13) and the time of flight of the liquid (F) from the at least one nozzle (13) until it hits at least one part of the target object (P1) in order to determine the activation time of the spraying operation for the at least one nozzle (13) onto at least one part of the target object (P1).

14. The device (10) as claimed in one of the preceding claims, wherein the device is part of an agricultural machine, a robot or a drone, or is able to be connected thereto.

15. A method (100) for applying a liquid (F) to at least one part of a target object (P1) in an agricultural field (LF), comprising steps (a) to (c):in step (1), moving a device for applying a liquid (F) to at least one part of a target object (P1) in an agricultural field (LF),in step (2), conveying a liquid (F) from at least one storage container (12) in the direction of at least one nozzle (13) during the movement,in step (3a), causing at least one sensor unit (16), at the time t1, to generate at least one recording and forward this at least one recording to the computing and control unit (11),in step (3b), causing at least one position determination unit, at the time t1, to determine the position of the at least one sensor unit (16) and forward these position data to the computing and control unit (11),in step (3c), determining at least one part of a target object - if present - in the at least one recording by carrying out image processing and georeferencing the at least one part of the target object using the position data of the at least one sensor unit (16) from the time t1 in order to determine the distance S1 from the at least one nozzle (13) to the at least one part of the target object in the recording from the time t1,in step (3d), causing the at least one position determination unit, at the time t2, to determine the position of the at least one nozzle (13) and forward these position data to the computing and control unit (11), wherein the time t2 is a time later than t1,in step (3e), determining the distance S2 from the at least one nozzle (13) to the at least one part of the target object (P1) based on the position data of the at least one nozzle from the time t2 and the position data of the at least one georeferenced part of the target object in the recording from the time t1,in step (3f), determining the speed v2 and direction r2 of the at least one nozzle (13) from the time t2 at least partially on the basis of the position data of the at least one nozzle (13) from the time t1 and from the time t2,in step (3g), scheduling the activation time of the spraying operation for the at least one nozzle (13) onto at least one part of a target object (P1) at least partially on the basis of the distance S2 from the at least one nozzle (13) to the at least one part of the target object (P1) from the time t2 and the speed v2 and direction r2 of the at least one nozzle (13) from the time t2.

16. The method (100) as claimed in claim 15, furthermore comprisingin step (3h), activating the at least one nozzle (13) at least partially on the basis of the scheduled activation time of the spraying operation and applying the liquid to at least one part of a target object (P1).

Images