DE102019201983A1 - Method for determining a condition of a crop planting and agricultural vehicle - Google Patents

Abstract

Method for determining a state (H) of a field crop (16) by means of an agricultural camera support (10) on which a camera (20) is mounted, which provides a field of view (22) which is at least partially directed at the field crop (16) with the steps of: taking at least one digital image (30) by means of the camera (20); Performing a pattern recognition on the image data (34) of the image (30); Detecting the state (H) of the field crop planting (16) from at least one recognized pattern, the pattern recognition being carried out using a neural network (60).

Description

The present invention relates to a method for determining a condition of a crop planting or a crop field by means of an agricultural camera support on which a camera is mounted which has a field of view directed at least partially onto the crop planting.

The present invention also relates to an agricultural vehicle with a camera and with a control device on which a method of the type described above is implemented.

Agriculture has taken considerable steps towards modernization in recent years. The focus is primarily on cultivating agricultural areas in a locally differentiated and targeted manner. The goal is generally to optimize the profitability within a field as permanently as possible.

During the growth of the crop planting, fertilizer should, for example, only be applied as required, possibly as a function of a condition of the field crop planting over the area of the field. Furthermore, efforts are made to drive agricultural vehicles as autonomously as possible and / or to use drones to counteract soil consolidation.

The autonomous vehicle control can take place, for example, via GPS navigation. In addition to this, it is known to use image data for the automatic guidance of an agricultural vehicle.

For example, the document discloses DE 10 2016 209 437 A1 an automatic steering system for guiding an agricultural vehicle over a field on which a lane was left in a previous work process, the steering system having a camera looking at the lane, an image processing system for processing image signals from the camera and a steering control. The image processing system is programmed to convert the image signal from the camera into a binary image and to recognize the lane based on a morphological operation applied to the binary image using a structural element such as vertical lines or rectangles that represents the lane.

The document DE 103 51 861 A1 discloses another method for automatic steering of an agricultural machine in which a camera generates an image signal from which a pixel file is generated. From this, in turn, texture information relating to the texture of the surroundings of the pixels in the pixel file is generated. Pixels of the pixel file are classified taking into account the texture information in order to generate binary information as to whether or not the pixel is to be assigned to an area to be processed. A steering signal based on the results of the classification is then generated. The steering means of the harvesting machine are adjusted according to the steering signal so that the machine is steered automatically.

As a result of the automatic steering, the agricultural vehicle can be safely guided on a tramline which is provided for an agricultural season or permanently in the field, for repeated maintenance measures to be carried out on the plants.

In a standard tramline procedure, exactly the same tramline is used every year. With the help of GPS and data on the machine working width, etc., the field is divided into zones that are driven on and areas not driven on.

As a rule, a large number of add-on parts and / or hangers can be attached to the agricultural vehicle. So-called sprayers are often used for fertilization. To optimize the spraying process, ultrasonic sensors are often used on the sprayers to measure the height of the vegetation. These sensors are usually attached to the sprayer's arms. The plant height determined in this way can be used to ensure an optimal distance between a syringe of the sprayer and the crop planting. The height can also be used to automatically adjust a mower.

Against this background, it is an object of the invention to provide an improved method for determining a condition of a field crop planting and an improved agricultural vehicle.

The above object is achieved by a method for determining a crop planting by means of an agricultural camera carrier, on which a camera is mounted, which has a field of view directed at least partially onto the crop planting, with the steps: recording at least one digital image by means of the Camera; Performing a pattern recognition on the image data of the image; Detecting the state of the crop planting from at least one recognized pattern, the pattern recognition being carried out using a neural network.

The above object is also achieved by an agricultural vehicle with a camera and with a control device on which a method of the type according to the invention is implemented.

With the method according to the invention, a state of a field crop can be determined without special sensors or the like having to be provided for this purpose, such as ultrasonic sensors. The method according to the invention can therefore be carried out inexpensively.

In many cases, a camera is in any case mounted on an agricultural camera carrier, in particular on an agricultural vehicle such as a tractor, which camera has a field of view that is at least partially directed towards the crops. In many cases, a camera of this type is used for navigation or for automatic steering of the vehicle and / or for supporting the automatic steering of the vehicle.

The camera support can be an agricultural vehicle, but it can also be an attachment part on an agricultural vehicle. Furthermore, the agricultural camera carrier can be a trailer for an agricultural vehicle or the like. In general, however, it is also conceivable to use a drone as an agricultural camera carrier that is flown unmanned over the field at a comparatively low height (e.g. maximum 5 m) without contact with the ground.

In particular, the camera is a digital camera. The digital image captured by the camera can be a digital black and white or digital color image. If it is a color image, it is possible to use only the image data of one color (e.g. red, green or blue) to carry out the method, or of all three colors that make up a color image.

Before the pattern recognition is carried out, the digital image can be subjected to digital filtering by means of which, for example, a contrast increase is carried out or the like. Furthermore, the image data can be subjected to preliminary processing in which, for example, optical distortions or aberrations, etc., which are characteristic of the special camera or the special lens used in the camera, are calculated out. In general, it is also possible to carry out automated image processing on the recorded image in order to improve the contrast, etc., and / or to carry out a white balance, possibly depending on the predominant color temperature recorded (cloudy sky, sunshine or other temperature parameters).

The digital image can, for example, be an image with more than 1 megapixel (each pixel can be exclusively a gray value or can contain three individual pixels for a green value, a blue value and a red value).

In order to carry out the pattern recognition it is usually sufficient if the digital image has less than 10 megapixels.

The state of the field crop planting is preferably such a state that can be derived from a pattern that is carried out in the context of pattern recognition on the image data of the image.

Since the pattern recognition and use of a neural network is carried out, it is preferred if the recording of the at least one digital image takes place under comparable boundary conditions under which the neural network was learned in. This can be a position of a camera on the agricultural camera support, for example. Furthermore, the same camera or a camera with at least the same viewing angle should be used (same focal length!). Finally, the field of view of the camera should be set comparably, both with regard to the alignment in the horizontal and the alignment in the vertical.

The task is thus completely solved.

In a particularly preferred embodiment, the neural network is a convolutional neural network. Such networks are well known and also known as convolutional neural artificial networks, see
https://de.wikipedia.org/wiki/Convolutional_Neural_Network.

According to a further preferred embodiment, it is preferred if a target region is first selected from the recorded image.

The target region is preferably a preferred section from the recorded image data.

The field of view of the camera is preferably such a field of view that enables the camera to contribute to automatic steering even during processes. As a result, the field of view is more of a “wide angle” field of view.

To determine the condition of the crop planting, however, only a relatively small section of the digital image is usually necessary, since, for example, pattern recognition is only based on data that are in relative proximity to the camera, for example at a distance less than 15 meters from the camera, especially less than 10 meters.

Accordingly, the target region can be selected in such a way that such a section is selected from the field of view of the camera in which the target region generated in this way extends maximally up to the above value.

For example, this allows the number of pixels to be reduced to less than half the number of pixels in the digital image, preferably less than a quarter, and in particular less than a fifth of the number of pixels in the digital output image.

Furthermore, it is advantageous if, when using the convolutional neural network, a plurality of convolution operations are carried out on the image data.

For this purpose, a convolution matrix or a filter kernel is usually used, which is moved over the digital image data. The convolution matrix is preferably a 3x3 matrix, a 5x5 matrix or a 7x7 matrix.

The input for a neuron of the network determined by means of the convolution matrix is then converted into an output via an activation function. In a subsequent step, a selection operation, known as “pooling”, is preferably carried out after each convolution operation, during which superfluous information is discarded.

After a few repetitive units of respective convolution operations and selection operations, the network can generate one or more so-called fully connected layers that are used for classification or approximation.

Edge detection usually takes place during the steps of the convolution operations and the evaluation operations. The pattern includes in particular at least one edge.

It is particularly preferred if, after a plurality of folding operations, there is at least one fully connected layer that represents a classification or approximation of the state of the crops.

In other words, the convolutional neural network is trained in such a way that not only a pattern has been recognized in at least one fully connected layer as the end of the network, but the state of the crops has already been inferred.

Here, the convolutional neural network can output discrete values as an approximation, for example state 1 , Status 2 , Status 3 , etc. However, the convolutional neural network (also referred to as CNN in the following) preferably outputs classes of discrete size, for example state A from 0 to 0.1, state B from 0.1 to 0.2, etc.

The training of the CNN preferably includes the use of a large number of recorded digital images and assigned real measurement data (“labeled data”).

The number of images used in training should be as large as possible and still be practical. The number can, for example, be in a range of greater than 2,000 and less than 50,000, in particular in a range of more than 4,000 and less than 15,000 images. The optimization and training processes are carried out using suitable learning mechanisms.

In the method according to the invention, the CNN is consequently used not only for pattern recognition, but also for approximation / classification of the condition of the crops.

Overall, it is particularly advantageous if a pattern to be recognized during the pattern recognition has an edge of the crop planting towards which the field of view of the camera is directed.

An edge of the field crop planting is understood to mean a transition from the field crop planting to an unplanted part of the field. The pattern to be recognized can in particular be an upper edge and / or a lower edge of such an edge of the field crop planting, i.e. in particular an area in which the edge plants of the field crop planting emerge from the ground or with their tips (e.g. ears) end up.

It is preferred here if the edge of the crop planting is adjacent to a tramline.

The state of the crop planting may be a fruit ripeness degree, a fruit size, a color, etc.

It is particularly preferred, however, if the state of the crop planting is a plant height of the crop planting.

This state can in particular be used to set a height of a syringe or a distance between the syringe and the tips of the plant when the vegetation is sprayed by means of a sprayer.

According to a further overall preferred embodiment, the camera is mounted on the camera support in such a camera position that the field of view of the camera is directed obliquely from above onto a lane in the crops.

The field of view directed obliquely from above onto the lane makes it possible to clearly see the edge of the crop planting at the transition to this lane. The field of view is preferably such that crops or plants adjacent to the lane can be seen starting from the ground up to the tip of the respective plant; In the case of a two-lane tramline with two lanes, it is preferred if the camera is laterally offset with respect to a longitudinal central axis of the camera carrier or of the vehicle. The lateral offset can, for example, be in the range from 5 cm to 150 cm, in particular in a range from 10 cm to 100 cm.

This makes it possible to record the edge of the field crop planting of one of the two lanes of the tramline in an optimized manner such that the edge of the field crop planting is relatively large in relation to the surroundings.

With the method according to the invention, it is possible with little additional effort to co-determine so-called “ground truth values” for the height of growth, for example by means of ultrasonic sensors, by lidar or manually with the aid of a measuring tape.

Furthermore, a control device is provided which is set up to carry out the method according to one of the previously described embodiments. The control device can have various interfaces for receiving and outputting the corresponding signals. In the context of the invention, a device of the control device for executing a specific function can be understood to mean a specific preparation, for example programming, of the control device for executing the function.

It goes without saying that the features mentioned above and those yet to be explained below can be used not only in the respectively specified combination, but also in other combinations or alone, without departing from the scope of the present invention.

• 1 eine schematische Darstellung eines landwirtschaftlichen Fahrzeugs am Rand einer Feldfrucht-Bepflanzung;
• 2 ein mittels einer Kamera des Fahrzeugs der 1 aufgenommenes digitales Bild;
• 3 eine Zielregion des Bildes der 2;
• 4 eine schematische Darstellung eines erkannten Musters in der Zielregion der 3;
• 5 ein aus dem erkannten Muster der 4 ermittelter Zustand der Feldfrucht-Bepflanzung;
• 6 eine Draufsicht auf ein landwirtschaftliches Fahrzeug, das in einer zweispurigen Fahrgasse in einer Feldfrucht-Bepflanzung fährt;
• 7 ein mittels einer Kamera des Fahrzeugs der 6 aufgenommenes Bild;
• 8 eine Zielregion des digitalen Bildes der 7;
• 9 ein erkanntes Muster in der Zielregion der 8 mit einer Darstellung des ermittelten Zustandes der Feldfrucht-Bepflanzung; und
• 10 eine schematische Darstellung eines Verfahrensablaufs ausgehend von einer Bild-Vorbearbeitung über die Anwendung eine Convolutional Neural Networks und das daraus abgeleitete Klassifizierungsergebnis.

Exemplary embodiments of the invention are shown in the drawing and are explained in more detail in the following description. Show it:

• 1 a schematic representation of an agricultural vehicle on the edge of a crop planting;
• 2 one by means of a camera of the vehicle 1 captured digital image;
• 3 a target region of the image of the 2 ;
• 4th a schematic representation of a recognized pattern in the target region of FIG 3 ;
• 5 one from the recognized pattern of the 4th ascertained condition of crop planting;
• 6th a plan view of an agricultural vehicle traveling in a two-lane tramline in a crop plantation;
• 7th one by means of a camera of the vehicle 6th captured image;
• 8th a target region of the digital image of the 7th ;
• 9 a recognized pattern in the target region of the 8th with a representation of the determined condition of the crop planting; and
• 10 a schematic representation of a process sequence based on image preprocessing using a convolutional neural network and the classification result derived therefrom.

In 1 an agricultural vehicle is shown in schematic form in the form of a vehicle and is generally designated 10. The vehicle 10 stands in a field 12 , more precisely on its arable land 14th . In an area in front of the vehicle 10 is in the field 12 a field crop planting 16 , for example in the form of a wheat field, a corn field, a rape field, etc.

The crop planting consists of single plants with a stalk 16a and a plant head or end, for example a plant ear 16b . The field crop planting 16 has a state in the form of a plant height H starting from the arable soil 14th up to the top of the plant ends 16b enough.

The vehicle 10 is a vehicle coordinate system with an x-axis that is a longitudinal axis of the vehicle 10 corresponds to a y-axis which is a transverse axis of the vehicle 10 corresponds, and assigned to a z-axis which corresponds to a height direction. The height of the plant H runs parallel to the z-direction.

An axle of the rear tires of the vehicle 10 is in one position x ₀ .

On the vehicle 10 is a digital camera 20th permanently mounted, in an x position of x ₁ . An alternative mounting position is at 20 ' shown.

The camera 20th shows a field of vision 22nd which is generally forward facing, so that the crop planting 16 is recorded, including its edge, on which the crop planting 16 ends (seen in x-direction).

Between the camera 20th and the edge of the crop plantation 16 a distance of at least one meter is provided.

A lower edge of the field of vision 22nd falls in the illustrated embodiment at a position x ₂ with the soil 14th together. The position x ₂ lies in front of a position x ₃ which is the x position of the edge of the crop planting 16 represents.

In other words, is the field of view 22nd the camera 20th adjusted so that the edge of the crop planting 16 starting from the arable land 14th is completely recorded up to the top of the individual plants.

In addition, it should be noted that the agricultural vehicle 10 a control device 24 having, preferably with the camera 20th connected is. Furthermore, the agricultural vehicle 10 preferably a steering device 26th on which an automatic steering of the vehicle 10 on the basis either exclusively of data from the camera 20th and / or on the basis of data from a GPS system 28 , which is also preferably in the vehicle 10 is integrated.

The steering device 26th and the GPS system 28 are also located with the control device 24 in connection. The control device 24 can also be designed to have one on the vehicle 10 mounted sprayer 29 to be adjusted with regard to its height, as indicated by a case in 1 is shown, preferably as a function of the plant height H .

In 2 is a means of using the camera 20th captured image 30th to see. The picture 30th shows part of a sky 32 as well as the field crops 16 in the area of their edge. There is also a small part of the arable land 14th to recognize. The picture 30th has a width B _B and a height B _H on. The picture 30th has a multitude of pixels 34 on, one of which is in 2 is indicated schematically. The image can, for example, have a number of pixels greater than 500,000 and less than 20,000,000, with each pixel being assigned either a single gray value or three color values in the form of a red, a green or a blue value.

The camera 20th and the digital images recorded with it 30th be at the vehicle 10 preferably used for navigation or for the purpose of autonomous driving, for example to use an in 1 not shown in detail to recognize tramline. The tramline can be recognized, for example, as in the documents described above DE 10 2016 209 437 A1 and DE 103 51 861 A1 is described.

On the other hand, the camera serves 20th in the present case also for determining a state of the crop planting 16 . The condition can be a degree of fruit ripeness, a fruit size, a color or the like. However, the state is preferably the height of the plant, on the basis of which the present determination method is described by way of example.

To determine the height of the plant H, only a section of the image is preferably needed 30th of interest. The section, which is also used as the target region 36 or "Region of Interest" is in 3 shown enlarged. It can be seen that here the plant height H 'is relative to the height of the target region 36 assumes a relatively large value, so that a comparatively good detection of this state is possible.

The digital image 30th and / or the target region 36 can also be subjected to further pre-processing, which can include, for example, a filtering, a white balance, an increase or decrease in contrast, an increase or decrease in the dynamic range, etc. Furthermore, it is possible to filter gray values, for example based on the color of the crops . For example, a yellow filter can be used to filter out blue in this way, so that the sky in the black and white image then appears black as it does in the target region 36A of the 4th is indicated schematically at 32.

The types of pre-processing of the image 30th or the target region derived from it 36 depends largely on the type of crop planting and other environmental conditions.

In many cases, the preprocessing also includes a compensation of intrinsic errors of the camera, for example a correction of an aberration, a distortion of a lens of the camera, etc. This is for a target region 36B schematically in 5 shown.

Finally, pre-processing can also be a function of external environmental conditions, for example a function of a color temperature (e.g. recording under cloudy skies or sunshine, etc.).

After this pre-processing of the image 30th or the target region 36 the digital two-dimensional image file generated in this way is fed to a convolutional neural network that has been previously trained to recognize the height of the plant.

The convolutional neural network, which is referred to below as CNN for short, has been trained on the basis of a large number (for example more than 1,000 and preferably less than 50,000) comparable image data, specifically using real measurement data for these image data. In other words, the 2 comparable images were recorded and the CNN was then trained to the effect that the correct real height H, measured by other means (either manually or by means of sensors or the like) was used as the training parameter for the respective images.

The CNN leads to the image data of the image or the target region 36 a pattern recognition by and detects the state in the form of the plant height H of the field crop planting from at least one recognized pattern. Both the implementation of the pattern recognition and the detection of the state are carried out by the CNN.

The CNN outputs an approximation of H as a discrete variable, for example a plant height in the form of a numerical number, which is taken from a number range that contains, for example, a limited number of numerical values. These are spaced, for example, so that the distance corresponds to a difference in plant height of more than 0.5 cm and less than 50 cm. For maize as a field crop, the relevant difference is in a range from 5 cm to 50 cm. For smaller plantings, for example wheat, the difference is, for example, in a range from 1 cm to 10 cm.

By restricting the number range to, for example, more than 5 and less than 50, in particular less than 30 different values, the training of the CNN with the above-mentioned limited number of image data can be performed.

As an alternative to outputting discrete values for the plant height H, it is possible to train the CNN in such a way that it outputs classes of plant heights, for example 10 cm to 20 cm or 20 cm to 30 cm or 30 cm to 40 cm or ...

The number of classes is preferably in a range from 5 to 30, in particular in a range from 5 to 15.

By means of the method described above for determining a condition of a field crop planting, the condition can be recorded or derived from image data from a camera, specifically by means of a CNN. Alternatively, another neural network can also be used, with a CNN for image data being the preferred artificial neural network.

It is therefore possible to save sensors such as ultrasonic sensors that are attached to the sprayer 29 be attached in order to measure the height H during operation. Another advantage is that the height of the vegetation can be determined in advance while the vehicle is in motion, and the height of the sprayer can then be adjusted, taking into account the distance covered and, for example, the distance from the measured section of the crop plantation to the sprayer 29 can be adjusted in the x-direction.

Another embodiment of a method for determining a plant height of a field crop is described below, which generally corresponds to the method of FIG 1 to 5 corresponds. The same elements are therefore identified by the same reference symbols. The main differences are explained below.

In 6th can be seen that an agricultural vehicle 10 , which is designed as a two-lane vehicle, along a tramline 40 drives that one left lane 42 and a right lane 44 includes. The tramline 40 is one in the crop planting 16 provided tramline, which is regularly used for cultivating the planting 16 is driven on and consequently in the area of the lanes 42 , 44 corresponds directly to the arable soil or the arable soil height.

The vehicle 10 has a central axis 46 in the x direction, ie in the longitudinal direction. The camera 20th has an axis of field of view 48 which runs obliquely to the central axis 46 , so at an angle of more than 1 degree and preferably less than 30 degrees to the central axis 46 is aligned. Furthermore, it is alternatively or additionally preferred if the camera 20th opposite the central axis 46 is laterally offset, in the y-direction. The side distance 50 between the central axis 46 and the camera 20th can for example be in a range from 5 cm to 100 cm, preferably in a range from 5 cm to 50 cm.

This asymmetrical arrangement or alignment removes one of the two lanes from the field of view 22nd so captured that an edge 52 of the crop planting 16 at the transition from this planting to the lane (here the left lane 42 ) is clearly visible, starting from the arable land 14th to the top of the crop planting 16 in the area of this edge 52 .

One by means of the camera 20th captured image 30th (according to the picture 30th of the 2 ) is in 7th shown.

It can be seen that the two lanes 42 , 44 can be seen, taking from the left lane 42 a predominant part of the arable land 14th can be seen. It is also for the left lane 42 the edge 52 starting from the lane 42 up to the top of the crop planting 16 easily recognizable, especially in a lower part of the image 30th .

Hence becomes a target region 36 from the lower part of the picture 30th selected as it is in 8th at 36A is shown. In the picture the 8th the right lane can still be seen to some extent, but it no longer plays a role in further pattern recognition and status determination. 9 corresponds 5 and shows a target region 36B .

It is crucial that the representation of the 8th and / or the 9 and especially from the edge 52 of the crop planting 16 with the aid of a CNN can determine the height H of the crop planting, in the same way as above with reference to the 2 to 5 described.

The height H can again be output as a discrete value or as a plant height class.

10 shows in schematic form the determination method presented above.

In one step 62 image preprocessing takes place, as described in detail with reference to the 2 and 3 has been described and which in the same way to the 7th and 8th is applicable.

The pre-processed image file (or its target region 36 ) into a CNN 60 entered, within its convolution operations 66 , Selection operations 68 (Pooling) and further steps known from the prior art relating to CNN are carried out until one or more fully connected layers are ultimately implemented 70 from which a classification of the state results or which represents such a classification of the state (indicated at 64), whereby the term classification of the state H can mean that discrete values are represented as well as classes in the form of plant height Areas as described above.

In a preferred variant, a pattern recognition first takes place in the CNN, from which the tramline or at least one lane can be derived. This fully connected layer, which represents data with regard to the tramline, can then be used, for example, by an automatic steering system and then also by a further CNN or further folding operations and selection operations up to further fully connected layers from which the classification of the Plant height H of the crops can be seen.

During the determination process, ground truth values for the plant height can be determined with a little extra effort, e.g. by ultrasonic sensors, by lidar or manually using a tape measure. These ground truth values can be used for ongoing training of the CNN.

By restricting the image data to a target region, the runtime of the CNN can be positively influenced.

List of reference symbols

10

agricultural vehicle

12

Field

14th

Arable land

16

Field crop planting

16a

Straw

16b

Ear of wheat

20th

Camera (digital)

22nd

Field of view

24

Control device

26th

Steering device

28

GPS system

29

Sprayer

30th

image

32

sky

34

pixel

36

Target region

40

Tramline

42

left lane

44

right lane

46

Central axis ( 10 )

48

Field of view axis

50

Side spacing

52

Edge ( 16 )

60

Convolutional Neural Network

62

Pre-processing image data

64

Classification / approximation result

66

Folds

68

Selection operations (pooling)

70

Fully connected layer

H

Stature

B _H

Height image 30th

B _B

Wide image 30th

QUOTES INCLUDED IN THE DESCRIPTION

This list of the documents listed by the applicant was generated automatically and is included solely for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Patent literature cited

• DE 102016209437 A1 [0006, 0067]
• DE 10351861 A1 [0007, 0067]

Claims (11)

Method for determining a state (H) of a field crop (16) by means of an agricultural camera support (10) on which a camera (20) is mounted, which provides a field of view (22) which is at least partially directed at the field crop (16) has, with the steps: - Recording at least one digital image (30) by means of the camera (20); - Carrying out a pattern recognition on the image data (34) of the image (30); - Detection of the state (H) of the field crop planting (16) from at least one recognized pattern, the pattern recognition being carried out using a neural network (60). Procedure according to Claim 1 wherein the neural network is a convolutional neural network (60). Procedure according to Claim 1 or 2 , wherein a target region (36) is first selected from the recorded image. Procedure according to Claim 2 or 3 wherein a plurality of convolution operations (66) are performed on the image data. Procedure according to Claim 4 wherein after a plurality of folding operations (66) there is at least one fully connected layer (70) which represents a classification or an approximation of the state (H) of the crop planting (16). Method according to one of the Claims 1 - 5 wherein a pattern to be recognized during the pattern recognition has an edge (52) of the field crop planting (16) to which the field of view (22) of the camera (20) is directed. Procedure according to Claim 6 wherein the edge (52) of the crop planting (16) is adjacent to a lane (42). Method according to one of the Claims 1 - 7th wherein the state (H) of the crop planting (16) is a plant height (H) of the crop planting. Method according to one of the Claims 1 - 8th wherein the camera (20) is mounted on the camera support (10) in such a camera position that the field of view (22) of the camera (20) is directed obliquely from above onto a lane (42) in the crop planting (16) . Control device (24) which is set up to implement a method according to one of the Claims 1 to 9 execute. Agricultural vehicle (10) with a camera (20) and with a control device (24) on which a method according to one of the Claims 1 to 9 is implemented.

Images