Method for operating a machine for harvesting and / or separating root vegetables, related machine, and related computer program product

Abstract

The present invention relates to a method for operating a machine (2) for harvesting root vegetables and / or separating root vegetables (24) from additional material being transported with the root vegetables (24), including at least soil in the form of soil smears and / or soil aggregates (28) and possibly leaves and / or stones, wherein at least one electromagnetic, particularly optical, or acoustic image capture unit (12) captures at least one inspection image of at least a portion of the material being moved by at least one conveying element, particularly a sieving belt (10), relative to a machine frame (6) of the machine (2), and an evaluation device generates a setting signal for setting at least one operating parameter of the conveying element and / or additional conveying element of the machine (2) based on at least one inspection data set generated and / or formed by the inspection image, wherein at least one characteristic representative of the sieving ability of the transported soil is determined by the evaluation device and used to set the operating parameter. The present invention also relates to a machine for harvesting root vegetables and a computer program product.

Description

[Technical Field]

[0001] The present invention relates to a method for operating a machine for harvesting root vegetables and / or separating them from further material conveyed therewith, including at least soil in the form of soil smears and / or soil aggregates, and possibly leaves and / or stones. The method further comprises capturing, by means of at least one electromagnetic, in particular optical, or acoustic image capture unit, at least one inspection image of at least a portion of the material being moved by at least one conveying element, in particular a sieving belt, relative to the machine frame of the machine. Furthermore, an evaluation device is used to set the operating parameters of the machine.

[0002] In U.S. Patent Application Publication No. 2017 / 013773, the soil texture and / or soil friability are detected, for example, via optical or acoustic sensors, in order to adapt the depth of the digging element depending on the soil information. This is based on the knowledge that, even with the same settings, the working depth of the digging element becomes shallower when moving from soft soil to hard soil, which can lead to damage to root crops. Therefore, by setting the digging depth of the harvester depending on the soil condition before digging the harvested crop, damage to the harvested crop is reduced and a better yield is achieved.

[0003] The object of the present invention is to use soil information for the treatment of materials in a machine.

[0004] This problem is solved by the subject matter of claim 1. Preferred embodiments of the invention can be read from the dependent claims as well as the description. Furthermore, this problem is solved by a machine according to claim 15 and by a computer program product according to claim 16.

[0005] According to the invention, a method for operating a machine for harvesting root vegetables and / or separating them from further material being transported with them, comprising at least soil in the form of scattered soil and / or soil aggregates and possibly leaves and / or stones, is proposed, which first involves taking, by means of at least one electromagnetic, in particular optical or acoustic image capture unit, at least one inspection image of at least a portion of the material being moved by at least one conveying element, in particular a sieving belt, relative to the machine frame of the machine, and using an evaluation device to generate, on the basis of the inspection image and / or on the basis of an inspection data set formed by said inspection image, a setting signal for setting at least one operating parameter of the conveying element or further conveying elements of the machine, wherein at least one characteristic representative of the sieving capacity of the transported soil is determined by the evaluation device and used for setting the operating parameter.

[0006] According to the setting signal, the action of one or more adjusting or driving means of the conveying element is changed in order to vary the sieving of the soil.The invention therefore relates to conveying elements in the form of separating equipment provided for sieving the soil, in particular here to sieving belts, hedgehog belts and / or roller arrays.

[0007] In particular, the image capture unit orients its sensor toward a conveying element configured as a sieving belt that conveys the material by the machine as a first or second sieving belt after the material has been picked up. This embodiment of the method according to the invention thus analyzes the soil very early after picking, in particular when it is captured in the first half or first third of the sieving belt, in order to be able to react very quickly to changes in soil conditions by adapting the operating parameters and thus to be able to adapt at least a portion, in particular a large portion, of the sieving section, including one or preferably several sieving belts, immediately or even preventively. The setting of the operating parameters preferably takes place within a time frame of at most one minute, preferably at most 30 seconds, after the capture of the test image. That is, the evaluation device is configured to complete the evaluation of the test image within this time frame. Furthermore, in particular, the adaptation of the operating parameters is at least initiated or performed within this time frame.

[0008] Soil components or aggregates of soil components with a minimum diameter of 5 mm or more are considered soil aggregates. For example, in the case of non-spherical aggregates, the equivalent diameter can be assumed as the diameter. Soil aggregates always consist of a large number of interconnected particles, including, in particular, sand, silt, and / or clay. Slurries are particle fractions with particle sizes up to 2 mm, and, according to the above definition, are connected particles with diameters less than 5 mm.

[0009] Similarly, the optical or acoustic image capture unit or further such image capture units may have their sensors oriented in the area behind the conveying elements in order to be able to react to any remaining soil aggregates, so that, for example, if there are too few relatively small aggregates identified and too many large aggregates identified, the screening width can be reduced, for example, and the beating performance of the beaters acting on the screening belt can be improved in order to reduce the large aggregates.

[0010] Preferably, the composition of the material present in the inspection image, including scattered soil and / or soil aggregates, in particular soil clods, is determined, in particular using a classification method, where soil clods are generally considered to be larger aggregates, i.e., blocks of soil with a diameter of more than 5 cm. The term "soil clod" may be used as a synonym for elongated soil clods. Based on such classification, for example, by evaluating the color values ​​of the inspection data set, the individual components of the transported observation material can be identified. Therefore, for the observation of soil aggregates, a simple method can be used to match the inspection image or a part of the inspection image that contains only soil aggregates or at least soil aggregates relative to the majority of components (see below).

[0011] Preferably, the features representative of sieving ability include one or more values ​​representative of the size, shape, moisture, strength, or color of one or more soil aggregates, and / or one or more values ​​representative of one or more, particularly statistical distributions, of the size, shape, strength, or color of a number of soil aggregates, for example, via such values, a size class can be assigned to all soil aggregates present in an inspection image or inspection dataset for purposes of better automated processing without the need to fit to individual soil aggregates.

[0012] Preferably, the size of the observed soil aggregates, in particular the clods, is adapted. For example, these may be diameter, equivalent diameter, projected area, or volume. In a further alternative or complementary embodiment of the invention, aggregate shape or color parameters can be extracted, since color and shape change accordingly under different soil types and varying soil moisture, which represent major influence factors on sieving ability. The same applies to strength, which may be derivable from color, shape, and size. For example, the breaking load can be used as a measure for the strength of soil aggregates.

[0013] In a further alternative or complementary embodiment of the method according to the invention, one or more, in particular statistically aggregated, features or distributions formed via values ​​for the size, color, strength and / or shape of a large number of, in particular different, soil aggregates are used as sieving ability features.

[0014] Such statistical features summarizing the distribution of many individual features may be, for example, the mean, standard deviation / variance, median, percentile, or first, second, or kth order.

[0015] According to a further development of the invention, the size, shape or color are classified into a predetermined number of classes, preferably five, with the values ​​monotonically increasing or decreasing from the first to the last class. The average membership or most probable class of the observed soil aggregates of the test data set is then used as a feature representing the sieving capacity. For example, if there are small aggregate size classes on average, for example, due to the fact that the soil has been cultivated in a mostly dry sandy soil, the sieving capacity can be reduced, for example, by setting a small sieving rod or roller spacing.

[0016] Preferably, in a further configuration of the method according to the invention, the screening performance characteristics obtained in a plurality of successive evaluation cycles are filtered or offset against one another before determining the operating parameters or the settings of the operating parameters on their basis. In particular, the use of a low-pass filter or a moving average value is suitable here for smoothing out possible outliers.

[0017] An inspection dataset may be formed by an inspection image or a portion of an inspection image. It may also be a dataset resulting from processing and / or analysis of an inspection image or a portion of an inspection image. Furthermore, an inspection dataset may include the inspection image itself, a processed inspection image, and / or a dataset resulting from analysis of the inspection image. The same is true for the inspection image, the processed inspection image, and portions of the dataset resulting from analysis of at least a portion of the inspection image.

[0018] At least one characteristic for determining the sieving capacity, obtained based on the test data set, is used to set the operating parameters of or one of the conveying elements, particularly the sieving belt, provided for sieving the soil. Subsequently, a quantitative determination of the sieving capacity characteristic is made after a qualitative determination of the components present in the harvested crop based on color values, as described, for example, in DE 102018127844 A1. By using the method according to the present invention, the conveying element can be improved and configured so that a desired amount of soil aggregates, e.g., clods or clod sizes, are present in the area of ​​the conveying section present after picking. Thus, the harvested crop in the form of root vegetables is optimally separated or gently protected during transport by the device, depending on the desired soil aggregates, particularly soil aggregate size. Therefore, by using the method according to the present invention, the throughput of the machine during operation is increased while the risk of damage to the root vegetables is reduced, which leads to improved economic efficiency in the use of the machine.

[0019] The image capture units used to form an image of the material transported on the transport element are referred to as electromagnetic image capture units and have sensors that operate to capture electromagnetic waves, in particular light waves. They can be sensors that can be used to capture one-dimensional or multidimensional images. For example, they can be one or more radar sensors that can receive waves reflected from the material, whose frequencies are in the range of 10 to 150 GHz. Optical image capture units are designed to capture light in the visible, ultraviolet, and / or infrared regions, in particular.

[0020] Optical image capture units are particularly contemplated as 2D, 2.5D, or 3D cameras, such as RGB cameras, time-of-flight cameras, black-and-white or grayscale cameras, or stereo cameras. Similarly, light cross-sectioning, structured light, and plenoptic cameras can be used to capture the harvested crop. Acoustic image capture units can be particularly in the form of arrays of distance-measuring acoustic sensors, especially ultrasonic sensors, which are equally suitable for generating input data for the evaluation device, i.e., inspection images, by the continuous movement of the harvested crop along a sensor array attached to the machine frame. The same applies to arrays of optical distance sensors and arrays of mechanical contact sensors. Most RGB cameras have two-dimensional image sensors in the form of CCD or CMOS sensors.

[0021] It is also clear that in the method according to the invention and the device according to the invention a combination of electromagnetic and acoustic image capture units is possible and that the image capture unit may have one or more electromagnetic and / or acoustic sensors.

[0022] Preferably, the features are determined based on the input dataset generated by or formed by the test dataset using neural network analysis, histogram analysis, and / or Structure from Motion (SfM) analysis, which are particularly well suited to the large amounts of data generated when observing crop flow, in particular during harvesting operations of potato or beet harvesters.

[0023] In particular, in a development of the method according to the invention, the neural network is designed as a convolutional neural network which classifies each input data set into one of a number of classes representing different sieving ability feature values. This type of neural network has been found to be particularly good for identifying soil aggregates during harvesting operations and for identifying the sieving ability features assigned to these soil aggregates.

[0024] In particular, the evaluation device comprises one or more CPU units and / or one or more graphic processor units, in particular in the form of GPUs (Graphical Processing Units) or GPGPUs (General Purpose Graphical Processing Units), and / or a processor unit based on an FPGA (Field Programmable Gate Array). This representation of the evaluation device allows for a particularly resource-friendly and particularly local evaluation of the test data sets. It goes without saying that an evaluation device configured as an EDP device or configured by such a device also comprises the usual means for example for power supply, interfaces, a main memory, and non-volatile program and data memory.

[0025] According to the invention, in order to improve the screening performance of the conveying element, in particular the hedgehog belt or sieving belt, especially under difficult harvesting conditions when the picked soil or picked soil only breaks down into a few large aggregates, a beater, for example in the form of a rotor beater or swing beater, can be configured. This causes an additional rocking movement of the conveying element, which results in impact destruction of soil aggregates, for example in the form of soil clods, and therefore improves screening performance. Further or supplementary operating parameters of the conveying element configured as a sieving belt are, in particular, the sieving belt speed, the sieving belt pick-up speed, the set height of at least one triangular roller, the set height of any drop step, the frequency of one or more beaters, for example the amplitude of a swing beater, the position of the beater relative to the sieving belt or the position of the adjusting means acting on the sieving belt, and / or the inner diameter of the sieving belt. The operating parameters of the sieving belt are therefore operating variables such as speed, frequency, amplitude or position, which can possibly be set by a unit acting on the sieving belt. Setting signals are therefore signals output or initiated by an evaluation device, which accordingly affect the setting of these variables. In the case of conveying elements configured as hedgehog belts, the aforementioned setting means may also be present in part, in particular for setting the belt speed. In the case of conveying elements in the form of roller arrays, their rotational speeds and their mutual distances can be set.

[0026] A particularly camera-based analysis of the crop flow on the conveying elements, in particular on the sieving belt, leads to the detection of features that allow the average aggregate size to be estimated, so that, based on this information, beaters or other means that influence the sieving performance of the sieving belt can be automatically activated or deactivated in a need-oriented manner and adjusted in their function, thereby, for example, allowing a number of setting means that are arranged one after the other in relation to the conveying direction of the sieving section, in particular parallel to this conveying direction, to be variably set.

[0027] Preferably, the image capture unit captures at least the front third of the sieving belt in particular. Depending on the position of the camera's viewing angle, it is advantageous to transfer the optically distorted image of the section of interest into a rectangular representation.

[0028] For particularly good and particularly accurate capture of the harvested crop independent of the environmental conditions, the observed sieving belt section or the harvested crop transported on the sieving belt can be illuminated by means of an illumination unit.

[0029] Preferably, for the determination of the aggregate size of the soil or soil aggregates, e.g., clod size, by the evaluation device, an area of ​​at least 75%, preferably up to 90%, more preferably up to 95% of the test image or test data set is selected, and even more preferably an area that exclusively contains soil or soil clods and is particularly connected, preferably rectangular. Even if these areas are not visible in each image, especially at high travel speeds during harvesting operations, it has been shown that a correspondingly limited area for determining the clod size gives particularly good results, and even small areas, in particular of at least 15 cm x 15 cm, are sufficiently representative for possible setting or changing of operating parameters.

[0030] Such regions are selected in an automated manner, in particular by examining the color information contained in the test data set, possible edges or gradients of the 2D or 3D image, and possibly using statistical analysis, in particular by using pixel-by-pixel classification based on color information.

[0031] The portion of the inspection data set representing the connected, preferably at least substantially soil-containing rectangular area is either directly provided for the determination of at least one soil mass size, or is processed as an input data set and provided to a neural network analysis, a histogram analysis, and / or a structure from motion analysis in which the image area is assigned at least one soil mass size that is used to set operating parameters.

[0032] In particular, when using a neural network, preferably a convolutional neural network (CNN) including a convolutional layer, the input data set can be converted, for example, into vector format. Topologically, a CNN can be a sequential or recurrent network. A convolutional layer is a processing level within the network that applies a convolutional filter to an input matrix. As with other layers, the convolutional filter has flexibility in the form of weighting. Here, the convolutional filter is used to extract image features, and subsequent classification, for example, is performed based on these image features. However, to make the application of CNN practical for the present invention, several improvements had to be made in its development. To reduce the memory requirements of the network, split weights are used for the neurons of the convolutional filter, assuming that the interest levels of image features are uniform regardless of their location in the image. Furthermore, split weights are robust to translation, rotation, scaling, and intensity variance. Additionally, pooling layers, particularly those that truncate values ​​by region, can be used between convolutional layers. Preferably, max pooling is used, whereby all values ​​in a small region (eg 2x2 or 4x4) are discarded except for the largest and therefore most significant value.

[0033] In order to keep the calculation of the CNN as simple as possible, preferably, a ReLU (rectified linear unit) is used as the activation function, which is always zero in the negative domain and increases linearly in the positive domain. This improvement makes the CNN particularly suitable for the practical operation of the machine according to the invention. Additionally, computer architectures that can be used for parallel calculations are also advantageous. The neurons in a layer are independent of each other and can therefore be calculated particularly in parallel. Therefore, architectures with many parallel calculation units, for example in the form of GPUs, are well suited for the CNN.

[0034] Preferably, the image segment is selected so that it has a maximum value of the soil density distribution, i.e., so that it has a maximum value of the corresponding density distribution that substantially contains soil clods or soil. Such a distribution is generated, in particular, based on a pixel-by-pixel classification in which all pixels of the observed image area are classified according to the probability of belonging to the possible object groups of root vegetables, soil / clods, and possibly stones and / or leaves. A further classification variable could also represent an "empty sieving belt," i.e., the absence of possible harvested crops. Therefore, preferably, the area with the maximum number of pixels representing soil is used to select the image segment. In this case, the maximum value of "soil density" is the optimal area to be provided as an input variable to an evaluation method for detecting one or more sieving ability features, preferably based on a neural network, in particular a CNN. Preferably, image segments having a soil or earth density of less than 75% are not used for determining the feature representative of sieving ability, so that the determination of the at least one sieving ability feature is not impaired by excessively many foreign objects, such as stones, leaves, or root vegetables, which may be present. Thus, for example, image segments can be defined automatically depending on a lower threshold value for the minimum soil density.

[0035] It is trivial to pre-train the neural network using a large number of test images or test datasets of the same size that correspond to the appropriate image classification. For images of similar size, the edge regions of smaller images can be filled with zeros by so-called zero padding. Image size variations of up to 10-15% do not lead to a significant deterioration in the image analysis results. For example, for a large number of test datasets, each one is manually assigned to one of five size classes.

[0036] Specifically, when a CNN is used, test images or image regions of a test dataset can be assigned one or more categories of soil aggregate size. For example, a value on a scale of 1 to 5 can be assigned to each image. During the CNN training phase, this sieving ability value is determined by a human expert for a representative number of sample images from which the CNN can learn its internal weights. During the online prediction phase, this value, automatically estimated by the CNN, is then used to control or set operating parameters.

[0037] Based on the available technology, it has proven advantageous to define a particularly rectangular input data set for the neural network, each with an edge length between 100 and 400 pixels, preferably between 150 and 250 pixels, for a large number of harvesting conditions within a large trial window. This size represents a very good compromise between the computational power required to perform the setting of the operating parameters during the running movement within less than 30 seconds, preferably less than 3 seconds, after imaging, taking into account the dominant variance, and the required duration.

[0038] Furthermore, using this input image size, it is possible to find a sufficiently large image area that is occupied almost exclusively by soil or soil aggregates in a sufficiently large percentage (greater than about 75%) of the camera image during machine operation. These image sizes should also be viewed in relation to the resolution of the image sensor system used, e.g., camera resolution, particularly in the range of 0.5 to 10 megapixels, and preferably 1.0 to 1.5 megapixels.

[0039] Preferably, the evaluation device evaluates the test data set at least partially locally on the machine or on a directly connected towing vehicle. In particular, insofar as a large database is required due to the large number of configurable parameters, the evaluation device can also at least partially, preferably completely, evaluate the test data set on a wirelessly connected server, although in this case a correspondingly fast and stable data connection is also required for continuous operation.

[0040] A hybrid form of such evaluation is also contemplated, whereby part of the hardware used for the evaluation may reside on a local machine and further part may reside at a remote location.

[0041] In some cases, an environmental sensor, such as a soil sensor, in particular a moisture sensor, can be used in a further embodiment of the present invention to additionally determine the moisture content of the soil aggregates and use it in the evaluation device to set the operating parameters. The moisture sensor can operate on an electrical or optical basis. In addition, a sensor for measuring the electrical conductivity can provide further information about the soil condition, which can be used to set the operating parameters of the screening section.

[0042] In a further embodiment of the method according to the invention, the machine is equipped with a position sensor, for example a GNSS receiver, and has map material of the area to be harvested stored directly on the machine or provided via a remote server using a mobile network connection. In this case, the local soil type stored in the map material in combination with the machine position can be a further auxiliary variable for setting the operating parameters of the screening section.

[0043] Likewise, in one embodiment of the method, the machine may be connected to a remote server via which additional weather data, i.e. humidity and temperature information, can be retrieved, which can additionally be incorporated into the determination of the operating parameters of the screening section.

[0044] Preferably, the determination of the operating parameters is part of the machine's control loop, and in particular, weather, i.e., humidity and temperature, and / or soil type and / or harvesting strategy are additionally used as input or reference parameters. In such a control loop, additional, usually one-dimensional input variables that can be captured at higher frequencies can be introduced, such as the filling state of the conveying elements or load values ​​of the drive elements, such as pressure, torque, current consumption, etc. Using such input variables, operating parameters of the machine, in particular the screening section, can be controlled at high frequencies. Preferably, such a control loop can additionally control the harvesting depth and / or the travel speed.

[0045] Here, reference parameters such as weather, i.e. humidity and temperature and / or soil type and / or harvesting strategy, and in particular one or more characteristics determined according to the invention, are then used to represent the soil screening capacity in a cascade control sense, in particular for optimizing the parameters of a cascaded control loop, and thus to compensate for disturbance variables that cannot be captured based on the input variables of the high-frequency control loop, such as variations in soil type. Preferably, in this case the cycle time of the high-frequency control loop is in the range of 1 to 100 milliseconds, whereas the capture of the screening capacity characteristics from the test data set is carried out with a cycle time of 100 milliseconds to 30 seconds.

[0046] In particular, all input parameters that may be relevant for setting the operating parameters are linked to one another in a corresponding database, in particular in the form of a locally or externally stored data bank, where additionally, the sieving performance characteristics and the operating parameter values, as well as in particular environmental variables, may be brought into empirical or possibly analytical relationships. The setting of the operating parameters can also be carried out here under various conditions using the database.

[0047] In one embodiment of the method according to the invention, the evaluation device, designed as an EDP device, is configured to record image data and / or other sensor data and / or sieving performance characteristics and / or other calculation intermediate results and / or operating and / or environmental parameters, in addition to calculations, particularly in conjunction with the machine's location. These data can be stored locally on the evaluation device accompanying the machine or transmitted to a central server via a mobile connection. These data can also be used in the form of maps to display the sieving characteristics on the work field. Furthermore, especially when these data are recorded by multiple machines and centrally compiled, they can also be used to improve the algorithm logic for capturing sieving performance characteristics and, in some cases, the control logic for determining operating parameters.

[0048] The problem set out at the outset is also solved by a machine for harvesting and / or separating root vegetables, which machine comprises at least one electromagnetic, in particular optical, or acoustic image capture unit, a conveying element movable relative to the machine frame of the machine and in particular configured as a sieving belt, an evaluation device, and means for setting the conveying element or further conveying elements, wherein the machine is suitable for carrying out the steps of the method according to claim 1 as well as the further features of the method according to the invention described above or further below. In particular, the means for setting the conveying belt are means for setting the sieving belt as described above or further below.

[0049] This problem is also solved by a computer program product comprising instructions for causing a machine for harvesting and / or separating root vegetables according to the invention as described above or below to carry out the method steps as described above or below.

[0050] Further advantages or details of the invention can be seen from the following description of the drawings. [Brief explanation of the drawings]

[0051] [Figure 1] 1 is a side view of a machine according to the invention; [Figure 2] FIG. 2 is a perspective view of a part of the object according to FIG. [Figure 3] 2 shows a part of the object according to FIG. 1 captured by an optical image capturing unit; [Figure 4] FIG. 10 illustrates the selection of an area of ​​an inspection image. [Figure 5] FIG. 5 shows a region classification means according to FIG. 4; [Figure 6] FIG. 10 shows a means for influencing the sieving belt. [Figure 7] FIG. 10 shows further means for influencing the sieving belt. [Figure 8] FIG. 10 shows further means for influencing the sieving belt. [Figure 9] FIG. 10 shows further means for influencing the sieving belt. [Figure 10] 1 is a flow chart of a method according to the present invention; [Figure 11] 4 is a further flow chart of a further method according to the invention;

[0052] The individual technical features of the embodiments described below can be combined with the features of the preceding embodiments and of the independent and possibly further claims to achieve the object of the invention. Wherever useful, elements that act functionally in the same way are provided with the same reference signs.

[0053] According to the present invention, the machine 2 is configured to harvest root vegetables in the form of potatoes and is therefore configured as a potato harvester. Material in the form of soil or soil aggregates, root vegetables, leaves, and / or stones picked up in the area of ​​the receiving bin 4 is transported in conveying direction 1A via a conveying element in the form of a sieving belt 10 attached to the rear of the machine frame 6 and a further frame part 8. The sieving belt 10 is directly connected to the receiving bin 4 (FIG. 2). The crop transported in direction 1A using the sieving belt 10 is captured by a first optical image capture unit 12 in the form of an RGB camera fixed to the machine frame part 9 and oriented obliquely relative to the sieving belt 10 in the direction of the receiving bin 4. Illumination means 14 illuminate the sieving belt in the area of ​​a second image capture unit 12 arranged behind the first image capture unit 12 in the conveying direction 1A.

[0054] The area of ​​the machine 2 captured by the first image capture unit 12, located on the right side of FIG. 2, is shown in FIG. 3 and is free of soil. In particular, the individual sieving rods 16 of the sieving belt 10, located directly downstream of the horizontal ploughshare 18 of the receiving bin 4, are visible. Below the sieving belt 10, the machine frame 6 contains a setting means for setting the operating parameters for the operation of the sieving belt 10. This may be a roller 20, shown in FIG. 3, which may also be part of a further setting means, depending on the configuration of the machine 2 according to the invention. Using the evaluation device, an image section 22 is automatically selected (in FIG. 4, it contains as few root vegetables 24 as possible and impurities 26 (here, leaves)). According to the invention, the image section 22 contains at least 90% soil aggregates 28 and is already perpendicularly oriented, whereas the rest of the image is still shown slightly perspectively distorted.

[0055] The identification of leaves and soil aggregates is performed using pixel-by-pixel classification based on color values ​​captured by the optical image capture unit 12, including values ​​representing gray and / or actual color, which are compared to reference values ​​or ranges of reference values. This form of differentiation allows for a qualitative identification of components on the inspection image, and in particular, a pixel within a pre-determinable or pre-set threshold is assigned to a class of (harvested) material (soil / soil aggregates, leaves, roots, stones).

[0056] To the extent that region 22 is identified, the test data set or portion of the test data set representing this region is fed to the neural network, possibly in a version adapted to the neural network's input requirements. This neural network, in particular a CNN, assigns at least one soil aggregate size to the image region, and in further embodiments of the invention also the proportions of the size distribution of the various sieve belt segments within the image. The segments in Figure 4 show soil aggregate sizes, including particularly large soil aggregates, as can be seen in the right-hand graphical portion of Figure 5. Moving from this right-hand graphical portion to the left, further size classes of soil aggregates are shown that are recognized by the neural network and with which it was pre-trained.

[0057] Depending on the aggregate size thus defined, the operating parameters can be varied, for example, the beat amplitude or the movement frequency of the swing beater 30 shown in FIG. 6. Its swinging movement transmits impacts to the sieving belt 10, which leads to the breaking up of soil aggregates, especially soil clods. Alternatively or additionally, the frequency of the rotor beater 32 shown in FIG. 7 or the position of the triangular wheel 34 shown in FIG. 8 can be changed in relation to the machine frame 6 supporting the sieving belt. Similarly, the setting rail 30 (FIG. 9) can change the distance of the sieving belt 10 to the belt 3, thereby moving the sieving rod units formed of two sieving rods 16 connected via connectors 38, which in turn changes the clearance between the mutually following rods and the mutually following sieving rod units.

[0058] The flow of the method according to the invention according to Fig. 10 starts with a first method step 40, in which an inspection data set 42 is generated using the image capture unit 12, which is then qualitatively segmented using pixel-based classification in step 43, thereby allowing assignment 44 of individual image regions or pixels of the inspection data set to potatoes, leaves, soil, soil, etc. Then, in step 46, a region or image segment 22 containing only soil and corresponding soil aggregates is selected. This region 22 is analyzed using a CNN in step 50, via which the segmented image is assigned a size class according to Fig. 5 in result 52. Then, in step 54, operating parameters are changed, possibly by outputting or initiating setting signals, and the sieving performance of the sieving belt 10 is subsequently adapted.

[0059] The setting of the sieving performance of the sieving belt 10 and thus the setting of the sieving section according to step 54 is preferably also part of a control loop 60 (FIG. 11), in which a machine control unit 62 including an evaluation device accesses a local or external data bank 64 from which, on the one hand, allocation rules for setting the operating parameters of the sieving belt according to step 54, a harvest depth setting 66, and / or a running speed 68 are obtained. Furthermore, the machine control unit 62 can process a large amount of further information, including a filling state identification 70 of the sieving belt start and / or a filling state identification 72 of the sieving belt rear and / or, optionally, pressure information 74 of the separating device and / or, optionally, clogging identification 76 of the separating device. Finally, in step 78, actual values ​​of the individual operating parameters of the individual functional assemblies can be captured and processed as input information for the machine control unit 62.

[0060] Typically, the evaluator 80 is part of the machine control unit 62. Additional inputs for the machine control unit 62, besides the soil aggregate size estimate 52, are a harvesting strategy 82 predetermined by the operator and / or environmental information regarding weather and soil type coming from the capture unit 84. Circle 86 symbolizes the impact of the size class identification 52, harvesting strategy identification 82, and environmental variable identification 84 performed by the evaluator on the harvesting performance of the machine 2 mapped by steps 70-78. For example, for a maximum yield harvesting strategy, once the maximum aggregate size class is identified, the swing beater amplitude, sieve rod spacing, and belt speed are maximized, whereas for a more gentle strategy, the amplitude is increased less and the belt speed is simultaneously reduced.

Claims

1. 1. A method for operating a machine (2) for harvesting root vegetables and / or for separating root vegetables (24) from further material conveyed therewith, the further material including at least soil in the form of soil aggregates (28), comprising: At least one electromagnetic or acoustic image capture unit (12) captures at least one inspection image of at least a portion of the material moved by at least one sieving belt (10) relative to the machine frame (6) of the machine (2), and an evaluation device generates, based on the inspection image and / or on at least one inspection data set formed by the inspection image, a setting signal for setting at least one operating parameter of the sieving belt (10) and / or a further sieving belt (10) of the machine (2), and by means of the setting signal, the action of one or more adjusting or driving means of the sieving belt (10) is changed in order to vary the sieving of the soil, At least one characteristic representative of the sieving ability of the co-conveyed soil is determined by the evaluation device and used to set the operating parameters, and a quantitative determination of at least one sieving ability characteristic is performed after qualitative determination of components present in the inspection image and in the harvested crop using a classification technique; The method, wherein the characteristics include one or more values ​​representing the size, shape, moisture, intensity, or color of one or more of the soil aggregates (28) and / or include one or more values ​​of the size, shape, moisture, intensity, or color of a number of the soil aggregates (28).

2. 2. The method of claim 1, wherein the features are determined by the evaluation device based on an input dataset generated by or formed from the test dataset using neural network analysis, histogram analysis, and / or structure from motion analysis.

3. 3. The method of claim 2, wherein the neural network used in the neural network analysis is designed as a convolutional neural network that classifies each of the input data sets into one of multiple classes representing different values ​​of a screening ability feature.

4. 4. The method of claim 1, wherein the components of the material present in the inspection image are determined using a classification technique.

5. 5. The method according to claim 1, wherein an area (22) of the inspection image or the inspection data set containing at least 75% of the soil aggregates (28) is selected for determining the characteristics by the evaluation device.

6. 6. The method of claim 5, wherein the portion of the inspection data set representing the region (22) is provided directly or processed as the input data set and provided to the neural network analysis, the histogram analysis, and / or the structure from motion analysis, in which the region (22) is assigned the features used to set the operating parameters.

7. 7. The method according to claim 1, wherein the evaluation device evaluates the inspection data set at least partly locally on the machine (2) or on a directly connected tractor.

8. 8. The method according to claim 1, wherein the evaluation device evaluates the test data set on a wirelessly connected server.

9. 9. The method according to claim 1, wherein the operating parameters of the sieving belt (10) are a sieving belt speed, a sieving belt pick-up speed, a set height of at least one triangular roller, a set height of a drop step, a beater frequency, a beater amplitude, a beater position, and / or an inner diameter of the sieving belt.

10. 10. The method according to claim 1, wherein the moisture of the soil aggregates (28) is determined by means of a moisture sensor and used in the evaluation device for setting the operating parameters.

11. 11. The method according to any one of the preceding claims, wherein the determination of the operating parameters is part of a control loop of the machine (2).

12. 12. The method of claim 11, wherein the control loop is used to control harvest depth and / or travel speed.

13. 13. The method according to claim 1, wherein the setting of the operating parameters is performed using a database, in which the characteristics and operating parameter values ​​are mutually linked and stored.

14. A machine (2) for harvesting and / or separating root vegetables (4), said machine (2) comprising: At least one electromagnetic image capture unit (12); a sieving belt (10) movable relative to the machine frame of said machine (2); an evaluation device and means for setting the sieving belt (10) or a further sieving belt (10), A machine (2) configured to perform the steps of the method according to any one of claims 1 to 13.

15. A computer program comprising instructions for causing a machine according to claim 14 to carry out the method steps according to any one of claims 1 to 13.

Images