Classifying device for grown natural products and method for classifying grown natural products

Abstract

The invention relates to a classifying device (10) for classifying grown natural products (O), said classifying device being suitable for reliably automatically classifying a natural product (O). In order to solve the problem, the classifying device (10) has the following features: - at least one means for generating an optical recording (12) of a natural product (O) to be classified from two or more different perspectives, - an evaluation unit (14) which is designed to compare the optical recordings with stored reference information items (36) and to assign the natural product (O) to be classified to one of a plurality of virtual classes on the basis of the comparison, and - an output unit (16) which is designed to output the natural product (O) assigned to the virtual class in a physically classified manner in accordance with the evaluation.

Description

[0001] I059PC2401

[0002] - 1 -

[0003] Classifying device for grown natural products and methods for classifying grown natural products

[0004] Description

[0005] The invention relates to a classification device for classifying grown natural products, comprising

[0006] - at least one means of producing an optical image of a natural product to be classified from two or more different perspectives,

[0007] - an evaluation unit which is designed to compare the optical images with stored reference information and to assign the natural product to be classified to one of several virtual classes based on the comparison, and

[0008] - an output unit which is designed to physically classify the natural product assigned to the virtual class according to the evaluation.

[0009] Furthermore, the invention relates to a method for classifying grown natural products.

[0010] Classification in the form of sorting natural products is known, for example, from the food industry, to divide foodstuffs processed as bulk goods into two predetermined classes, for example quality classes.

[0011] The terms sorting and classifying must be distinguished in that classifying involves dividing a good into three or more classes, whereas sorting involves dividing a good into only two classes. For example, a natural product can be sorted into one of the two classes "good" or "bad" I059PC2401

[0012] - 2 - can be assigned. In contrast, the assignment can be graded by means of classification, whereby a grown natural product can be assigned to one of the classes "very good", "good", "adequate" or "poor".

[0013] Sorting goods using a sorting device is a well-known practice. For example, color sorters are used in the food industry to separate individual elements of a granular product based on color differences. These color sorters typically feature optical image recognition for identifying the granular product and a mechanical separation device for sorting out individual elements. Using known color sorters, for instance, individual grains of cereal with a distinctive color can be sorted out from a mass of grains falling like a curtain, particularly by blowing them out.

[0014] Sorting goods into two classes is technically relatively simple to implement, since image recognition requires only a single threshold value for a single evaluation criterion, such as brightness or color shade, and evaluating the goods against this threshold. The physical division into two classes is also technically relatively simple because the elements of a good assigned to the first class can be directly extracted from the set of elements, and the remaining elements directly form the second class. Therefore, the elements of the second class require no further processing for sorting.

[0015] Compared to sorting, classifying goods is technically challenging. This is because, for classification, the goods to be classified must first be assessed according to multiple evaluation criteria and / or multiple threshold values ​​in order to determine the three or more classes. Furthermore, physically assigning an element of the goods to be classified into one of three or more classes requires more complex technical means than sorting. In particular, a mere I059PC2401

[0016] - 3 - Multi-stage, successive sorting using known mechanical separating devices is regularly unsuitable for classifying goods because such successive sorting is characterized by high costs, time losses during processing, and / or quality losses in the goods being classified. The invention described here must be explicitly distinguished from multi-stage, successive sorting.

[0017] Challenges arise particularly in the classification of naturally grown products. Naturally grown products are those that have been produced through an essentially biological process. They must therefore be distinguished from both engineered products and non-organically grown natural products. Classifying naturally grown products is challenging because these products often exhibit significant variation with respect to a given assessment criterion relevant for classification in each individual case. Significant variation means that there can be widely differing characteristics with respect to a given assessment criterion. Assessment criteria with significant variation include, for example, the color(s), shape, and / or size of the individual elements within the quantity of the natural product.The shape of a hazelnut kernel, for example, can be essentially spherical, hemispherical, or fragmented into a multitude of random shapes, thus exhibiting considerable variation. Furthermore, classifying natural products is challenging because these products can easily be damaged by mechanical action during the classification process. Damage caused by classification must be avoided or at least minimized.

[0018] Due to the aforementioned technical challenges in classifying natural products, this classification is generally carried out manually by human workers, based on the visual assessment of a single item within a quantity of a natural product. Such manual classification is time-consuming and can be unreliable due to a lack of objective evaluation criteria. I059PC2401

[0019] - 4 -

[0020] The invention is based on the objective of providing a classification device for classifying grown natural products, which is suitable for classifying a natural product quickly, reliably and gently.

[0021] The invention achieves its objective with the features of the independent claims. Further practical embodiments and advantages of the invention are described in connection with the dependent claims.

[0022] The classification device for classifying grown natural products has the following features:

[0023] - at least one means of producing an optical image of a natural product to be classified from two or more different perspectives,

[0024] - an evaluation unit which is designed to compare the optical images with stored reference information and to assign the natural product to be classified to one of several virtual classes based on the comparison, and

[0025] - an output unit which is designed to physically classify the natural product assigned to the virtual class according to the evaluation.

[0026] The classification device can be designed, in particular, such that individual elements of a quantity of a natural product to be classified are individually optically detected, individually assigned to a virtual class, and individually classified and output. Furthermore, it can be possible, in particular, for individual elements of a quantity of a natural product to be classified to be handled individually using the classification device described here.

[0027] The means for generating an optical image is designed to classify the natural product from at least two different I059PC2401

[0028] - 5 -

[0029] The device optically captures images from different perspectives. For this purpose, it has at least one optical sensor. This means of generating an optical image captures the natural product optically without contact. The different perspectives are distinguished by the fact that different surface sections of the natural product to be classified are captured from each perspective. To this end, the means of generating an optical image can be designed such that images are captured from different viewing angles of the natural product. The at least one optical sensor can, for example, be moved around the natural product, and images can be successively captured from different viewing angles. Additionally or alternatively, the means of generating an optical image can have a plurality of optical sensors arranged offset from one another and directed at the natural product.Alternatively or additionally, the natural product can be rotated in front of the at least one optical sensor of the means for generating an optical image, with successive optical images of the natural product being taken. Preferably, the angle between two adjacent different perspectives is 45° to 180°, and more preferably 75° to 105°. This enables the recording of a large surface area within a short time, with the images from the different perspectives partially overlapping, so that the optical images can be combined with one another using an image processing unit described below, and / or at least a proportion of features optically perceptible in a plurality of images can be spatially assigned to one another.

[0030] The evaluation unit is designed to compare optical images of the grown natural product from different perspectives with stored reference information and to assign the natural product to be classified to one of several virtual classes based on this comparison. The optical images can be transmitted to the evaluation unit, particularly as digital data. For this purpose, the means for generating an optical image includes an I059PC2401

[0031] - 6 -

[0032] The data transmission interface facilitates data transfer between the optical image capture device and the evaluation unit. The evaluation unit has a corresponding first data transmission interface. Furthermore, the evaluation unit comprises the image processing unit, a storage unit, and a processing unit, such as a computer processor, to process and optionally store the transmitted data.

[0033] The storage unit can contain, for example, the stored reference information used to compare the data from the optical images of the natural product, the image processing unit, and / or process instructions to be executed by the computing unit for the automated comparison of the images with the stored reference information, including predetermined evaluation criteria. Furthermore, the data transmitted from the optical imaging device to the evaluation unit and / or data generated from the optical images can optionally be stored in the storage unit. Storing the data typically requires a certain amount of time, so the evaluation of the optical images and, consequently, the classification may be slower than without data storage. However, storage can still be advantageous, for example, to improve the evaluation process, which will be explained in more detail below.Against this background, the classifying device can be operated in two different operating modes.

[0034] If the process rules for automated image matching and / or the image processing unit are stored on the storage unit, the process rules and / or the image processing unit can be loaded into the processing unit for performing a classification of natural products. Alternatively, the process rules and / or the image processing unit can also be permanently contained in the processing unit. I059PC2401

[0035] - 7 -

[0036] In a first operating mode, the process instructions may, for example, stipulate that for the automated comparison of the optical images of the natural product with the stored reference information, the data of the optical images transferred to the evaluation unit are loaded directly into the image processing unit without saving the optical images themselves. In a second operating mode, the process instructions may optionally additionally stipulate that the data of the optical images transferred to the evaluation unit are saved in the storage unit and that this saved data is then loaded into the image processing unit.

[0037] In the image processing unit, the optical images are automatically compared with the stored reference information. During this comparison, at least some of the optically perceptible features in an image are recognized, these features are assigned to at least one predetermined evaluation criterion, and information about the recognized features is further processed as data, taking the assigned evaluation criterion into account. For example, the outer contour of a natural product can be recognized as an optically perceptible feature, assigned to the predetermined evaluation criterion "geometric shape," and the information "the geometric shape of the natural product is hemispherical" can be further processed in relation to this criterion. Additionally or alternatively, the perceptible features and the associated information can relate to, for example, brightness, color, color distribution, and / or size.Alternatively or additionally, the surface of the product visible in the images can be analyzed. In particular, the central area of ​​the image is used for surface analysis, and the outer area is used for analyzing the contour of the object being analyzed.

[0038] Preferably, for the automated comparison of the optical images with the stored reference information, a plurality of evaluation criteria per image are examined, and the assignment of an image to a virtual class is preferably carried out by results I059PC2401

[0039] - 8 - comparisons regarding a plurality of evaluation criteria for admission can be combined.

[0040] Additionally or alternatively, the image processing unit can be configured to combine multiple optical images of a natural product from different perspectives, for example, to create a 360° view. Again, additionally or alternatively, the image processing unit can be configured to visually divide a single optical image, depicting a natural product from multiple perspectives, into multiple representations, each showing a single perspective of the natural product, and to further process each of these representations of the natural product from one of the different perspectives as a separate, dependent image within the framework of automated comparison.

[0041] Separate, dependent images of a natural product taken from different perspectives can be assigned identifiers, allowing these images to be linked to one another. These identifiers can optionally be stored on the storage unit along with the separate, dependent images and / or the data generated from them. Based on these identifiers, images showing the same element of a set of natural products from different perspectives, and / or the results of comparing images showing the same natural product from different perspectives, can be combined as needed, as described above. This makes it possible to avoid classification errors that arise because a feature—for example, a hemispherical shape—is not visually perceptible from a first perspective.

[0042] The evaluation unit also includes a second data transmission interface, which serves for data transmission between the evaluation unit and the output unit. In the first operating mode, the process specifications may stipulate that the result of the assignment of an I059PC2401

[0043] - 9 -

[0044] The second interface directly transfers the image of a natural product to one of several virtual classes. In this second operating mode, the process specifications may stipulate that the result of the assignment and the information recognized in an image are additionally stored in the storage unit. Alternatively, in this second operating mode, the result of the assignment and the information recognized in an image can first be stored in the storage unit, and then the result of the assignment can be transferred from the storage unit to the output unit. In the storage unit, the result of the assignment and the information can be stored, in particular linked to the optical image to which the information is to be assigned, in order to generate a complete data set for an element of the set of natural products.

[0045] The output unit is designed to physically output the natural product assigned to a virtual class, classified according to the evaluation. For this purpose, in one practical embodiment, the output unit includes a manipulator. The manipulator moves the natural product to be physically classified into one of several containers, each corresponding to the virtual class into which the evaluation unit has assigned the natural product. Thus, each of the several containers holds only natural products of a single, predetermined class.

[0046] The manipulator can, in particular, have a deflecting device that contacts the natural product to be classified and which can be aligned, for example, with the different containers. The alignment of the contacting deflecting device can be effected by means of an actuator, wherein the actuator is controlled, for example, by the evaluation unit, in particular by the processing unit, depending on an assignment result of a physically classifiable natural product to a virtual class. The contacting deflecting device can, for example, have a section in the form of a funnel, a funnel section, or, in particular, a hollow cone section, wherein the funnel, the funnel section I059PC2401

[0047] - 10 - or the hollow cone section is rotatable about an axis of rotation. The funnel, the funnel section or the hollow cone section may, in particular, be designed such that the natural product to be classified and entering the funnel, the funnel section or the hollow cone section can be deflected by a wall of the funnel, the funnel section or the hollow cone section in a direction transverse to the axis of rotation.

[0048] In another practical embodiment, a distribution element is provided as a contacting deflection device, designed as an impact surface for a flying natural product. For example, a distribution plate can be designed to rotate in such a way that it interacts with a specific number of dispensing openings, causing an impacting natural product to rebound in a specific direction upon a single impact and thus be directed to a specific container. For instance, such a distribution element can be designed to assign a natural product precisely to one of five containers evenly spaced around a full 360-degree circular area. In this case, the distribution element targets a specific area of ​​360 degrees divided by 5, i.e., 72 degrees.For this purpose, a distribution plate can be designed in such a way that an impacting natural product is conveyed directly into the desired 72-degree target area by a single impact on the distribution element.

[0049] By appropriately designing guide channels and / or pipe sections, with which a classified natural product is conveyed from the distribution element to a container, the arrangement of the containers can also be arbitrarily different.

[0050] If the manipulator has a deflection device that can be driven rotaryally in two directions, the next desired position of the deflection device can be reached from any position with a maximum adjustment angle of 180 degrees, thus saving adjustment time. If five equally sized ranges of 72 degrees each are provided, as described above, the maximum achievable adjustment angle is even greater.

[0051] - 11 - only 144 degrees, with three equally sized areas of 120 degrees, the maximum adjustment angle to be reached is 120 degrees, so that a very fast adjustment of the position of the distribution element is possible in each case.

[0052] As an alternative to the contact deflection device, the manipulator can have a contactless deflection device. The contactless deflection device can have multiple compressed air outlets, which are arranged, for example, on the inside of an annular element. By releasing compressed air from one or more of the compressed air outlets, the natural product to be classified, which enters the contactless deflection device, can be blown in a predetermined direction.

[0053] Additionally, the manipulator may have one or more guideways or channels on which or in which the natural product to be classified is conveyed.

[0054] The classification device described here enables particularly fast, reliable, and gentle classification of natural products. Especially in the first operating mode, which allows classification without saving optical images or data derived from them, 10 to 20 items of a given quantity of material can be classified per second.

[0055] In practice, the means for generating an optical image of a natural product to be classified can be designed to capture the natural product from at least four different perspectives. Capturing the image from four different perspectives enables classification of the natural product based on a complete 360° view. The optical images from the four perspectives can advantageously overlap. In particular, it can be designed to capture the entire surface of the natural product and thus record all features relevant for classification during comparison (I059PC2401).

[0056] - 12 - to take the stored information into account. This can further reduce the risk of incorrect classification.

[0057] For the sake of completeness, it should be noted that the means of generating an optical image can also be designed to capture five or six perspectives or more perspectives, for example four horizontal perspectives (front, back, right, left) and one or two vertical perspectives (top, bottom).

[0058] However, experiments have shown that using four images, generated with two single-mirror arrangements and two double-mirror arrangements, is particularly efficient for a high-quality, complete view of the entire surface of a natural product. This allows four perspective images to be created from a single camera viewpoint, which can then be combined into a single photograph.

[0059] The camera and the mirrors are preferably arranged in such a way that the entire surface of the object to be analyzed is captured with overlaps in the outer edge area of ​​the respective image.

[0060] This can be achieved in particular by means of a mirror arrangement, in which two horizontal beams of light are first directed past the object from the rear right and rear left onto the object to be analyzed, and two beams of light are first directed by means of a first mirror into a position above or below the object to be recorded and then by means of a second mirror in a predominantly vertical direction from the top right and from the bottom left (or vice versa) onto the object to be recorded.

[0061] The means for generating an optical image of the natural product to be classified can include a camera and a mirror optic. The aforementioned optical sensor is then, for example, the camera's sensor. It is also possible that several cameras, and thus several sensors, are provided. The mirror optic can have one or more mirrors.

[0062] - 13 - wherein the at least one mirror directs light emitted by the natural product onto the sensor of the at least one camera. The mirror optics can be designed such that the natural product is optically magnified. This magnification allows features relevant for classification to be more easily perceived in the images during the comparison process.

[0063] The camera can be specifically aligned with the mirror optics, and the mirror optics can be designed such that a natural product to be classified can be captured from different perspectives by the camera in a single photograph. For this purpose, the mirror optics can comprise a plurality of mirrors, with at least some of the mirrors facing a common viewing area from different directions. For example, these mirrors direct light emitted from the common viewing area onto different areas of the camera's sensor. The light can be directed onto the camera's sensor directly or indirectly. Indirect light guidance can be achieved, for example, by means of an intermediate deflecting optic, in particular a prism, a deflecting mirror, or fiber optics.Due to the described arrangement of the mirrors, the natural product to be classified, when it is in the viewing space, is viewed from a plurality of different perspectives simultaneously, and the natural product to be classified can be captured from a plurality of perspectives with a single photograph.

[0064] In this context, it is particularly advantageous if the image processing unit is designed to separate a single image depicting a natural product from multiple perspectives in such a way that each representation of the natural product from one of the different perspectives can be further processed as a separate, dependent image within the framework of automated matching. I059PC2401

[0065] - 14 -

[0066] In another practical embodiment, a flight path is provided upstream of the output unit, which natural products to be classified can traverse while flying, particularly falling. The means for generating an optical image is designed to capture the natural product to be classified within the flight path. In this case, due to the flying state of the respective natural product, for example, when a hazelnut is in free fall, each side of the natural product can be captured and thus viewed simultaneously, because no side is obscured by a surface. Preferably, the flight path is a space within the device extending vertically in the direction of gravity, which the natural product traverses from top to bottom due to the influence of gravity, passing in front of an entry point into the output unit.The flight path is preferably located downstream of an exit area of ​​a separation device and upstream of the entry area of ​​the output unit, and in this case is spatially situated between a separation device arranged above and an output unit arranged below.

[0067] In a practical embodiment, the flight path is completely enclosed in an upper section and / or a lower section, particularly above and below the area where the at least one optical image is captured, by means of a guide structure, thus preventing the particles from leaving the natural product. For this purpose, a hollow cylindrical profile or a hollow profile with another cross-section can be provided, with round cross-sections being preferred.

[0068] The circumference of the guide structure is preferably between 1.2 and 5 times the maximum diameter of the natural products to be classified. For efficiency reasons, it is particularly preferred that the circumference of the guide structure in the inlet area of ​​the output unit is approximately 3 times the maximum diameter of the natural products to be classified. I059PC2401

[0069] - 15 -

[0070] Preferably, the flight path is less than 100 cm, more preferably less than 80 cm, and particularly preferably 60 cm, 50 cm, or even only 30 cm. The flight time for unaccelerated natural products is accordingly between 0.25 seconds at 30 cm and 0.4 seconds at 80 cm. Even flight paths of 10 cm with a flight time of 0.14 s or 5 cm with a flight time of 0.1 s are practically feasible, especially if the at least one image is reliably acquired within this time and, if the method is also to be implemented, the output unit can be brought into an output configuration assigned to the assignment in time before leaving the flight path. This requires high-speed data transfer including evaluation and control of the output unit.

[0071] In a preferred embodiment of the device according to the invention, the steps described above, i.e., creating a recording, evaluating and optionally bringing the output unit into an associated output configuration, can be carried out within a maximum of 67 ms, i.e. 0.067 s, so that approximately 15 natural products per second can be classified with the device.

[0072] In another practical embodiment, at least a portion of the area in which images are generated by the optical imaging device has a configurable and / or monochrome background. This includes, for example, a receiving device for inserting or otherwise securing a background element, such as a colored strip of material. The securing can be achieved, for example, by means of clamp-like elements or a guide rail into which the material strips are inserted. Alternatively or additionally, magnetic fastening, hook-and-loop fastening, or another suitable type of fastening is also considered. I059PC2401

[0073] - 16 -

[0074] In addition to the aforementioned options for a physical background element, an electronically generated background design can also be provided, in particular a background element in the form of a backlight – which may also be adjustable. Preferably, the background is chosen to create a particularly high contrast between the natural products to be classified and the background, especially to enable the natural products to be distinguished from the background with sharp and precise contours. A black background has proven advantageous for hazelnuts.

[0075] With electronic backgrounds, the background can also be dynamically selected and displayed by first determining the color of the natural product in an early flight phase and then, depending on this, setting a different color as a background with particularly good contrast, so that a suitable (dynamic) background is already active during at least one shot.

[0076] In practice, the evaluation unit can be trained to perform machine learning and / or to assign the natural product to be classified to the virtual class using artificial intelligence. The artificial intelligence can, for example, be stored on the aforementioned storage unit and loaded into the processing unit and / or executed by the image evaluation unit during the automated comparison of the optical images with the stored information or the image evaluation unit. If the assignment of the natural product to be classified to the virtual class is carried out using artificial intelligence, the stored reference information can include, in particular, feature patterns of known natural products. Feature patterns are frequently occurring combinations of several feature values ​​learned by the artificial intelligence with respect to several evaluation criteria.Artificial intelligence can be trained to compare the features visually perceptible in the recordings with the feature patterns and to decide which features are recognizable and to which evaluation criterion these features are assigned I059PC2401.

[0077] - 17 - are how a characteristic is pronounced in relation to an associated evaluation criterion and / or into which class of the majority of virtual classes the natural product shown in a photograph is to be classified.

[0078] The use of artificial intelligence is particularly advantageous for classifying natural products. This is because, due to the aforementioned variations in each evaluation criterion that must be considered when classifying natural products, developing algorithms for evaluating a single criterion is highly complex. It is comparatively simpler to train an artificial intelligence using known characteristic patterns of familiar natural products. Furthermore, the decision regarding the classification of a natural product can be made more quickly using artificial intelligence than with a complex algorithm.In this context, the evaluation unit can be configured such that the artificial intelligence can be trained using 2 to 1000 images, particularly 200 to 500 images, and preferably 300 images, of natural products assigned to known and predetermined classes. Additionally or alternatively, the classification device can be operated in a second operating mode for training purposes. In this mode, the data from the optical images transmitted to the evaluation unit, the results of the classification, and / or the information recognized in an image are stored in the memory unit. In a second step, the classification result can be evaluated by a human operator based on an assessment of the stored data from the optical images and / or the information recognized in an image.This assessment can be communicated to the artificial intelligence so that the artificial intelligence can learn.

[0079] Classification is performed based on objective criteria by comparing optical images of the grown natural products using artificial intelligence. Furthermore, the classification can be fully automated, very I059PC2401

[0080] - 18 - accurately and very quickly with consistent quality. These advantages can be particularly pronounced due to the combination of artificial intelligence matching with the generation of images of the natural product from at least two perspectives.

[0081] In practice, the classification device can also have a storage container in which the quantity of the natural product to be classified can be stored.

[0082] Furthermore, in practice, the classification device can include at least one separation unit for the natural product to be classified. The separation unit serves to separate individual elements of the natural product to be classified from a quantity of the natural product and to feed these individual elements to the means for generating an optical image. Advantageously, the separation unit automatically separates an individual element of the natural product to be classified from the quantity of natural products and automatically feeds it to the means for generating an optical image.

[0083] In particular, the at least one separation device can comprise at least one spiral conveyor and preferably a vibratory spiral conveyor. A vibratory spiral conveyor comprises at least one spiral track on which the elements of the quantity of natural products can be conveyed, and a vibratory drive. The spiral track can be integrated into the storage container. It is set into vibration by means of the vibratory drive. This causes the elements to be conveyed along the spiral track. In particular, a controlled feed of elements for classification can be achieved in this way.

[0084] Alternatively or in addition to a spiral conveyor, at least one conveyor belt, at least one roller, at least one pulley and / or other conveying device may also be provided. I059PC2401

[0085] - 19 -

[0086] The separation device can further include an additional conveying device, in particular a conveyor belt, which is arranged between one end of the spiral track and the means for generating optical images of the natural products to be classified. The additional conveying device can transport the elements of the natural product to be classified from the end of the spiral track to the means for generating optical images. In particular, the spiral conveyor and the additional conveying device can be configured such that the conveying speed of the additional conveying device is greater than the conveying speed of the spiral conveyor. This allows a predetermined distance to be set between two elements successively fed to the means for generating optical images.As a result, the available duration for assigning individual elements to one of the classes can be flexibly set in this way.

[0087] The separation unit can be designed to handle natural products ranging in size from 0.05 cm to 30 cm, particularly from 0.3 cm to 10 cm. For this purpose, the spiral track and the additional conveying device are dimensioned accordingly. The separation unit can be specifically designed to handle nuts, preferably hazelnuts.

[0088] Additionally or alternatively, the output unit can be configured to output the natural product assigned to the virtual class in a radial direction. For this purpose, the manipulator described above can be configured to move the natural product, after its virtual assignment to one of several classes, from the position in which the virtual assignment takes place, in one of several radial directions. The selection of the radial direction in which the natural product is moved is based on the result of the virtual assignment. Moving the natural product in a specific direction can be accomplished, for example, by a deflection device. I059PC2401

[0089] - 20 -

[0090] If the dispensing of the classified natural product is radial, a plurality of containers for the classified storage of the natural product dispensed by the dispensing unit can be arranged circumferentially around the dispensing unit. Alternatively or additionally, guide channels and / or pipe sections leading to containers are arranged circumferentially around the dispensing unit.

[0091] The invention also relates to a method for classifying grown natural products, according to which a) in a first method step a natural product to be classified is fed to a flight path, b) during the passage of the flight path, at least one image is created with a means for generating an optical image of the natural product to be classified from two or more different perspectives, c) with an evaluation unit the optical images are compared with stored reference information and the natural product to be classified is assigned to one of several virtual classes on the basis of the comparison, and d) with an output unit the natural product assigned to the virtual class is physically output in a classified form according to the evaluation by bringing the output unit into an output configuration associated with the assignment before leaving the flight path.

[0092] Reference is hereby made again to the advantages already described above in connection with the device, which also apply to the method. Furthermore, reference is made to the further development possibilities of the method resulting from the device, such as the timely movement of a manipulator, designed as a contacting deflection element, before it leaves the flight path. In this regard, switching time can be saved in particular if the deflection element can be driven rotaryally in two directions. I059PC2401

[0093] - 21 -

[0094] Further practical embodiments and advantages of the invention are described below in connection with the drawings. They show:

[0095] Fig. 1 shows a first embodiment of the classifying device in a view from an oblique top view;

[0096] Fig. 2 shows the classifying device from Fig. 1 in a side view;

[0097] Fig. 3 shows the classifying device from Fig. 1 in a sectional side view with a plurality of natural products to be classified;

[0098] Fig. 4 shows a schematic representation of functional elements of the classifying device from Fig. 1;

[0099] Fig. 5 shows a second embodiment of the classifying device in a view from an oblique top view;

[0100] Fig. 6 shows the classifying device from Fig. 5 in a cutaway side view with two natural products to be classified;

[0101] Fig. 7 shows an enlarged representation of the area of ​​the classifying device designated VII in Fig. 6 in a sectional side view with two natural products to be classified;

[0102] Fig. 8 shows an enlarged view of the area of ​​the classifying device designated VIII in Fig. 7 in a sectional view with two natural products to be classified.

[0103] Fig. 9 shows a third embodiment of a classifying device in a perspective view, I059PC2401

[0104] - 22 -

[0105] Fig. 10 shows the classifying device from Fig. 9 in a sectional view and

[0106] Fig. 11 shows only one camera with a schematic representation of the four viewing directions of images taken with the camera.

[0107] In the figures, matching reference symbols in different figures denote identical or functionally equivalent technical elements. For clarity, not all reference symbols are included in the figures, even though the elements may be present.

[0108] Since Figures 1 to 4 show the same embodiment of the classifying device for classifying grown natural products, these figures are described together below.

[0109] The classifying device 10 shown in Figures 1 to 4 for classifying natural products comprises a means for generating an optical image 12 of a natural product to be classified from two or more different perspectives. A natural product to be classified is an object 0 that can be captured by the means for generating an optical image 12. This image can be output as digital data. Furthermore, the classifying device 10 includes an evaluation unit 14. The evaluation unit 14 is configured to receive optical images or data generated by the means for generating an optical image 12 and to compare them with stored reference information, as described in detail below in connection with Figure 4.Based on this comparison, a natural product 0 depicted in a photograph and to be classified is assigned to one of several virtual classes by means of the evaluation unit 14. The classification device 10 also includes an output unit 16, which is configured to receive the result of the assignment to one of the virtual classes and to physically output the natural product 0 assigned to the virtual class in the class determined by the evaluation. I059PC2401.

[0110] - 23 -

[0111] The means for generating an optical image 12 comprises a camera 18 with an optical sensor, a mirror optic 20 which includes at least the mirrors 20a and 20b shown in the figures, and an object detector 22.

[0112] Camera 18 is attached to a mounting device and can be positioned approximately 30 cm away from the mirror optics 20. Camera 18 is a digital camera that transmits optical images to the evaluation unit 14 in the form of digital data.

[0113] The mirror optics 20 and the object detector 22 are arranged in an enclosure of the means for generating an optical image 12 in Figures 1 and 2, which depict the first embodiment of the classifying device 10, and are therefore not visible in detail. However, the mirror optics 20 and the object detector 22 can be seen in Figure 3 and in Figures 6 and 7, which show the second embodiment of the classifying device 10.

[0114] The mirror optics 20 and the object detector 22 are identically designed and arranged in both embodiments.

[0115] The mirror optics 20 comprise four mirrors, of which only two, 20a and 20b, are shown in Figures 3, 6, and 7 due to the sectional side view. The four mirrors of the mirror optics 20, including mirrors 20a and 20b, are arranged offset from one another and inclined, and they are aligned from different directions toward a common viewing space 24. The mirrors of the mirror optics 20 can, in particular, be arranged in pairs opposite each other around the viewing space 24. This arrangement allows the viewing space 24 to be viewed from a different perspective by each of the mirrors of the mirror optics 20. The mirrors of the mirror optics 20 are inclined to one another such that each of the four mirrors reflects light emitted from the viewing space 24 directly onto the sensor of the camera 18 through an opening in the housing of the optical recording device 12. I059PC2401

[0116] - 24 -

[0117] In another embodiment, not shown here, the mirror optics 20 can comprise fewer or more than four mirrors instead of four, for example, two, three, five, six, or ten mirrors. The mirrors are preferably arranged equidistant from one another on a circle or on a spherical surface around the viewing area 24, so that a large part of the surface of a natural product to be classified can be captured. Alternatively, the mirror optics 20 can also include a deflecting optic, in particular a prism, a deflecting mirror, or a fiber optic, between the mirrors and the sensor, onto which the light from each of the mirrors is reflected and which then directs the light to the sensor (not shown in the figures). Additionally or again alternatively, the mirrors of the mirror optics can be curved mirrors.Regardless of the design of the mirror optics 20, the light emitted from the viewing area 24 is reflected onto the sensor of the camera 18.

[0118] The camera 18 is functionally connected to the object detector 22 such that the camera 18 is automatically triggered when the object detector 22 detects an object 0, in particular a natural product 0. In the classification device 10 described here, the object detector 22 is specifically designed as an optoelectronic system for generating a light barrier, which communicates with the camera 18. When a natural product O, which moves, for example, along the arrow P1 shown in Figures 3 and 7, crosses and interrupts the light barrier, the object detector 22 sends an electrical signal to the camera 18 to trigger it.The latency between the detection of a natural product O and the triggering of the camera 18 is such that the arrangement of the object detector 22 and the viewing space 24 captured by the camera 18 is such that a moving natural product O detected by the object detector 22 is located in the viewing space 24 at the time the camera 18 is triggered.

[0119] So that the camera 18 can see the image reflected by the mirrors of the mirror optics 20

[0120] To detect light, camera 18 has its sensor pointing in the direction of the I059PC2401

[0121] - 25 -

[0122] The sensor is arranged in a mirror-aligned manner – or, if provided, in the direction of the deflecting optics. The sensor has different areas (not shown), the number of which corresponds to the number of mirrors in the mirror optics 20. In the embodiment described here, the sensor has four areas, in particular four quadrants. The position and inclination of each of the four mirrors provided in the embodiment described here is adjusted such that each of the different areas of the sensor of the camera 18 captures the light from one of the different perspectives.

[0123] If an object 0, in particular a natural product 0, is located in the viewing space 24, the natural product 0 can be captured from four different perspectives with the camera 18 in a single optical image thanks to the classification device 10 described here. As a result, particularly fast classification of the natural product 0 is possible based on a complete 360-degree view. In particular, it may be possible to capture the entire surface of the natural product 0 and thus take into account all features of the natural product 0 relevant for classification when comparing them with the stored reference information. This reduces the risk of incorrect classification.

[0124] The evaluation unit 14 is designed to compare optical images taken by the means for generating an optical image 12 with stored reference information and to assign a natural product to be classified to one of several virtual classes based on this comparison. The operation of the evaluation unit 14 is shown schematically in Figure 4.

[0125] The means for generating an optical image 12 includes a data transmission interface 26, which serves to transmit data to a first data transmission interface 28 of the evaluation unit 14. The evaluation unit 14 further includes an image processing unit 30, an I059PC2401

[0126] - 26 -

[0127] Storage unit 32 and a computing unit 34, for example designed as a computer processor.

[0128] The storage unit 32 contains the image processing unit 30, the reference information 36 stored as data, with which the optical images of the natural product produced by the means for generating an optical image 12 are compared, and process instructions 38 executed by the computing unit for the automated comparison of the images with the stored reference information 36.

[0129] When the classification of a quantity of elements of a natural product is started, the process instructions 38 for the automated comparison of the recordings are loaded into the computing unit 34 and followed. For example, the operator of the classifying device 10 may first be asked to input whether the classifying device 10 should be operated in a first operating mode, which enables rapid classification, or whether the classifying device 10 should be operated in a second operating mode, which enables training of the classification.

[0130] For example, when the classifying device 10 is operated in the first operating mode, the image processing unit 30 and the reference information 36 are loaded from the storage unit 32 into the computing unit 34. Furthermore, when the evaluation unit 14 receives optical images in the form of data from the camera 18, the process instructions 38 in the first operating mode provide that the images transmitted to the evaluation unit 14 are loaded directly into the image processing unit 30 running on the computing unit 34 and that the images are processed by the image processing unit 30 without storing them.

[0131] During processing, an optical image containing the previously described representations of the natural product from different perspectives is separated by the image processing unit 30. Each representation from one of the different perspectives is then processed further as a separate, dependent image. Each I059PC2401

[0132] - 27 - these separate, dependent images may optionally be assigned an identifier by means of which the images of a natural product from different perspectives can be matched to each other.

[0133] Subsequently, the image processing unit 30 automatically compares the optical images with the stored reference information 36. Optically perceptible features are detected in the images, these features are assigned to at least one predetermined evaluation criterion, and the values ​​of the detected features are further processed taking the assigned evaluation criterion into account. In the embodiment described here, for example, optically perceptible features relating to the evaluation criteria "color", "geometric shape", and "size" are detected in each of the separate, dependent images.

[0134] The recognition of features, the assignment of the recognized features to the evaluation criteria, and the identification of feature values ​​are carried out primarily by means of artificial intelligence performed by the image analysis. The assignment of the natural product shown in the images to a virtual class is also carried out by artificial intelligence, based on the aforementioned recognition steps.

[0135] The artificial intelligence can, for example, recognize feature patterns in the optical images and compare them with the stored reference information 36 loaded into the processing unit 34. The reference information 36 loaded into the processing unit 34 can, in particular, include datasets on feature patterns of known natural products. The features optically perceptible in the images are compared by the artificial intelligence with the known feature patterns, and based on this comparison, the artificial intelligence decides which features are recognizable in the images, which evaluation criterion these features are to be assigned to, and how pronounced a feature is with respect to a corresponding evaluation criterion. Based on these decisions, I059PC2401

[0136] - 28 -

[0137] The artificial intelligence uses comparison to determine which of the majority of virtual classes the natural product shown in the recording belongs to.

[0138] For example, the reference information can contain 36 limit values ​​for each of the aforementioned evaluation criteria "color," "geometric shape," and "size," the combination of which constitutes a feature pattern. Features related to these evaluation criteria are recognized by the artificial intelligence in the images because the AI ​​has been trained on these or similar features and associated evaluation criteria using known images of natural products. The AI ​​has learned from this training and stored the corresponding information as part of the reference information 36. The AI ​​recognizes the various manifestations of a feature in an optical image because it has also learned possible manifestations of a feature using known images.The characteristics of a feature visible in an optical image are interpreted by artificial intelligence, for example, in relation to learned threshold values. Based on the result of this comparison, the natural product depicted in the image, which is to be classified, is assigned to one of several virtual classes.

[0139] Using the aforementioned identifiers, the results of comparing images showing the same natural product from different perspectives can be combined. This combination serves as the basis for a final overall classification of the natural product. This final overall classification allows for the identification and reduction of classification errors resulting from a feature being poorly visible or not visible at all from one perspective.

[0140] By considering feature patterns and a multitude of evaluation criteria across multiple images of a natural product from different perspectives, without storing the images, artificial intelligence enables the classification of a natural product I059PC2401

[0141] - 29 - to a virtual class in a very short time and very reliably. In particular, the assignment of a natural product to a virtual class can thus be carried out within a few milliseconds, for example less than 25 milliseconds. Thus, for example, fifteen elements of a set of a natural product can be classified virtually and physically per second.

[0142] As mentioned above, the evaluation unit 14 for classifying natural products using artificial intelligence can be trained to perform machine learning. For this purpose, the classifying device can, for example, be operated in the second operating mode. In this second operating mode, the process instructions 38 stipulate that the images transmitted to the evaluation unit 14 are loaded into the image processing unit 30 running on the computing unit 34, and that the images are processed by the image processing unit 30. Furthermore, the process instructions 38 stipulate that the data of the optical images transmitted to the evaluation unit 14 are additionally stored in the storage unit 32 as stored images 40.As part of the comparison of the optical images with the reference information, the features visually perceived by the artificial intelligence in the images, the result of their assignment to evaluation criteria, and the result of the features' values ​​in relation to the evaluation criteria are also stored. After the data is saved, the artificial intelligence can be informed in the second operating mode whether the decisions made were correct or incorrect. The artificial intelligence learns based on this information. The artificial intelligence can be trained, in particular, using 300 optical images of known natural products. This training, in combination with the setup described here, leads to a reliable classification of natural products.

[0143] After a natural product has been assigned to a virtual class, the assignment result is displayed using a second I059PC2401

[0144] - 30 -

[0145] Data is transferred from data transmission interface 42 of the evaluation unit 14 to a data transmission interface 44 of the output unit 16.

[0146] For the purpose of physically outputting the natural product assigned to a virtual class in accordance with the evaluation, the output unit 16 has a manipulator 46, which can be seen in particular in Figures 1 to 3 with reference to the first embodiment of the classifying device 10. The manipulator 46 moves the natural product to be physically classified into one of several containers 48, which are assigned to the virtual class into which the natural product was classified by the evaluation unit 14. Thus, each of the several containers 48 contains only natural products of a single, predetermined class. The containers 48 are arranged below the output unit 16 in a circumferential direction around the output unit 16.

[0147] In the first embodiment of the classifying device 10 shown in Figures 1 to 3 and described here, the manipulator 46 has a hollow shaft rotatable about an axis of rotation and a deflecting device 50 designed as a hollow cone section that contacts the natural product to be classified. The hollow cone section 50 is connected to the rotatably mounted hollow shaft and can thus be aligned with the different containers 48 by means of a recess provided in a wall of the hollow cone section 50. The hollow shaft is connected to an actuator that can rotate the hollow shaft and the hollow cone section 50 about the axis of rotation. The actuator is in turn controlled by the computing unit 34 depending on the assignment result of a natural product to be classified to a virtual class.In the first embodiment of the classifying device 10 described here, the output unit 16 further comprises a plurality of guide channels 52 corresponding to the number of virtual classes. Each of the guide channels is arranged between a position in which the recess of the aligned hollow cone section 50 can be positioned and a container 48 associated with this position. The guide channels 52 can be arranged, in particular, in a radial direction.

[0148] - 31 - be designed to point away from the deflecting device 50 towards the containers 48.

[0149] The design of the deflection device 50 as a hollow cone section allows it to be aligned from a first position to a second position within a few milliseconds, for example, within less than 25 milliseconds. The rotational speed can be up to 7.2° per millisecond, for example, when rotating 180° within 25 milliseconds. The actuator is designed accordingly to execute this rotational speed. If, for example, fifteen elements of a set of natural products are to be classified per second, approximately 25 milliseconds are available in which an element to be classified is virtually assigned to the set of natural products and the deflection device 50 is aligned, and approximately 39 milliseconds in which the element to be classified can enter and exit the deflection device 50 after being assigned a virtual class.

[0150] In order to supply the natural products to the inspection chamber 24, the classification device 10 has an automatic separation unit 54 for the natural product to be classified. With the separation unit 54, an element can be taken from a quantity of a natural product stored in a storage container 56 and this element can be supplied to the automatic classification.

[0151] The separation device 54 comprises a vibratory spiral conveyor and an additional conveying device 58, which is designed, for example, as a conveyor belt. The vibratory spiral conveyor comprises a spiral track 60 formed on the wall of the storage container 56, on which the elements of the quantity of natural products can be conveyed, and a vibratory drive arranged, for example, below the storage container 56. Due to vibrations of the spiral track 60 generated by the vibratory drive, the elements of the quantity of natural products are successively conveyed along the spiral track 60 from the I059PC2401

[0152] - 32 -

[0153] The contents are removed from the storage container 56 and conveyed to the conveyor belt 58. The conveyor belt 58 can be operated at a higher conveying speed than the vibratory bowl feeder, so that the distance between two successively conveyed elements increases during the transition from the vibratory bowl feeder to the conveyor belt 58. This allows the distances between two elements to be flexibly adjusted, in particular so that fifteen elements of a quantity of a natural product can be classified per second. At a free end of the conveyor belt 58, an element to be classified can fall due to gravity, be detected by the object detector 22, and then proceed into the viewing chamber 24.

[0154] Figures 5 to 8 show a second embodiment of the classifying device 10. The second embodiment is very similar to the first embodiment, therefore only the differences between the two will be discussed. Otherwise, the two embodiments can be identical.

[0155] The second embodiment differs from the first embodiment in that the output unit 16 has an alternative manipulator 46. The manipulator 46 of the second embodiment has a contactless deflection device 62, which comprises a plurality of compressed air outlets 64 arranged in a ring, as can be seen particularly in Figure 8. The compressed air outlets 64 can, for example, correspond to at least the number of predetermined virtual classes. The compressed air outlets 64 are distributed radially and, in particular, equidistantly around the ring arranged below the viewing area 24. By means of the processing unit 34, the release of compressed air from one or more of the compressed air outlets 64 can be selectively controlled depending on the assignment result of a natural product to be classified to a virtual class.A natural product 0 to be classified is carried by the compressed air flowing from an activated compressed air outlet 64 in the direction of the arrows P2 shown in Figure 8 in the direction of the I059PC2401.

[0156] - 33 - also shown in Figure 8, the arrow P3 moves out of the manipulator 46 into one of several guide channels 52. The guide channel 52, into which the natural product 0 is moved, is located, in particular, on a side opposite the activated compressed air outlet 64. Each of the guide channels 52 leads radially away from the viewing area 24 to a container 48.

[0157] In contrast to the first embodiment, in the second embodiment the camera 18 is positioned in a different location, namely next to the storage container 56. The position of the camera 18 can, in principle, be chosen arbitrarily, as long as the light emitted by the mirror optics 20 can reach the sensor of the camera 18.

[0158] Figure 9 shows another embodiment of a classifying device 10 with a storage container 56 and a spiral track 60 analogous to the embodiments described above. Also analogous to the embodiments described above, the natural products are separated and then move individually along a flight path F (see Fig. 10), which extends to the entry area of ​​a dispensing unit 16. During the flight path F, at least one image is captured by means of an optical image 12, in particular by a camera 18.

[0159] A light barrier 72 detects that a natural product 0 is in the flight path F and triggers the creation of a recording using the camera 18.

[0160] In contrast to the preceding embodiments, the output unit 16 is a manipulator 46 with a deflecting device 50 that can be driven rotationally in two directions, which is explained below with reference to Figure 10. It is a contacting deflecting device 50 formed from a distribution element shaped such that a falling natural product O strikes it once and is then either directed into a container 48 or first into a pipe section 66 and via this pipe section 66 to a specific container.

[0161] - 34 -

[0162] The deflection device 50 is designed to direct the product 0 towards the outer circumference. In the illustrated embodiment, the deflection device 50 is implemented as a distribution element in the form of a curved distribution plate, which is rotatably movable and, depending on its position, allows the product to rebound only into a specific segment of the circular ring towards the outer circumference. In the illustrated embodiment, the product is divided into five different containers 48, which is why 72 degrees are provided for each segment of the circular ring. The deflection device 50 is accordingly designed such that the product 0 is conveyed towards the desired segment of the circular ring upon impact with the deflection device 50. The deflection device 50 is adjusted by rotation using an actuator (not shown).

[0163] Figure 11 shows an isolated representation of a means for generating an optical image 12 in the form of a camera 18. With this camera 18, an exemplary falling natural product 0 can be recorded from four viewing directions according to arrows B1, B2, B3, and B4. The path of a light beam 70 from the camera 18 to an exemplary mirror optic 20 is shown in Figure 11. The mirror optic 20 is a simple mirror optic, which here—viewed from the camera 18—is arranged behind the natural product 0. From this mirror optic, the light beam 70 is directed directly onto the natural product 0 according to arrow B1.Further, uninscribed light beams are reflected by means of other, unindicated mirror optics in such a way that the light rays hit the natural product O in accordance with arrows B2, B3 and B4, so that the entire surface of the natural product O with overlaps is visible in the resulting images.

[0164] The features of the invention disclosed in the present description, the drawings, and the claims can be essential for realizing the invention in its various embodiments, both individually and in any combination. The invention is not limited to the described embodiments. It can be found in I059PC2401

[0165] - 35 -

[0166] The scope of the requirements and taking into account the knowledge of the responsible expert may vary.

[0167] I059PC2401

[0168] Reference symbol list

[0169] 10 Classification from the attention

[0170] 12 Means of producing an optical image

[0171] 14 evaluation units

[0172] 16 output units

[0173] 18 Camera

[0174] 20 Mirror optics

[0175] 20a, 20b Mirrors of the mirror optics

[0176] 22 Object detector

[0177] 24 Viewing area

[0178] 26 Data transmission interface of the means for generating an optical image

[0179] 28 First data transmission interface of the evaluation unit

[0180] 30 image processing units

[0181] 32 storage units

[0182] 34 computing units

[0183] 36 stored reference pieces of information

[0184] 38 procedural rules

[0185] 40 saved recordings

[0186] 42 Second data transmission interface of the evaluation unit

[0187] 44 Data transmission interface of the output unit

[0188] 46 Output unit manipulator

[0189] 48 containers

[0190] 50 contacting deflection device

[0191] 52 guide channels

[0192] 54 Separation device

[0193] 56 storage containers

[0194] 58 additional transport equipment, conveyor belt

[0195] 60 spiral track

[0196] 62 contactless deflection device

[0197] 64 compressed air outlets

[0198] 66 Pipe section

[0199] 70 light beams I059PC2401

[0200] - 37 -

[0201] 72 Light barrier

[0202] B1 Arrow (viewing direction 1 from the camera)

[0203] B2 Arrow (viewing direction 2 from the camera)

[0204] B3 Arrow (viewing direction 3 from the camera)

[0205] B4 Arrow (viewing direction 4 from the camera)

[0206] F Flight route

[0207] 0 object; natural product

[0208] P1 Arrow (direction of movement in free fall)

[0209] P2 Arrow (flow direction of compressed air)

[0210] P3 Arrow (trajectory when deflected by compressed air)

Claims

I059PC2401 - 38 - Patent claims 1. Classifying device (10) for classifying grown natural products (0), comprising - at least one means of producing an optical image (12) of a natural product (0) to be classified from two or more different perspectives, - an evaluation unit (14) which is designed to compare the optical images with stored reference information (36) and to assign the natural product (0) to be classified to one of several virtual classes on the basis of the comparison, and - an output unit (16) which is designed to physically classify the natural product (0) assigned to the virtual class according to the evaluation.

2. Classifying device (10) according to claim 1 , characterized in that the means for generating an optical image (12) of the natural product (0) to be classified is designed to capture the natural product (0) from at least four different perspectives.

3. Classification device (10) according to one of claims 1 or 2, characterized in that the means for generating an optical image (12) of the natural product (0) to be classified comprises a camera (18) and a mirror optic (20).

4. Classifying device (10) according to the preceding claim, characterized in that the camera (18) is aligned with the mirror optics (20) and the mirror optics (20) are designed such that the natural product (0) to be classified can be recorded from the two or more perspectives of the camera (18) by means of a single recording.

5. Classifying device (10) according to one of the preceding claims, characterized in that the output unit (16) is provided with a I059PC2401 - 39 - A flight path (F) is provided, which can be traversed by the natural products (0) to be classified, wherein the means for generating an optical image (12) is designed in such a way that it detects the natural product (0) to be classified in the area of ​​the flight path (F).

6. Classifying device (10) according to one of the preceding claims, characterized in that at least a subsection of the area in which images are generated by means of generating an optical image (12) has a configurable and / or monochromatic background design.

7. Classification device (10) according to one of the preceding claims, characterized in that the evaluation unit (14) is designed to be able to perform machine learning and / or to assign the natural product (0) to be classified to the virtual class by means of artificial intelligence.

8. Classification device (10) according to one of the preceding claims, characterized in that at least one separation device (54) is provided for the natural product (0) to be classified.

9. Classifying device (10) according to the preceding claim, characterized in that the at least one separation device (54) comprises at least one spiral conveyor.

10. Classifying device (10) according to one of the two preceding claims, characterized in that the separation device (54) is designed to handle grown natural products (0) with a size of 0.3 cm to 10 cm.

11. Classifying device (10) according to one of the preceding claims, characterized in that the output unit (16) is configured to output the natural product (O) assigned to the virtual class in a radial direction.

12. Classifying device (10) according to the preceding claim, characterized in that a plurality of containers (48) are provided for the classified storage of the product dispensed by the dispensing unit (16). I059PC2401 - 40 - natural product (0) and / or guide channels (52) and / or pipe sections leading to containers (48) are arranged in a circumferential direction around the output unit (16).

13. Classifying device (10) according to the preceding claim, characterized in that the output unit (16) has a manipulator (46).

14. Classifying device (10) according to the preceding claim, characterized in that the manipulator (46) has a deflecting device (50) that can be driven rotatorily in two directions of rotation.

15. Method for classifying grown natural products (O), characterized in that a) in a first method step a natural product (O) to be classified is fed to a flight path (F), b) while passing the flight path (F) at least one image is created from two or more different perspectives using a means for generating an optical image (12) of the natural product (O) to be classified, c) with an evaluation unit (14) the optical images are compared with stored reference information (36) and the natural product (O) to be classified is assigned to one of several virtual classes on the basis of the comparison, and d) with an output unit (16) the natural product (O) assigned to the virtual class is physically output in a classified form according to the evaluation by bringing the output unit (16) into an output configuration associated with the assignment before leaving the flight path (F). * * * * * * *

Images