Method and device for automatically determining at least one quality parameter that characterizes the processing of part of a carcass

Abstract

The invention relates to a method for automatically determining at least one quality parameter that characterizes the processing of part of a carcass, comprising the following steps: conveying the part of the carcass, as input product, in a conveying direction using a conveyor (11) to a first processing machine (12) for carrying out a first processing step for processing the part of the carcass to form an output product, recording input sensor data from the part of the carcass using at least one sensor apparatus (12) prior to carrying out the first processing step, classifying partial data relevant to the processing quality of the first processing step from the input sensor data, ascertaining the at least one quality parameter on the basis of the partial data and providing the at least one quality parameter to be displayed and / or forwarded to a downstream processing machine designed to carry out at least one further processing step on the basis of the ascertained quality parameter. The invention also relates to a corresponding device.

Description

[0001] METHOD AND DEVICE FOR AUTOMATICALLY DETERMINING AT LEAST ONE CHARACTERIZING THE PROCESSING OF A CARCASS PART

[0002] QUALITY PARAMETERS

[0003] Description

[0004] The present invention relates to a method for automatically determining at least one quality parameter characterizing the processing of a carcass part. Furthermore, the invention relates to a device for automatically determining at least one quality parameter characterizing the processing of a carcass part.

[0005] Such methods and devices are used in the automated processing and / or handling of slaughtered animals or parts of slaughtered animals, particularly in the processing of slaughtered poultry. In the following, we will always refer to carcass parts, which are used for simplification as a synonym for parts of a slaughtered animal carcass, body parts, and / or slaughtered animals.

[0006] It is well known in the art to use sensors to detect carcass parts or parts of the carcass prior to their automatic processing and to determine the processing-relevant parameters of the carcass parts in order to then optimally adapt the subsequent automatic processing process to the respective carcass part. The sensors are designed for this purpose, for example, as optical or X-ray sensitive imaging sensors in order to determine, for example, the external dimensions, positions and location of bone structures or the like of the carcass part. The processing-relevant parameters thus recorded then form the basis for adapting the respective processing process. In this way, it is possible to automatically control processing tools and adapt them to the conditions of the respective carcass part or the product to be processed.A disadvantage of the known methods and devices is that they are exclusively designed to sensor-basedly detect physical parameters inherently determined by the carcass to be processed, such as weight, dimensions, external contour, bone position, and the like, and to use these parameters as a starting point for further processing. The extent to which these parameters influence the respective upcoming processing process remains completely unconsidered.

[0007] The object of the present invention is therefore to propose a method that ensures an assessment of the carcass part to be processed with regard to its suitability for processing before a processing step is carried out. Furthermore, the object of the invention is to propose a method that ensures such an assessment of the highest possible quality for at least one part product obtained from the carcass after processing. Furthermore, the object is to propose a corresponding device.

[0008] The object is achieved by the method mentioned at the outset in that it comprises the steps of conveying the carcass part as an input product in a conveying direction by means of a conveying device to a first processing machine for carrying out a first processing step for processing the carcass part into an output product, detecting input sensor data from the carcass part by means of at least one sensor device before carrying out the first processing step, classifying partial data relevant to processing quality for the first processing step from the input sensor data, determining the at least one quality parameter on the basis of the partial data and providing the at least one quality parameter for display and / or forwarding to a downstream processing machine set up to carry out at least one further processing step on the basis of the determined quality parameter.By determining at least one quality parameter, it is possible for the first time to assess the quality of a carcass part with regard to its processing requirements. The carcass part to be processed is thus subjected to an evaluation that is always related to the respective processing step. In this way, the quality parameter provides a value that provides information about the carcass part with regard to its processability and / or suitability for this purpose. A useful embodiment of the invention is characterized in that the classification of the partial data is carried out algorithmically using predetermined selection parameters. This offers the advantage that the classification of the partial data is based on predefined algorithms that can be adjusted by specifying the selection parameters for the respective processing step.The advantage of algorithmic classification is, among other things, that the complexity of the calculation can be kept relatively low, which has a beneficial effect on the computing power required. Another advantage is that such algorithmic classification delivers clearly reproducible results. Furthermore, prior training is not required. Selection parameters can, for example, be different sections or areas of the carcass. If the carcasses are poultry carcasses, for example, the selection parameters can be different areas, such as the breast, especially the right and left breast fillet areas, legs, neck, hip, etc.

[0009] Typical carcass parts, particularly in the area of ​​poultry carcass processing, include wings and / or legs. These include, in particular, wing tips or tipped wings, the shoulder, elbow, hip, and knee joints, and the usual poultry products defined by these joints, such as leg quarters and anatomical legs, as well as the products obtained from them during further processing, such as drums, chops, drumsticks, and / or tights. When processing poultry carcasses, the products obtained include, for example, front halves, breast caps, whole fillets, half fillets, breast fillets, or tenderloins.

[0010] A preferred development of the invention is characterized in that the partial data is classified using a pre-trained neural classification network. This offers the advantage that artificial intelligence is used for classification, so that the development and definition of classification algorithms is not necessary. Furthermore, such a neural classification network is fundamentally capable of classifying and classifying even unclear or ambiguous input sensor data. Furthermore, this offers the advantage of delivering consistent results with regard to the quality parameters to be determined, despite the great variability of biological products with regard to the input data. A further expedient development of the invention is characterized in that the determination of the at least one quality parameter is carried out algorithmically based on predetermined quality assessment parameters.Such predetermined quality assessment parameters include, for example, the shape, weight, condition of the product surface, degree of superficial injuries, such as blood stains, etc. The aforementioned advantages of algorithmic classification also apply analogously to the algorithmic determination of at least one quality parameter.

[0011] A further advantageous embodiment of the invention is characterized in that the at least one quality parameter is determined using a pre-trained neural quality assessment network. The aforementioned advantages of the neural classification network also apply analogously to the neural quality assessment network for determining the at least one quality parameter.

[0012] Typical quality parameters, particularly in poultry processing, include residues from the previous processing step. For example, during the defeathering step, the proportion of feathers remaining on the poultry carcass can be a quality parameter. Another quality parameter can be the degree of processing-induced injuries to the product surface that have a negative impact on quality. Examples of this include detached skin, exposing the underlying muscle tissue. Bloodstains in the skin and upper tissue area, bone and / or joint fractures, particularly on the extremities of the carcasses, are other examples that can be used as quality parameters.

[0013] According to a further preferred embodiment of the invention, the quality assessment network comprises a predetermined number of neural subnetworks, each of the neural subnetworks being pre-trained to determine at least one of the quality parameters. In other words, the subnetworks advantageously form expert networks that are designed and configured to determine the respective quality parameter.

[0014] According to a further preferred embodiment, the input sensor data always includes image data of the slaughtered animal carcass part. Advantageously, the input sensor data can thus be captured contactlessly. In particular, it is possible to record the image data of the slaughtered animal carcass parts during their continuous conveyance.

[0015] A preferred development of the invention is characterized in that the sensor device records the image data in the wavelength range of visible, infrared, ultraviolet, and / or the spectral range of X-rays. By selecting the respective wavelength range, it is possible to obtain image data that represents an optimal starting point for determining the respective quality parameter.

[0016] A further advantageous embodiment of the invention is characterized in that the input sensor data at least partially depicts the carcass part in three dimensions. This offers the advantage that the dimensions of the carcass part, or at least partial dimensions thereof, can be recorded and used for further evaluation.

[0017] According to a further preferred embodiment of the invention, an estimated weight and / or carcass dimension of the starting product is determined as at least one of the quality parameters. Advantageously, core quality parameters that are of central importance for reliability and meaningfulness are determined in this way.

[0018] According to a further preferred embodiment, at least one of the quality parameters is a quality value assigned to the starting product, which represents a measure of the expected final quality of the starting product after its processing. Advantageously, the quality value thus determines a quantity that provides information about the expected quality of the starting product following processing. The quality value is thus a measure of the expected final quality after the respective processing step.

[0019] A further advantageous embodiment of the invention is characterized in that, as at least one of the quality parameters, an overall quality value is determined, which is composed of a predetermined first component related to the estimated weight and / or carcass dimension and a predetermined second component related to the quality value. In other words, the overall quality value is determined by means of a linear combination of the estimated weight and / or carcass dimension according to the predetermined first and second components. This offers the advantage that the influence of the estimated weight and carcass dimension can be specified on a customer-specific basis and adapted to the respective needs.

[0020] A suitable embodiment of the invention is characterized in that the neural classification network is trained by means of the following steps: providing a plurality of carcasses, detecting the input sensor data from each of the carcass parts by means of the at least one sensor device before carrying out the first processing step, classifying partial data relevant to processing quality for the first processing step from the input sensor data by evaluating the input sensor data by a human training person, defining classification areas set up for the classification of the relevant partial data on the basis of the partial data by the human training person, providing training data for training the neural classification network by inputting the input sensor data as input data and the respective classification areas as target data for the neural classification network,Repeated adjustment of the classification network's weights based on the difference between the target data and the output data generated by the neural network. This advantageously makes it possible to pre-train the classification network's functionality with regard to optimal classification of the relevant data subsets.

[0021] A preferred development of the invention is characterized in that the neural qualification network is trained as follows: providing a plurality of carcasses, capturing the input sensor data from each of the carcass parts by means of the sensor device before carrying out the first processing step, classifying partial data relevant to the processing quality for the first processing step from the input sensor data by evaluating the input sensor data by a human trainer, determining at least one quality parameter based on the partial data by the human trainer for each of the carcass parts, providing training data for training the neural quality assessment network by inputting the input sensor data as input data and the respective quality parameters as target data for the neural quality assessment network,Repeated adjustment of the weights of the quality assessment network based on the difference between the target data and the output data generated by the neural network. This advantageously makes it possible to pre-train the functionality of the quality assessment network with regard to optimal determination of the respective quality parameters.

[0022] Furthermore, the object is achieved by the device mentioned at the outset in that a conveyor device configured to convey the carcass part as an input product in a conveying direction to a first processing machine, the processing machine configured to carry out a first processing step for processing the carcass part into an output product, at least one sensor device configured to record input sensor data from the carcass part before carrying out the first processing step, a classification device configured to classify partial data relevant to the processing quality for the first processing step from the input sensor data,wherein the classification device is designed and configured to determine the at least one quality parameter on the basis of the partial data and to provide the at least one quality parameter for display and / or forwarding to a downstream processing machine configured to carry out at least one further processing step on the basis of the determined quality parameter.

[0023] The advantages achievable with the present invention have already been described in detail in connection with the method according to the invention. To avoid repetition, we also refer to the advantages mentioned there in connection with the device according to the invention, which apply equally to the method claims, which are essentially drafted analogously to the method. Therefore, only selected aspects of the device according to the invention will be discussed separately below.

[0024] A preferred development of the invention is characterized in that the classification device is designed such that the classification of the partial data is carried out algorithmically on the basis of predetermined selection parameters.

[0025] According to a further preferred embodiment of the invention, the classification device is designed such that the partial data is classified using a pre-trained neural classification network. A further expedient embodiment of the invention is characterized in that the classification device is designed such that the at least one quality parameter is determined algorithmically based on predetermined quality assessment parameters.

[0026] According to a further preferred embodiment, the classification device is designed such that the determination of the at least one quality parameter is carried out by means of a pre-trained neural quality assessment network.

[0027] A further expedient embodiment of the invention is characterized in that the quality evaluation network comprises a predetermined number of neural sub-networks, wherein each of the neural sub-networks is pre-trained to determine one of the quality parameters.

[0028] According to a further preferred embodiment of the invention, the input sensor data in any case comprise image data of the carcass part.

[0029] A further expedient embodiment of the invention is characterized in that the at least one sensor device is configured such that the image data are recorded in the wavelength range of visible, infrared, ultraviolet and / or the spectral range of X-ray radiation.

[0030] According to a further preferred embodiment, the input sensor data, at least in part, map the carcass part three-dimensionally.

[0031] A further preferred development of the invention is characterized in that the classification device is designed to determine an estimated weight and / or a carcass dimension of the starting product as at least one of the quality parameters.

[0032] According to a further preferred embodiment of the invention, the classification device is designed to determine, as at least one of the quality parameters, a quality value assigned to the starting product, which represents a measure of the expected final quality of the starting product after its processing. A further expedient embodiment of the invention is characterized in that the classification device is designed to determine, as at least one of the quality parameters, an overall quality value, which is composed of a predetermined first component related to the estimated weight and / or carcass dimension and a predetermined second component related to the quality value.

[0033] According to a further preferred embodiment, the neural classification network is trained according to the aforementioned steps.

[0034] A further expedient embodiment of the invention is characterized in that the neural quality determination network is trained according to the aforementioned steps.

[0035] Further preferred and / or expedient features and embodiments of the invention emerge from the dependent claims and the description. Particularly preferred embodiments are explained in more detail with reference to the accompanying drawings. The drawing shows:

[0036] Fig. 1 is a schematic representation of the method according to the invention and the device according to the invention,

[0037] Fig. 2 is a schematic representation of the process of training the neural networks for classifying partial data,

[0038] Fig. 3 is a schematic representation of the process of training the neural networks to determine the quality parameters,

[0039] Fig. 4 a schematic representation for determining the quality parameters using algorithms and

[0040] Fig. 5 shows a schematic representation for determining quality parameters using pre-trained neural networks. The method and device according to the invention are explained below by way of example with reference to the schematic view shown in Fig. 1. The method and device are designed and configured for the automatic determination of at least one quality parameter characterizing the processing of a carcass part.

[0041] The carcass part is conveyed by means of a conveyor device 10 in a conveying direction to a first processing machine 11. The carcass part thus forms an input product of the first processing machine 11, which is configured to perform a first processing step by means of which the carcass part is processed as an input product into an output product.

[0042] By means of a sensor device 12, input sensor data from the carcass part is acquired in a capture step 13 before the first processing step is carried out. For example, in the capture step 13, image data is captured and extracted, which advantageously includes the visible light spectrum 14, the UV spectrum 15, the IR spectrum 16, the X-ray spectrum 17, and / or depth or 3D information 18. Further preferably, the aforementioned image components containing spectral or depth and 3D information are summed to form a composite image by means of an addition unit 19. The individual components can be weighted as desired according to a predetermined rule in order to particularly emphasize features in the image data that are relevant for the respective processing.

[0043] In a subsequent classification step 20, processing quality-relevant partial data are determined from the input data for the first processing step and are preferably generated as object-specific partial data 22. The generation of the object-specific partial data 22 also includes, in particular, a division 21 of the input data classified in the classification step 20.

[0044] In a quality parameter determination step 23, the at least one quality parameter is subsequently determined based on the partial data 22. This at least one quality parameter represents a measure 24 for the expected (final) quality of the starting product after its processing. In a provision step 25, the at least one determined quality parameter is provided for display and / or forwarding to a downstream processing machine configured to carry out at least one further processing step based on the determined quality parameter.

[0045] Preferably, the partial data is classified algorithmically – as shown in Figure 1 with arrow 26 – using predetermined selection parameters. Algorithmic means that the partial data is classified using predefined algorithms. For this purpose, for example, selection parameters are specified and thus optimally adjusted for the respective processing step. Selection parameters are, for example, different sections or areas of the slaughtered animal carcass. If, for example, the slaughtered animal carcasses are poultry carcasses, the selection parameters are, for example, different areas, such as the breast, in particular the right and left breast fillet areas, legs, neck, hip, or the like. More preferably, the partial data is classified – as shown in Figure 1 with arrow 27 – using a pre-trained neural classification network.Further preferably, the determination of the at least one quality parameter is carried out algorithmically based on predetermined quality assessment parameters.

[0046] Fig. 4 shows an example of an advantageous embodiment of the method 28 for determining quality parameters by means of at least one algorithm. The object-specific partial data 22 are made available, for example, to the partial algorithms 29, 30, 31 as input data. By means of these partial algorithms, a first quality parameter 32, a second quality parameter 33 and an Nth quality parameter 34 are determined, where N is a natural number greater than zero. For example, the first quality parameter 32 is an estimated weight, while the second quality parameter represents the quality of the input product with regard to its expected final quality after processing. A total quality parameter 37 can be determined from the aforementioned first and second quality parameters 32, 33 by linear combination, ie by weighting with respective weighting factors 35, 36.The formation of such a total quality parameter 37 is not limited to the shown summation of two quality parameters 32, 33, but comprises any number of quality parameters as summands. It is also possible not to include individual quality parameters 31 in the summation, but to provide them as a further quality parameter 34. Fig. 5 shows an example of an advantageous embodiment of the method 38 for determining quality parameters using at least one pre-trained neural quality assessment network. In particular, the quality assessment network comprises a predetermined number of neural subnetworks 39, 40, 41, wherein each of the neural subnetworks 39, 40, 41 is pre-trained to determine at least one of the quality parameters. The object-specific partial data 22 are made available, for example, to the neural subnetworks 39, 40, 41 as input data.

[0047] The first quality parameter 32, the second quality parameter 33, and the Nth or further quality parameter 34 are determined by means of the neural subnetworks 39, 40, 41. As already described, the first quality parameter 32 is, for example, the estimated weight, while the second quality parameter represents the quality of the input product with regard to its expected final quality after processing. The total quality parameter 37 can be determined from the aforementioned first and second quality parameters 32, 33 by linear combination, i.e., by weighting with respective weighting factors 35, 36. The formation of such a total quality parameter 37 is not limited to the shown summation of two quality parameters 32, 33, but includes any number of quality parameters as summands.It is also possible not to include individual quality parameters 31 in the summation, but to provide them as additional quality parameters 34. The neural subnetworks 39, 40, 41 are therefore designed as expert networks that are configured to determine the respective quality parameters.

[0048] Instead of summing, the quality parameters 32, 33, 34 and possibly further quality parameters can also be fed to another neural network - not shown in the drawing - which is set up to determine the sum quality parameter 37.

[0049] Preferably, both in the algorithmic and in the neural network-based determination of at least one of the quality parameters, the estimated weight and / or the dimension, i.e., the spatial extent / size, of the starting product is determined. Further preferably, the at least one quality parameter is determined as a quality value assigned to the starting product, which represents a measure of the expected final quality of the starting product after its processing. The aforementioned total quality parameter 37 preferably corresponds to an overall quality value, which is composed of a predetermined first component related to the estimated weight and / or carcass dimension and a predetermined second component related to the quality value.

[0050] Figure 2 schematically shows the training of the neural classification network. First, a plurality of carcasses are provided, and in a detection step 13, input sensor data from each of the carcass parts is detected by the at least one sensor device 12 before the first processing step is carried out. Subsequently, in a classification step 42, the input sensor data is evaluated and divided into partial data relevant for processing by a human training person. In other words, partial data relevant to processing quality for the first processing step is classified from the input sensor data by the human training person evaluating the input sensor data. This person defines classification areas 43 based on the partial data, in particular for classifying the relevant partial data.

[0051] From the input sensor data 13, in conjunction with the classification regions 43, corresponding training data 44 is provided for training the neural classification network. The training 45 of the neural classification network is then performed by inputting the input sensor data as input data and the respective classification regions 43 as target data for the neural classification network. The weights of the neural network are adjusted step by step by repeatedly adjusting the weights of the classification network based on the difference between the target data and the output data generated by the neural network.

[0052] Figure 3 schematically shows the training of the neural qualification network. First, in a recording step 13, input sensor data from each of the carcass parts is recorded using the at least one sensor device 12 before the first processing step is performed.

[0053] Subsequently, in the evaluation step 46, a human trainer evaluates the input sensor data and defines the relevant quality parameters for processing. Based on the partial data, this trainer defines classification areas 43, in particular, for classifying the relevant partial data. In other words, at least one quality parameter 32, 33, 34 is defined for each of the carcass parts by the human trainer based on the input sensor data.

[0054] The training 45 of the neural classification network is then carried out by inputting the input sensor data as input data and the respective quality parameters 32, 33, 34 as target data for the neural quality evaluation network and repeatedly adjusting the weights of the quality evaluation network based on the difference between the target data and the output data generated by the neural network.

[0055] All statements regarding the functioning of the method according to the invention also apply analogously to the functioning of the device according to the invention, and vice versa. To avoid repetition, only the device or the method will be discussed in detail in each section; however, the statements always apply to both the device and the method.

Claims

1. A method for automatically determining at least one quality parameter characterising the processing of a carcass part, comprising the steps of: Conveying the carcass part as an input product in a conveying direction by means of a conveying device (11) to a first processing machine (12) for carrying out a first processing step for processing the carcass part into a starting product, Acquiring input sensor data from the carcass part by means of at least one sensor device (12) before carrying out the first processing step Classifying partial data relevant for the first processing step from the input sensor data, Determining at least one quality parameter based on the partial data and Providing the at least one quality parameter for display and / or forwarding to a downstream processing machine configured to carry out at least one further processing step on the basis of the determined quality parameter.

2. Method according to claim 1, characterized in that the classification of the partial data is carried out algorithmically on the basis of predetermined selection parameters.

3. Method according to claim 1, characterized in that the classification of the partial data is carried out by means of a pre-trained neural classification network.

4. Method according to one of claims 1 to 3, characterized in that the determination of the at least one quality parameter is carried out algorithmically on the basis of predetermined quality evaluation parameters.

5. Method according to one of claims 1 to 3, characterized in that the determination of the at least one quality parameter is carried out by means of a pre-trained neural quality assessment network.

6. The method according to claim 5, characterized in that the quality evaluation network comprises a predetermined number of neural sub-networks (39, 40, 41), each of the neural sub-networks (39, 40, 41) being pre-trained to determine at least one of the quality parameters.

7. Method according to one of claims 1 to 6, characterized in that the input sensor data in any case comprise image data of the carcass part.

8. The method according to claim 7, characterized in that the sensor device records the image data in the wavelength range of visible, infrared, ultraviolet and / or the spectral range of X-ray radiation.

9. Method according to one of claims 7 or 8, characterized in that the input sensor data at least partially depict the carcass part three-dimensionally.

10. Method according to one of claims 4 to 9, characterized in that an estimated weight and / or a carcass dimension of the starting product is determined as at least one of the quality parameters.

11. Method according to one of claims 4 to 10, characterized in that as at least one of the quality parameters a quality value assigned to the starting product is determined, which represents a measure of the expected final quality of the starting product after its processing.

12. Method according to claim 10, characterized in that as at least one of the quality parameters a total quality value (37) is determined, which is composed of a predetermined first portion relating to the estimated weight and / or the carcass dimension and a predetermined second portion relating to the quality value.

13. Method according to one of claims 3 to 12, characterized in that the neural classification network is trained by means of the following steps: Providing a variety of carcasses, Capturing the input sensor data from each of the carcass parts by means of the at least one sensor device (12) before carrying out the first processing step (11) Classifying partial data relevant for the first processing step from the input sensor data by evaluating the input sensor data by a human training person, Determination of classification areas set up for the classification of the relevant partial data on the basis of the partial data by the human training person, Providing training data for training the neural classification network by entering the input sensor data as input data and the respective classification areas as target data for the neural classification network, repeatedly adjusting the weights of the classification network based on the difference between the target data and the output data generated by the neural network.

14. Method according to one of claims 5 to 13, characterized in that the neural qualification network is trained as follows: Providing a variety of carcasses, detecting the input sensor data from each of the carcass parts by means of the sensor device (12) before carrying out the first processing step (11), Classifying partial data relevant for the first processing step from the input sensor data by evaluating the input sensor data by a human trainer, Determination of at least one quality parameter based on the part data by the human trainer for each of the carcass parts Providing training data for training the neural quality evaluation network by inputting the input sensor data as input data and the respective quality parameters as target data for the neural quality evaluation network, repeatedly adjusting the weights of the quality evaluation network based on the difference between the target data and the output data generated by the neural network.

15. Device for automatically determining at least one quality parameter characterising the processing of a carcass part, comprising a conveyor device (10) configured to convey the carcass part as an input product in a conveying direction to a first processing machine (11), the processing machine (11) configured to carry out a first processing step for processing the carcass part into an output product, at least one sensor device (12) configured to acquire input sensor data from the carcass part before carrying out the first processing step, a classification device configured to classify partial data relevant to processing quality for the first processing step from the input sensor data,wherein the classification device is designed and configured to determine the at least one quality parameter on the basis of the partial data and to provide the at least one quality parameter for display and / or forwarding to a downstream processing machine configured to carry out at least one further processing step on the basis of the determined quality parameter.

16. Device according to claim 15, characterized in that the classification device is designed such that the classification of the partial data is carried out algorithmically on the basis of predetermined selection parameters.

17. Device according to claim 16, characterized in that the classification device is designed such that the classification of the partial data is carried out by means of a pre-trained neural classification network.

18. Device according to one of claims 15 to 17, characterized in that the classification device is designed such that the determination of the at least one quality parameter is carried out algorithmically on the basis of predetermined quality evaluation parameters.

19. Device according to one of claims 15 to 17, characterized in that the classification device is designed such that the determination of the at least one quality parameter is carried out by means of a pre-trained neural quality assessment network.

20. Device according to claim 19, characterized in that the quality evaluation network comprises a predetermined number of neural sub-networks (39, 40, 41), each of the neural sub-networks being pre-trained to determine one of the quality parameters.

21. Device according to one of claims 15 to 20, characterized in that the input sensor data in any case comprise image data of the carcass part.

22. Device according to claim 21, characterized in that the at least one sensor device (12) is configured such that the image data are recorded in the wavelength range of visible, infrared, ultraviolet and / or the spectral range of X-ray radiation.

23. Device according to one of claims 21 or 22, characterized in that the input sensor data at least partially depict the carcass part three-dimensionally.

24. Device according to one of claims 19 to 24, characterized in that the classification device is designed as at least one of the Quality parameters to determine an estimated weight and / or carcass dimension of the starting product.

25. Device according to one of claims 19 to 24, characterized in that the classification device is designed to determine, as at least one of the quality parameters, a quality value assigned to the starting product, which represents a measure of the expected final quality of the starting product after its processing.

26. Device according to claim 25, characterized in that the classification device is designed to determine, as at least one of the quality parameters, a total quality value (37), which is composed of a predetermined first portion with respect to the estimated weight and / or the carcass dimension and a predetermined second portion with respect to the quality value.

27. Device according to one of claims 15 to 26, characterized in that the neural classification network is trained according to the method according to claim 13.

28. Device according to one of claims 15 to 27, characterized in that the neural quality determination network is trained according to the method according to claim 14.

Images