Tissue structure analysis system

Abstract

A method for automatically processing meat products is provided and includes using a first vision system to acquire images and generate models of meat products, generating a database containing these images and models, acquiring an image of a meat product to be processed using a second vision system, transmitting the image to a prediction system configured to follow an algorithm to associate the image with the images and / or the models contained on the database and generate a prediction model of the meat product. The prediction model is then transmitted to an analysis system configured to generate an integrated model comprising combined features of the image and the prediction model, the analysis system being configured to define processing instructions for the processing of the meat product taking into consideration the combined features of the integrated model. The processing instructions are transmitted to a meat processing machine for processing the meat product.

Description

TISSUE STRUCTURE ANALYSIS SYSTEMTECHNICAL FIELD

[0001] The technical field generally relates to meat processing techniques, and more particularly relates to tissue structure analysis for improving processing operations.BACKGROUND

[0002] Modern meat packing plant automation systems heavily rely on different scanning techniques to increase the performance of its process. Various means of scanning, such as 3D scanning, either by laser triangulation or other means, is often used as input, and from which the process can be refined. Most of these scanning technologies can only scan the surface of the processed material, whereas many applications must account for the presence of underlying structures of the workpieces (e.g., pieces of meat and / or carcasses). Some technologies, such as X-Ray scan and ultrasound scan will detect underlying structures.

[0003] For instance, US patent No. 11,388,905 (to MAREL MEAT B.V.) describes a method of processing a half pig carcass consisting of detecting positions of various bones in the half pig carcass prior to cutting the half pig carcass using a cutting device. The detection of the bone positions, being indicative of the inner bone structure of the half pig carcass, is accomplished using an X-ray apparatus for capturing X-ray data. Other documents describing similar methods and making use of X-ray and / or ultrasound scanning techniques include US 11,399,550 (to JOHN BEAN TECH CORP), US 10,631,548 and US 11,785,955 (to SCOTT AUTOMATION & ROBOTICS PTY LTD), among others.

[0004] However, known scanning techniques and equipment such as X-ray and / or ultrasound machines come with important disadvantages including requiring a large footprint on the production line, slow operational speeds (e.g., slow data acquisition), high acquisition costs and similarly high installation and maintenance costs. It is thus noted that there is currently a lack of low-cost solutions for obtaining internal measurements that could assist the performance of various meat processing machines.

[0005] There are therefore various challenges and there is a need for enhanced technologies in this field.SUMMARY

[0006] According to an aspect, a method for automatically processing meat products is provided. The method includes acquiring at least one image of a meat product to be processed; transmitting the at least one image to a prediction system configured to follow an algorithm to:generate a prediction model of the meat product to be processed comprising at least one of:o predicted underlying tissue structures; ando predicted optimized or near-optimized processed specifications; transmitting the prediction model to an analysis system configured to generate an integrated model comprising combined features of the at least one image and the prediction model, the analysis system being further configured to define processing instructions for the processing of the meat product taking into consideration the combined features of the integrated model; and transmitting the processing instructions to a meat processing machine configured to automatically process the meat product for obtaining a final product.

[0007] According to a possible embodiment, the at least one image includes at least one of a 2-dimensional image, a 3-dimensional image, laser / illumination wavelengths and a laser line scatter data set.

[0008] According to a possible embodiment, the at least one image comprises external characteristics of the meat product to be processed.

[0009] According to a possible embodiment, the external characteristics comprise product specifications including any one or combination of a shape of the meat product, a size of the meat product, a weight of the meat product, dimensions of various parts of the meat product and relative dimensions between parts of the meat product.

[0010] According to a possible embodiment, the external characteristics comprise product features including any one or combination of a nature or type of the meat product, a quality of the meat product, a presence of defects, defect specifications including size and color, and relative defect location on the meat product.

[0011] According to a possible embodiment, the combined features of the integrated model includes relative positions between the external characteristics of the meat productand the predicted underlying tissue structures to predict a location of the underlying tissue structures within the meat product.

[0012] According to a possible embodiment, the processing instructions include generating one or more cutting paths along the meat product taking into consideration the location of the underlying tissue structures in order to avoid the underlying tissue structures during operation of the meat processing machine.

[0013] According to a possible embodiment, the combined features of the integrated model includes relative positions between the external characteristics of the meat product and the predicted optimized or near-optimized processed specifications to assist in generating the processing instructions adapted to enable operation of the meat processing machine to process the meat product for obtaining the final product having specifications similar to the predicted optimized or near-optimized processed specifications.

[0014] According to a possible embodiment, the algorithm includes pattern recognition to identify one or more external characteristics of the meat product to be processed.

[0015] According to a possible embodiment, the algorithm integrates artificial intelligence to identify one or more external characteristics of the meat product to be processed.

[0016] According to a possible embodiment, the artificial intelligence includes machine learning.

[0017] According to a possible embodiment, the machine learning includes deep learning, where multiple layers of processing are used to extract progressively higher-level features from data.

[0018] According to a possible embodiment, the method further includes, in a training step, providing the prediction system access to a database containing at least one of:• models of meat products with underlying tissue structures; and• models of processed meat products with optimized or near-optimized processed specifications,in order to train the algorithm of the prediction system in pattern recognition.

[0019] According to a possible embodiment, training the algorithm of the prediction system in pattern recognition includes gathering information in order to take into accountthe presence of bones, fat layers, cartilage and / or muscle groups of the meat products and / or processed meat products.

[0020] According to a possible embodiment, the training step is initiated prior to the step of acquiring at least one image of a meat product to be processed.

[0021] According to a possible embodiment, the training step is initiated and completed prior to the step of acquiring at least one image of a meat product to be processed such that access to the database is stopped.

[0022] According to a possible embodiment, the generated prediction model corresponds to a custom model associated to the meat product to be processed.

[0023] According to a possible embodiment, the custom model is added to the database as a new model to be accessed by the prediction system for subsequent meat products.

[0024] According to a possible embodiment, the final product is added to the database as a new model to be accessed by the prediction system for subsequent meat products.

[0025] According to a possible embodiment, each step of the method is accomplished without human intervention.

[0026] According to a possible embodiment, the models contained on the database are acquired using high-end vision systems configured to generate clear, concise and detailed images including CT scans, 3D scans, 2D scans, X-ray scans, ultrasounds or combinations thereof.

[0027] According to a possible embodiment, the method is implemented as part of a meat processing system comprising a conveying system configured to convey the meat product along a processing line; a vision system provided along the processing line configured to acquire the at least one image of the meat product; a meat processing station provided along the processing line downstream of the vision system and comprising the meat processing machine.

[0028] According to another aspect, a method for automatically processing meat products is provided. The method includes in a training step, training an algorithm of a prediction system by providing access to a database containing images and generated models of meat products acquired by a first vision system; acquiring at least one image of a meatproduct to be processed using a second vision system; transmitting the at least one image to the prediction system configured to follow the algorithm to generate a prediction model of the meat product to be processed; transmitting the prediction model to an analysis system configured to generate an integrated model comprising combined features of the at least one image and the prediction model, the analysis system being further configured to define processing instructions for the processing of the meat product taking into consideration the combined features of the integrated model; and transmitting the processing instructions to a meat processing machine configured to automatically process the meat product for obtaining a final product.

[0029] According to a possible embodiment, the method further includes establishing a meat processing system comprising a conveying system; the second vision system; the prediction system; the analysis system; and the meat processing machine.

[0030] According to a possible embodiment, the prediction model generated comprises at least one of : predicted underlying tissue structures of the meat product to be processed; and predicted optimized or near-optimized processed specifications of a predicted final product obtainable from the meat product to be processed.

[0031] According to a possible embodiment, the second vision system, prediction system and analysis system are configured to cooperate to enable omitting the first vision system from the meat processing system.

[0032] According to a possible embodiment, the second vision system, prediction system and analysis system are configured to replace the first vision system from the meat processing system, wherein the prediction models and / or the final products are added to the database as new models to be accessed for subsequent training steps.BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 is a representation of a meat processing system integrating a prediction system, according to an embodiment.

[0034] Figure 2 is a representation of a combination of external and internal features to create an integrated model, according to an embodiment.

[0035] Figure 3 is top view of a workpiece, showing a standard cutting path and a refined cutting path defined by the prediction system, according to an embodiment.

[0036] Figure 4 is an image of a workpiece, showing internal features with a refined cutting path based on the internal features, according to an embodiment.

[0037] Figure 5 is an image of a workpiece, showing annotations regarding internal and / or external features of the workpiece, according to an embodiment.

[0038] Figure 6A is another image of a workpiece, showing internal features thereof, according to an embodiment.

[0039] Figure 6B is the image shown in Figure 6A with a contour line defined around a bone of the workpiece, according to an embodiment.

[0040] Figures 7A to 7C illustrate a beef plate bone intercostal meat harvesting process, with an image of the workpiece showing external features (Figure 7A), an image of the workpiece showing internal features (Figure 7B) and features of the workpiece remaining following the harvesting process (Figure 7C), according to an embodiment.

[0041] Figures 8A to 8C illustrate a beef loin drop process, showing the workpiece (Figure 8A), the cutting line defined between the sirloin and the round (Figure 8B) and external features identified and used to perform the cut (Figure 8C), according to an embodiment.

[0042] Figure 9A is of a pork carcass, showing a cut line for separating the shoulder from the belly, according to an embodiment.

[0043] Figure 9B is of pork ribs following a rib pulling operation, according to an embodiment.

[0044] Figure 10 illustrates a pork shoulder separation process, showing a cut line for separating a shoulder butt from a picnic cut, according to an embodiment.

[0045] Figure 11 illustrates pork ribs lifted from a flank, showing some bone damage created during processing, according to an embodiment.DETAILED DESCRIPTION

[0046] As will be explained below in relation to various embodiments, the present disclosure describes apparatuses, systems and methods for various operations, such as operations for improving operations and efficiency of devices and machines of a processing plant. The devices and systems can be part of a meat processing plant, such as for the processing of animal carcasses for the production of specific meat pieces and products.

[0047] In the context of the present disclosure, the processing plant corresponds to a meat processing plant, where animal carcasses and meat pieces are processed and packaged. However, it should be noted that the systems and methods described herein can be implemented and used in relation with processing plants for various other items and in different fields.

[0048] In some embodiments, the present disclosure describes a scanning process for predicting the presence and location of underlying tissue structures in a meat product to be processed. More particularly, the scanning process is part of a prediction system operable to increase the performance of the related meat processing. For instance, the prediction system is used to gather information in order to take into account the presence of underlying tissue structures, such as bones, fat layers, cartilage or muscle groups, among others. The gathered information can be analysed and sent to a meat processing machine in order to complete its operation with improved accuracy. In other implementations, the prediction system can be required to enable operation of some meat processing machines, without which operation of the meat processing machines would not be viable.

[0049] The prediction system allows for predictions of the presence and location of underlying tissue structures, without resorting to complex scanning systems, such as X-Ray scans (e.g., computed tomography, fluoroscopy, radiography, etc.), magnetic resonance and ultrasound scan. The prediction system includes a combination of a measuring system, such as scanners (e.g., 2D and / or 3D scanners), a machine-learning algorithm and an analysis system. The scanner is adapted to create a corresponding 2D-or 3D-representation of any given meat piece, which is then sent to a processor of the analysis system configured to process the created representation and predict (e.g., calculate) the location of underlying tissue structures. The analysis system is then operable to superimpose the prediction over the created representation and creates processing instructions usable by meat processing machines to perform their operationand / or complete their operation with improved accuracy. Adjusting the operational process of meat processing machines to account for underlying tissue structures improves the quality of the final product, increases cost effectiveness and reduces waste.

[0050] With reference to Figure 1 , a schematic representation of a processing plant 5 is shown. The processing plant includes a processing plant floor provided with various machines mounted along a processing line 10 and operable to act upon a workpiece being transported along the processing line 10 in order to manufacture a final product 12. In the present disclosure, the processing plant 5 corresponds to a meat processing plant 15 having machines 16 and tools for the production of meat products. However, it should be noted that the devices, systems and methods described herein can be implemented in different plants for the production of different products. It should be noted that, as used herein, the expression “final product” can refer to the product manufactured at the end of the processing line, such as the product meant to be packaged and sold. Alternatively, in relation to a given meat processing machine, the “final product” can refer to the product produced by that specific machine. For instance, for a defatting machine, the final product corresponds to the piece of meat from which it removed the layer of fat. That “final product” can then move down the processing line, e.g., becoming the piece of meat to be processed by another machine.

[0051] In this embodiment, the machines 16 mounted along the processing line 10 include meat processing machines 18 operable to act upon the material and / or workpiece 20 being transported. The workpiece can include any one of an animal carcass (e.g., complete or partial) or a piece of meat (e.g., a piece or portion of an animal carcass). The workpiece can be fresh, unprocessed, cut or processed (e.g., partially) prior to being acted upon by the meat processing machines 18. In the present disclosure, the workpieces 20 typically correspond to meat pieces 21, although it is appreciated that other types of workpieces can be used.

[0052] The meat processing machines 18 can include mechanical processing machines 22 operable to act on meat pieces 21 to alter their state. For example, mechanical processing machines 22 can include water jets, circular saws, reciprocating saws, band saws, cutting blades, cutting wires, jaw cutters, ultrasonic knifes, ultrasonic saws, lasers, etc. In some embodiments, the mechanical processing machines are operable to act upon meat pieces 21 that are transported along the processing line 10. For instance, the mechanical processing machines 22 can be adapted to transform an animal carcass intoone or more meat pieces 21 based on implemented and / or predetermined operational parameters. In some embodiments, the animal carcass can correspond to a pig carcass, and the mechanical processing machines 22 can act upon the carcass to produce products such as pork belly, tenderloin, sirloin, ham, bone-in loin cuts, ribs, etc. The mechanical processing machines 22 can be partially- or fully-automated meat processing machines. In other words, the mechanical processing machines 22 can be operable without human intervention.

[0053] Still referring to Figure 1, the machines 16 mounted along the processing line 10 can further include data acquisition instruments 24 configured to collect data regarding the meat pieces 21 being conveyed along the processing line. Different types of data can be collected by the data acquisition instruments 24. Examples of data acquisition instruments 24 include cameras, scanners, scales, ultrasonic measuring devices, among others. The data collected by the data acquisition instruments 24 can be processed and analysed to improve the efficiency of the meat processing plant, and more specifically, improve the efficiency of individual meat processing machines. The performance and efficiency of the meat processing plant can rely heavily on the accuracy of the operations of the meat processing machines. For example, the grading, cutting and deboning operations can each be improved by providing more accurate models of the meat pieces 21 being conveyed along the processing line. It should also be noted that several automated applications / operations could be simply not viable without a mean to locate the underlying structures of the meat pieces.

[0054] In this embodiment, the data acquisition instruments 24 includes a scanner 26 operable to collect data regarding the shape and appearance of the meat pieces 21. The scanner 26 can include a 2D-scanner and / or a 3D-scanner 27 operable to create a corresponding model (e.g., 2D and / or 3D model) of each meat piece being transported along the processing line 10. In the present embodiment, the scanner corresponds to a 3D scanner 27 operable to create a 3D model. The model can include information regarding the shape, contour and / or size of meat piece being scanned. The model can also assist in determining which type of meat piece is being processed, such as a full carcass, a partial carcass, a processed piece of meat, etc. The model can also provide information regarding external features of the scanned meat piece, including defects, for example. It is noted that creating the model can improve the automatization of the meat processing plant, where the mechanical processing machines 22 can process (e.g., cut)the meat such that each piece of meat created has generally identical features (e.g., size, weight, etc.).

[0055] With reference to Figure 2, in addition to Figure 1, in this embodiment, the processing plant 15 includes a prediction system 30 and an analysis system 40. The prediction system 30 is configured to predict the presence and location of various underlying tissue structures within the meat pieces. The underlying tissue structures can include bones, fat layers, cartilage, muscle groups, etc. It is noted that the underlying tissue structures are generally removed, at least partially, from cuts of meats which are packaged, commercialized and sold. It should therefore be appreciated that determining, at least moderately, the presence and / or location of the underlying tissue structures can enable, facilitate and / or refine the operations of the mechanical processing machines 22, which can improve the quality of the created final product. However, it should also be noted that, in some instances, the underlying tissue structures can be kept and are thus part of the packaged, commercialized and sold product. For example, upon cutting and separating the primal shoulder of a hog, the cut is typically performed at a specific location and / or based on customer specifications. The cut may go between ribs or straight through a rib.

[0056] In this embodiment, the prediction system 30 includes a machine-learning algorithm configured to learn from data, identify patterns, make predictions, and / or perform tasks without explicit programming. The machine-learning algorithm can correspond to or be assisted by artificial intelligence (Al), defining an Al-assisted prediction system 33, which can be given access to a databases for various information / data. The Al-assisted prediction system 33 can use the data from these databases to train for predicting the presence of underlying tissue structures of interest. The data and information used for the Al training can include a large number of scans, such as CT scans, 3D scans, 2D scans, X-ray scans, ultrasounds, etc., of meat pieces. For example, in a pork processing plant, images (e.g., 2D images) and scans (e.g., 3D images) of pork body parts can be fed to the Al-assisted prediction system 33 in order to train the Al in predicting the presence of underlying tissue within these pork body parts. The pork body parts can include a shoulder, a leg, a head or any other suitable part of the animal. It should also be noted that the machine-learning algorithm can alternatively be implemented using standard programming (e.g., using user-written code), and can therefore not be assisted by artificial intelligence.

[0057] The data and information used can have various sources. For instance, one database can include a large number of models (2D and / or 3D) of meat pieces. These models can be created using high-end equipment configured to provide clear, concise and detailed information regarding meat pieces. Examples of high-end equipment include X-ray machines and associated systems, such as 2D X-Ray, dual energy X-Ray, computed tomography scanner, ultrasound imagery scanner, magnetic resonance imagery machine, etc. It should be noted that at least some of the above-mentioned high-end equipment are typically not used in production environments. These types of equipment have several limitations, such as processing speed, physical size, etc. , that limit their use in a production context. In some embodiments, the prediction system can be trained using multiple images of the same meat piece or carcass obtained from different imaging technologies. For instance, a first imaging system, such as high-end equipment not intended for production environments, can acquire non-visible or otherwise hard-to-obtain internal information about the meat piece, while a second imaging system (e.g., usable in the production environment) acquires visible or externally accessible information. By correlating the non-visible information from the first imaging system (e.g., the high-end equipment) with the visible information from the second imaging system, the prediction model can learn to infer or “predict” internal or otherwise non-visible characteristics of meat pieces based on the data available in the production environment.

[0058] The prediction system 30 can be trained by “studying” and learning from these models (e.g., the models created using the high-end equipment(s)) in order to recognize patterns and / or features of the meat pieces. In some embodiments, the information gathered from the high-end models can include external and / or surface features (e.g., defects, etc.), size, weight, nature of meat piece, underlying structures (e.g., bones, fat layers, cartilage, muscle groups, glands, membrane, joints, fat thickness, etc.). It should be noted that, as used herein, a “pattern” can be observed physically (e.g., directly and / or using a measuring or vision system) or observed mathematically by applying algorithms. Similarly, “pattern recognition” can refer to the process of recognizing patterns by using a machine learning algorithm for the classification of data based on knowledge already gained or on statistical information extracted from patterns. In this embodiment, the “knowledge already gained” can correspond to the data included in the databases accessible by the prediction and / or analysis systems and can be referred to as “training data”. The prediction and / or analysis systems can learn from the training data in order to provide results, as will be described below, in an accurate manner.

[0059] In some embodiments, the training data can include a training set of data (e.g., as discussed above) and a testing set of data. The training set typically consists of the set of images / models that are used to train the prediction system. Training rules and algorithms are used to give relevant information on how to associate input data with output decisions (e.g., associate existing models for the creation of the prediction model). The prediction system is trained by applying these algorithms to the dataset (e.g., the training set and / or the input data) such that the relevant information is extracted from the data, and results (e.g., prediction models) are obtained. The input data can include annotations to guide, assist or otherwise facilitate training of the prediction system.

[0060] In some embodiments, annotations can be completed and integrated in the images, models and / or other data making up the input data manually or algorithmically (e.g., autonomously, automatically, etc.). Particularly, images and / or models created using high-end equipment (e.g., the first imaging system) can be annotated to identify and differentiate the various parts shown in the images and / or models. For instance, images and / or models can be annotated to identify bones, muscles, fat, cartilage, etc. While these different parts can be visually differentiable by the prediction system (e.g., during training), providing annotations which clearly identifies at least some of the parts can assist and facilitate the training process. It should also be noted that annotations can provide general information by identifying each part with general terms, such as “bone”, “muscle”, “cartilage”, etc., as seen in Figure 5, for example. Alternatively, or additionally, the annotations can provide specific information by identifying each part with specific terms, such as “scapula”, “shoulder”, “subcutaneous fat (backfat)”, etc.

[0061] As seen in Figure 6A, an image of the workpiece 20 created using high-end equipment (e.g., the first imaging system) can be obtained. This image can be annotated manually, such as by an operator, or automatically by an algorithm. In this embodiment, the contours (C) of the scapula bone 25 are clearly annotated to create an annotated image, seen in Figure 6B. The annotated image can then be fed to the prediction system to enable “learning” the location of the scapula relative to other parts shown in the original (e.g., unannotated) image. It is appreciated that unannotated images can be used as training data, alone or together with annotated images to provide a wide range of images and information to the prediction system. As previously noted, the prediction system can also be configured to predict internal or otherwise non-visible characteristics of meat pieces (e.g. the location of the scapula) based on the data available in the productionenvironment. As such, it is noted that the prediction system can learn from annotated images showing internal structures (e.g., X-rays, etc.) to identify the location of these internal structures (e.g., the scapula) by analyzing images showing external structures (e.g., images obtained with systems usable in the production environment).

[0062] On the other hand, the testing set is used to test the system. It corresponds to the set of data used to verify whether the system is producing the correct output after being trained or not. Testing data is used to measure the accuracy of the system. It should be noted that the output data obtained from the testing dataset can subsequently be used as input data as part of the training dataset. As such, the prediction system can be adapted to learn from the output of the system it is integrated in. In other words, the system can be adapted to produce output data useable as input data, thereby learning from itself to improve its efficiency. For example, following the creation of the integrated model, the meat processing machines perform one or more actions and / or cuts on the meat piece to produce a final product (e.g., primal cut, rib pulling, neck lifting, etc.). The final product can be scanned and integrated in the database as training data to assess the specifications of the final product (e.g., quality, yield, etc.) and determine if improvements can be made, thereby training the system. In some embodiments, the newly integrated data can be annotated to facilitate the learning sequence of the prediction system to better adapt and improve the prediction system. It should also be noted that the final product can be scanned using the first imaging system (e.g., high-end equipment non-suitable for production environment) and / or the second imaging system (e.g., usable in the production environment).

[0063] In some embodiments, the training data and the testing data can originate from the same physical meat piece (e.g., the same cut, the same animal, etc.) in order to ensure coherence and traceability between the datasets used for model development (e.g., training) and model validation (e.g., testing). For example, and as previously noted, the high-end equipment can acquire a first dataset providing internal or otherwise non-visible anatomical information for a given piece of meat. The same piece of meat can then be scanned using the second measurement system to acquire a corresponding visible or externally accessible dataset, which can also be referred to as a “production environment dataset”. These paired dataset, which represent the same piece of meat (and therefore the same underlying tissue structures), can be associated, correlated, or otherwise linkedwithin the training set to enable the prediction system to learn the relationship between external features and internal structures.

[0064] Similarly, the testing data can be generated by processing the same piece of meat with meat processing machines following an application of the trained model to the production-environment dataset for that piece of meat. The created final product can validate whether the prediction of underlying structures matches the actual internal data previously obtained for that same piece (e.g., the training data). This validation corresponds to feedback provided to the prediction system to improve and / or guide future operations. For instance, if the prediction matches the training data and / or matches the final product, then subsequent operations can use this initial successful operation as a guide or template. Alternatively, should the prediction fail to match the training data and / or the desired final product, then feedback can be provided to the system to improve subsequent operations. Feedback can be provided automatically, for example, based on measured differences between the obtained final product and desired final product, or manually, by a human operator.

[0065] It should therefore be understood that using training and testing datasets that originate from the same individual meat piece ensures that the prediction system is trained and validated against anatomically consistent information. By pairing internal data obtained from high-end equipment with the corresponding production-environment data for the same specimen, the system can reliably learn and confirm the relationship between external features and underlying structures. This approach reduces variability, improves dataset coherence, and strengthens the overall accuracy and robustness of the prediction model.

[0066] In some embodiments, the training data can correspond to final products produced by a human operator / worker. These final products can be integrated in training datasets accessible by the prediction and / or analysis systems to train the system in reproducing and / or improving on the human-made cuts. For example, for processing a pork belly, a worker would have to manually remove fat, glands, cartilage, bone chips and bits of meat (e.g., lean meat) from the surface of the pork belly depending on desired / current specifications. Rather than designing software configured to detect the different parts to be cut according to defined reference points, the meat processing machines can be trained to replicate the final product, rather than the specific steps to reach the final product. In this embodiment, the processing system can be adapted to include a firstmeasuring system adapted to scan the part (e.g., the pork belly) before cutting. The worker can then perform manual operations to remove the fat / lean from the pork belly. Finally, a second measuring system (or a feedback loop to the first measuring system) can be configured to identify the volume removed by the operator and / or the post-cut specifications (e.g., size, shape, weight, etc.) of the pork belly.

[0067] It should therefore be noted that the various movements made by the operator or the initial state of the piece of meat are not of interest, used or analyzed to train the processing system. Instead, the end result is used and analyzed to train the processing system in order to reproduce similar or identical end results with subsequent pork bellies. For example, in this embodiment, the prediction system is adapted to generate a prediction model showing how a cut should be made (taking in consideration the presence of bones, cartilage, etc.) regardless of the surface characteristics of the piece of meat (e.g., presence of fat, defects, etc.). It should be noted that other tasks / operations could be similarly reproduced. For example, some products will be defatted, such as pork shoulder butts or pork loins. Traditionally, workers will defat the product using a cutting device such as a knife, annular blade or skinning tool. However, the workers can overtrim and remove too much fat for the allowed / pre-established quality specifications. Moreover, the workers can score the product and remove valuable lean. The prediction system could be used to learn from the good and the bad examples provided in normal operation settings for training the algorithm to improve performance. Alternatively, the prediction system could, later on, learn from its mistakes by using a QA scanning system and improve the machine learning model.

[0068] It should be understood that the high-end equipment used to create the training data (e.g., the set of images and models useful for training the prediction system) are no longer required on the processing line. In other words, the prediction system and analysis system can be adapted to replace these types of equipment and tools. As such, the high-end equipment, which can come with high costs (e.g., initial purchasing cost), increased maintenance (which also increases overall cost), increased footprint on the factory / production floor, etc., can be omitted and replaced with one or more data acquisition instruments. In some embodiments, the data acquisition instruments come with lower initial and maintenance costs, lower factory floor footprint, etc.

[0069] In some embodiments, the Al-assisted prediction system 33 can use 3D models 31 created by the 3D scanner 27 as input data in order to determine (e.g., predict) thepresence and / or location of underlying tissue structures. For instance, the Al-assisted prediction system 33 can create an approximation of a bone structure for a given 3D model 31 of a given pig carcass, along with a prediction of the presence and location of fat layers, cartilage and / or muscle groups connected together and to the bone structure. The Al-assisted prediction system 33 can therefore create a prediction model 35 in three dimensions showing one or more underlying tissue structures of the piece of meat. The information relating to the presence and / or location of underlying tissue structures can therefore correspond to the output data of the Al-assisted prediction system 33. More particularly, the Al-assisted prediction system 33 uses 3D models 31 created by the 3D scanner 27 to create corresponding prediction models 35. In some embodiments, the prediction models 35 can be created of standard 2D images, in color and / or shades of gray. Alternatively, or additionally, different laser / illumination wavelengths could be used along (e.g., simultaneously with) the 2D and / or 3D scanner to create the prediction models 35. In other embodiments, a laser line scatter data set can be used to create the prediction models 35.

[0070] It should be noted that the prediction model is built by the prediction system 30 based on acquired knowledge and / or previous training using databases. As previously mentioned, the prediction system 30 can use machine-learning algorithms, such as algorithms assisted by artificial intelligence. For instance, the prediction system 30 can integrate Deep Learning algorithms which allow computational models that are composed of multiple processing layers to learn representations of data with multiple levels of abstraction. It is noted that Deep Learning is a subset of machine learning that excels at identifying complex patterns within data. As such, deep learning can be a highly effective tool for pattern recognition tasks, where the goal is to classify or categorize input data based on recognized patterns. Once classified / categorized, the input data can be manipulated to generate output data, such as, in this embodiment, the prediction model.

[0071] During training, the prediction system 30 can be shown one or more images (e.g., between 1 and 10,000 images), such as 3D models, images of a piece of meat, annotated images and / or models, etc., and produces an output, for example, in the form of a vector of scores, one for each visual feature located and / or identified. In some embodiments, a desired outcome is to have the visual features have the highest score such that the created prediction model corresponds as faithfully as possible to the real and / or desired results. Generally, this is unlikely to happen before training the prediction system. During training,the error (or distance) between the output scores and the desired pattern of scores can be computed (e.g., automatically, manually or a combination thereof). The prediction system then modifies its internal adjustable parameters to reduce this error. These adjustable parameters, often called weights, are real numbers that can be seen as ‘knobs’ that define the input-output function of the prediction system. In a typical deep-learning system, there may be hundreds of millions of these adjustable weights, and hundreds of millions of labelled examples with which to train the machine. To properly adjust the weight vector, the learning algorithm computes a gradient vector that, for each weight, indicates by what amount the error would increase or decrease if the weight were increased by a tiny amount. The weight vector is then adjusted in the opposite direction to the gradient vector. The objective function, averaged over all the training examples, can be seen as a kind of hilly landscape in the high-dimensional space of weight values. The negative gradient vector indicates the direction of steepest descent in this landscape, taking it closer to a minimum, where the output error is low on average. Detailed understanding of DeepLearning can be found in the article : LeCun, Y., Bengio, Y. & Hinton, G. Deep learning. Nature 521, 436-444 (2015). https: / / doi.Org / 10.1038 / nature14539, which is hereby incorporated in its entirety by reference.

[0072] Alternatively, or additionally, the prediction system 30 can retain access to the databases (e.g., the training data) for building the prediction models. For instance, the prediction system 30 can compare the 3D model 31 of a given pig carcass on the processing line (e.g., testing data) to similar “training data 3D models”. The training data includes respective underlying tissue structures of the similar training data 3D models such that the prediction system is adapted to create a corresponding prediction model 35 for the pig carcass being conveyed on the processing line.

[0073] In this embodiment, the analysis system 40 is configured to combine the 3D model 31 and the prediction model 35 in order to create an integrated model 42. The integrated model 42 is useful in determining the dimensions of the pig carcass, including the relative dimensions of the different parts of the pig carcass, including the various parts that are meant to be removed and / or separated from the underlying tissue structures of the prediction model 35. The determined dimensions can then be used to produce operational instructions for the mechanical processing machines 22 of the production plant. The produced operational instructions enable the mechanical processing machines 22 to perform their actions, to improve the efficiency of their actions and to create products ofimproved quality. In other words, the analysis system 40 is operable to superimpose the prediction model 35 over the 3D model 31 to produce instructions usable by one or more meat processing machines to complete their operations with higher accuracy, among others. In some embodiments, the integrated model 42 can provide exact, or near-exact, positions of the underlying structures of each individual piece of meat.

[0074] In some embodiments, the analysis system 40 can define refined cutting paths for various cutting operations. For instance, and with reference to Figures 3 and 4, during operations, automated robotic devices typically operate by following a standard cutting path (SC), leaving some clearance between the blade and the underlying structure (e.g., ribs, cartilage, etc.) in order to prevent collisions. The analysis system 40 can provided a refined cutting path (RC) to increase the process yield of the automated robotic devices. It should be noted that, in addition to the adjusting the general direction of the cutting path, the depth of the cutting tool can be similarly adjusted to prevent collisions. Moreover, the analysis system 40 can provided a refined cutting path for each individual piece of meat being conveyed along the processing line. It is thus noted that the cutting path of the automated robotic device can be adjusted (e.g., customized) for each different piece of meat. It is therefore noted that the analysis system 40 can therefore assist in producing final products with improved yield.

[0075] Operations that can be improved using the prediction and analysis systems include pork primal cuts (e.g., shoulder cut, ham cut, etc.), beef loin drop, pork shoulder separation in butt and picnic, rib pulling operations, beef plate bone pulling (which consists of removing the beef plate bone from the plate or navel), beef plate bone intercostal muscle harvesting, among others. The beef plate bone intercostal meat harvesting is typically a labor-intensive process, where operators need to cut meat between bones using a straight knife. Predictive positions of the bones provided by the prediction system would provide important information usable to assist in automating at least some parts of this process. For reference, Figures 7A to 7C illustrate various steps of the beef plate bone intercostal meat harvesting process. Figure 7A can correspond to an image acquired using production line-suitable equipment (e.g., cameras, scans, etc.) whereas Figure 7B can correspond to an image acquired using high-end equipment. Figure 7C illustrates the processed components following the beef plate bone intercostal meat harvest (the meat having been removed / harvested).

[0076] For beef loin drop operations, and as seen in Figures 8A to 8C, it is noted that this beef application includes a cut 50 made between the sirloin and the round. Standard and / or typical operators use a visible portion of the beef 51 , such as the aitch bone 52 and tail bone 54 to perform the cut. However, these features rely on the splitting process and other previous processing steps to enable accurate and repeatable results. It is appreciated that being able to train the prediction system to recognize, identify and / or locate internal bones (e.g., the aitch bone 52 and tail bone 54) can improve the cut quality and processing sequence.

[0077] In some embodiments, and for some markets, the pork ribs (seen in Figures 9B and 11) are partially removed from the rib cage, therefore removing precious surface features and marks. Moreover, the final cut specification can include a distance from the scapula (i.e. , shoulder blade or blade bone), which corresponds to a hidden feature (e.g., an internal or otherwise non-visible anatomical information). It is appreciated that training the prediction system to determine a location of the scapula in reference to other external feature(s) would be the target for this particular cut (e.g., shoulder cut 55). As seen in Figure 9A, examples of external features could be a position of a leg 56, locations of vertebrae 58, back bone curvature 60, etc. In addition, for pork rib pulling operations, it is often the case that some fat is present and masking the surface / external features and / or creating a false fat-lean frontier. These can affect thickness estimation and subsequent cut locations. This situation occurs for various cuts, such as for the side cartilage of the side ribs, as well as the last bone of the side ribs, for example. It is appreciated that being able to predict locations of specific features would help the system be more robust and adapted against visual variations, while also being able to cut closer to the cartilage and / or bones, which ends up creating higher quality / yield products.

[0078] With reference to Figure 10, another example is the pork shoulder separation in butt and picnic. This cut 62 is done to separate the shoulder generally in half. The cut can be done by assessing and examining external features, such as a vein and / or a visible cut of the blade bone. The neck bones are completely removed at this point. However, these external features (bone location, status, quality, accessibility, etc.) depend on previous processing steps, for example, the shoulder primal cut is typically completed prior to performing the pork shoulder separation. The prediction system can thus be trained to efficiently, repeatedly and accurately locate the shoulder blade to provide robust trainingdata and improved cut accuracy. The identification of the location of the shoulder blade is illustrated in Figures 6A and 6B.

[0079] It should be noted that the attainable yield with automated removal of intercostal muscles is at least partially conditioned by the capacity to contour around the ribs with the blade. Similar to the rib pulling operations, without previously detecting the underlying structure (e.g., the rib contours), automating the beef plate bone intercostal muscle harvesting operation can result in leaving excess meat on the ribs and / or risking bone fragments in the meat due to collisions between the blade and the bones. Using the prediction and analysis systems provides a predicted position of the ribs to enable the blade to follow a refined cutting path.

[0080] As seen in Figure 5, the removal of pork snouts can also be assisted by implementing the prediction and analysis systems. Pork snout removal involves performing a cut through a precise path to avoid damage and devaluation of the final meat product. Incidentally, deviating from the cutting path may result in decreased product quality. In addition, deviating from the cutting path can cause increased blade wear as bone structures are contacted, leading to increased processing time due to having to deal with bone-and-blade collisions. Figure 11 shows an example of bone damages due to a side ribs tail length being too short and / or due to a poor bone thickness estimate. Specifically, Figure 11 shows that two ribs 70 and some cartilage have been cut in half lengthwise, as one half is visible on the flank and the other on the rib. In the picture, the rib is darker (e.g., dark red) relative to the surrounding structures and tissues. This can occur when the knife is not low enough to go under the ribs, which can be a result of poor rib detection. The knife therefore did not have enough clearance to cut through the meat and gradually reach the desired thickness.

[0081] In this embodiment, the training dataset used by the prediction system can include a database of pig head models, including respective underlying structures. The prediction system can then generate / create a corresponding prediction model of the underlying structure of the pig head being processed, which can then be matched to the model of said pig head in order to define refined cutting paths. It is also possible to create additional training data by : 1) scanning the pig head before performing the boning operation (e.g., using a vision system); 2) debone the pig head; and 3) scan the pig head once again to create an output data which can be implemented in the training set database. It is thus noted that the system can learn from its own results to improve its operations (e.g, speedof operations, efficiency, quality of cuts, etc.). It should thus be understood that, in some embodiments, the prediction model is a custom, new and / or unique model, and is not simply chosen from a selection of existing models contained on the database.

[0082] It will be appreciated from the foregoing disclosure that there is provided a solution configured to improve automated operations within a processing plant. The solution includes a 2D and / or 3D scanner, an Al-powered (or Al-assisted) prediction system and an analysis system. The scanner creates a representation of the meat piece that is then sent to the prediction system which processes the representations and predicts the location of underlying tissue structures based on machine learning (e.g., Deep Learning). The analysis system then superimposes the prediction over the representation and produces operational instructions usable by meat processing machines to complete processing operations with increased accuracy.

[0083] The databases used for training the prediction system include data obtained using a first measurement device configured to map out and identify external features and internal features of meat pieces. A second measurement device is then used to create a standard representation of meat pieces to be processed. It should be noted that the second measurement device can correspond or be similar to production measurement devices (i.e., measurement devices used on an operational production / processing line). The prediction system builds the prediction model which is then combined with the representation from second measurement device. Artificial intelligence and / or standard programming are integrated in the processing system to correlate the external reference points of the standard representation to internal structure and / or tissue of the prediction model.

[0084] It is therefore noted that the first measurement device would not be used in production equipment (e.g., on the operational production / processing line). For example, a typical cutting system in a beef slaughter line includes a relatively large X-ray scanning system configured to scan the full beef carcass. The X-ray scanning system, and related components and equipment, is located in a dedicated room, such as within the slaughterhouse, in a laboratory, in a mobile unit, etc., which requires major modifications to the production line, which is very expensive. The data acquired by the X-ray scanning system follows the carcass along the slaughter line. According to another example, known equipment typically employs an X-ray scanning system on a sub primal part in order to get bone locations for controlling waterjets and cutting the sub primal part. These operationscan become relatively costly for the same reasons as stated before (e.g., footprint, cost, maintenance, etc.). In addition, it is noted that movement of the carcass can occur as it travels along the slaughter line, between different operations and prior to any subsequent vision systems. As such, it should be understood that, in some instances, the data acquired by the X-ray scanning system can be inaccurate or simply unusable.

[0085] Instead, the solution proposed in the present disclosure includes the prediction and analysis systems, combined with the second measurement device, which are configured to reach similar, if not identical, results as those obtainable with the first measurement device (e.g., the X-ray system). The associated costs (e.g., acquisition, installation, operation, maintenance, etc.) and operational footprint can thus be reduced by foregoing the first measurement device(s) for the second measurement device(s).

[0086] According to one exemplary implementation of the present solution, a vision system (e.g., the 3D scanner) scans the meat piece to obtain a corresponding model of the meat piece. The model can correspond to a series of points dispersed on a surface to create a 3D cloud map type image, with each point on the 3D surface model being represented by a pixel, for example. Using the intensity of the pixels and an image processing algorithm (e.g., the analysis system), a cut path can be determined using measurement specifications such as meat thickness, tail length, etc. In this embodiment, the 3D model uses external reference points (i.e., points on the surface of the meat piece) jointly with statistics and / or stored data (e.g., training data) for determining bone depth, dimensions, thickness, etc., of the piece of meat.

[0087] It should be appreciated that being able to predict the internal / underlying features, such as bone structures, allows for improving consistency, accuracy and performance of the processing systems. In addition, underlying structure prediction can also enable the processing systems to be less dependent on surface appearance of the meat pieces, for example, the presence (or absence) of surface fat, debris on the surface, defects on the surface (e.g., portions of the piece missing, etc.). Improving the cuts performed can also prevent cutting bones, losing meat yield and / or lowering product quality.

[0088] It is noted that using an X-Ray scanning device or an ultrasound machine can enable for precise modelling of the internal / underlying structures of meat pieces. However, the system would require additional investment and complexity. Moreover, it is further noted that X-ray and / or ultrasound scans have some limitations, such as poor depth ofscans, relatively complex scans and corresponding interpretations, requiring contact scanning techniques (particularly for ultrasounds), requiring specialized maintenance, costs of initial setup, requiring large footprint within the plant, concerns regarding safety (x-ray), costs of operations, low frequency of scans which can fall behind in a high-production speed meat packing plant, among others. It should be noted that, having a slow operational speed (i.e., low frequency of scans) can prevent some systems from reaching the desired overall factory speeds / output rates by creating a bottleneck effect at one or more locations along the processing line.

[0089] The present disclosure may be embodied in other specific forms without departing from the subject matter of the claims. The described example embodiments are to be considered in all respects as being only illustrative and not restrictive. For example, the 3D scanner may be replaced or assisted by another scanning device or technology to generate an image set adapted to train the Al program. The 3D scanner and / or other scanning device, Al powered prediction system and analysis system may or may not be separate systems. In some embodiments, any of the scanning device(s), Al powered prediction system and analysis system can be integrated within a meat processing machine, while, in other embodiments, any of the scanning device(s), Al powered prediction system and analysis system can be part of the processing plant as a standalone machine.

[0090] The CT scan may be replaced by another tissue penetrating scan technology, such as ultrasonic or radiographic scan. 2D X-Ray scans can be used as tissue penetrating scan technology. In such cases, the generated 2D image is matched (e.g., in orientation and in position) with the 3D scan, for the benefit of the Al training, among others. In some embodiment, the 3D scan may be replaced with a 2D-scanning technology, such as a conventional camera, for example. Once again, the orientation and position of both the 2D and 3D scans is matched, for the benefit of the Al training.

[0091] In other embodiments, the prediction system can be made to learn from human interventions and operations. For instance, a piece of meat or carcass can be scanned using the first measurement system (e.g., high-end vision system(s)). Then, a human operator can process the piece and perform various operations (e.g., cuts, removals, cleaning, separating, etc.). A second scan is then performed, using a second measurement system (e.g., suitable for production lines) or using the first measurement system once again. Differences from the first and second scans can be identified andprovided as data to the prediction system. In other words, the prediction system can be trained by “studying” the differences between initial and processed pieces. The processed pieces can be manually processed (e.g., by an operator) or machine-processed using specialized tools. The differences can teach the system which part of the piece of meat to remove to replicate previously performed operations.

[0092] The described example implementations are to be considered in all respects as being only illustrative and not restrictive. In the present disclosure, an embodiment is an example or implementation of the described devices, systems and methods. The various appearances of “one embodiment,” “an embodiment” or “some embodiments” do not necessarily all refer to the same embodiments. Although various features may be described in the context of a single embodiment, the features may also be provided separately or in any suitable combination. Conversely, although the described devices, systems and methods may be described herein in the context of separate embodiments for clarity, it may also be embodied in a single embodiment. Reference in the specification to “some embodiments”, “an embodiment”, “one embodiment”, or “other embodiments”, means that a particular feature, structure, or characteristic described in connection with the embodiments is included in at least some embodiments, but not necessarily in all embodiments.

[0093] As used herein, the terms “coupled”, “coupling”, “attached”, ’’connected” or variants thereof as used herein can have several different meanings depending in the context in which these terms are used. For example, the terms coupled, coupling, connected or attached can have a mechanical connotation. For example, as used herein, the terms coupled, coupling or attached can indicate that two elements or devices are directly connected to one another or connected to one another through one or more intermediate elements or devices via a mechanical element depending on the particular context.

[0094] Similarly, positional descriptions such as “top”, “bottom”, “above”, “under”, “below”, “left”, “right”, “front”, “rear”, “parallel”, “perpendicular”, “transverse”, “inner”, “outer”, “internal”, “external”, and the like should, unless otherwise indicated, be taken in the context of the figures and should not be considered limiting.

[0095] In the above description, the same numerical references refer to similar elements. Furthermore, for the sake of simplicity and clarity, namely so as to not unduly burden the figures with several references numbers, not all figures contain references to all thecomponents and features, and references to some components and features may be found in only one figure, and components and features of the present disclosure which are illustrated in other figures can be easily inferred therefrom. The implementations, geometrical configurations, materials mentioned and / or dimensions shown in the figures are optional, and are given for exemplification purposes only.

[0096] In addition, although the optional configurations as illustrated in the accompanying drawings comprises various components and although the optional configurations of the described devices and systems as shown may consist of certain geometrical configurations as explained and illustrated herein, not all of these components and geometries are essential and thus should not be taken in their restrictive sense, i.e. should not be taken as to limit the scope of the present disclosure. It is to be understood that other suitable components and cooperations thereinbetween, as well as other suitable geometrical configurations may be used for the implementation and use of the described devices and systems, and corresponding parts, as briefly explained and as can be easily inferred herefrom, without departing from the scope of the disclosure.

Claims

1. CLAIMS1. A method for automatically processing meat products, comprising:acquiring at least one dataset of a meat product to be processed;transmitting the at least one dataset to a prediction system configured to follow an algorithm to:generate a prediction model of the meat product to be processed comprising at least one of:o predicted underlying tissue structures; ando predicted optimized or near-optimized processed specifications;transmitting the prediction model to an analysis system configured to generate an integrated model comprising combined features of the at least one dataset and the prediction model, the analysis system being further configured to define processing instructions for the processing of the meat product taking into consideration the combined features of the integrated model; andtransmitting the processing instructions to a meat processing machine configured to automatically process the meat product for obtaining a final product.

2. The method of claim 1 , wherein the at least one dataset includes at least one of a 2-dimensional image, a 3-dimensional image, laser / illumination wavelengths and a laser line scatter data set.

3. The method of claim 1 or 2, wherein the at least one dataset comprises external characteristics of the meat product to be processed.

4. The method of claim 3, wherein the external characteristics comprise product specifications including any one or combination of a shape of the meat product, a size of the meat product, a weight of the meat product, dimensions of various parts of the meat product and relative dimensions between parts of the meat product.

5. The method of claim 4, wherein the external characteristics comprise bones, fat layers, cartilage, muscles and / or muscle groups of the meat products and / or processed meat products.

6. The method of any one of claims 3 to 5, wherein the external characteristics comprise product features including any one or combination of a nature or type of the meat product, a quality of the meat product, a presence of defects, defect specifications including size and color, and relative defect location on the meat product.

7. The method of any one of claims 1 to 6, wherein the combined features of the integrated model includes relative positions between the external characteristics of the meat product and the predicted underlying tissue structures to predict a location of the underlying tissue structures within the meat product.

8. The method of claim 7, wherein the processing instructions include generating one or more cutting paths along the meat product taking into consideration the location of the underlying tissue structures in order to avoid the underlying tissue structures during operation of the meat processing machine.

9. The method of any one of claims 1 to 8, wherein the combined features of the integrated model includes relative positions between the external characteristics of the meat product and the predicted optimized or near-optimized processed specifications to assist in generating the processing instructions adapted to enable operation of the meat processing machine to process the meat product for obtaining the final product having specifications similar to the predicted optimized or near-optimized processed specifications.

10. The method of any one of claims 1 to 9, wherein the algorithm includes pattern recognition to identify one or more external characteristics of the meat product to be processed.

11. The method of any one of claims 1 to 10, wherein the algorithm integrates artificial intelligence to identify one or more external characteristics of the meat product to be processed.

12. The method of claim 11, wherein the artificial intelligence includes machine learning.

13. The method of claim 12, wherein the machine learning includes deep learning, where multiple layers of processing are used to extract progressively higher-level features from data.

14. The method of any one of claims 1 to 13, further comprising, in a training step, providing the prediction system access to a database containing at least one of:• models of meat products with underlying tissue structures; and• models of processed meat products with optimized or near-optimized processed specifications,in order to train the algorithm of the prediction system in pattern recognition.

15. The method of claim 14, wherein the training step further includes identifying differences between the models of meat products with underlying tissue structures and the models of processed meat products with optimized or near-optimized processed specifications, and providing the prediction system with data corresponding to the differences.

16. The method of claim 14 or 15, wherein the models of processed meat products with optimized or near-optimized processed specifications are created without human intervention.

17. The method of claim 14 or 15, wherein the models of processed meat products with optimized or near-optimized processed specifications are created using human intervention.

18. The method of any one of claims 14 to 17, wherein training the algorithm of the prediction system in pattern recognition includes gathering information in order to take into account the presence of bones, fat layers, cartilage and / or muscle groups of the meat products and / or processed meat products.

19. The method of any one of claims 14 to 18, wherein the training step is initiated prior to the step of acquiring at least one image of a meat product to be processed.

20. The method of any one of claims 14 to 19, wherein the training step is initiated and completed prior to the step of acquiring at least one image of a meat product to be processed such that access to the database is stopped.

21. The method of any one of claims 1 to 20, wherein the generated prediction model corresponds to a custom model associated to the meat product to be processed, wherein the custom model is added to the database as a new model to be accessed by the prediction system for subsequent meat products.

22. The method of any one of claims 1 to 21 , wherein the final product is added to the database as a new model to be accessed by the prediction system for subsequent meat products.

23. The method of any one of claims 1 to 22, wherein each step of the method is accomplished without human intervention.

24. The method of any one of claims 1 to 23, wherein at least some steps of the method are accomplished without human intervention.

25. The method of any one of claims 1 to 24, wherein the models contained on the database are acquired using a high-end measurement system configured to generate clear, concise and detailed datasets.

26. The method of claim 25, wherein the high-end measurement system include CT scans, 3D scans, 2D scans, X-ray scans, ultrasounds or a combination thereof.

27. The method of any one of claims 1 to 26, wherein the method is implemented as part of a meat processing system comprising:a conveying system configured to convey the meat product along a processing line;one or more measurement systems provided along the processing line and configured to acquire the at least one dataset of the meat product;a meat processing station provided along the processing line downstream of the vision system and comprising the meat processing machine.

28. The method of claim 27, wherein the underlying tissue structures correspond to components of the meat product at least partially obstructed from the one or more measurement systems provided along the processing line.

29. The method of claim 28, wherein the external characteristics correspond to components of the meat product at least partially accessible by the one or more measurement systems provided along the processing line.

30. A method for automatically processing meat products, comprising:in a training step, training an algorithm of a prediction system by providing access to a database containing datasets and generated models of meat products acquired by a first measurement system;acquiring at least one dataset of a meat product to be processed using a second measurement system;transmitting the at least one dataset to the prediction system configured to follow the algorithm to:generate a prediction model of the meat product to be processed;transmitting the prediction model to an analysis system configured to generate an integrated model comprising combined features of the at least one dataset and the prediction model, the analysis system being further configured to define processing instructions for the processing of the meat product taking into consideration the combined features of the integrated model;transmitting the processing instructions to a meat processing machine configured to automatically process the meat product for obtaining a final product.

31. The method of claim 30, further comprising establishing a meat processing system comprising:a conveying system;the second measurement system;the prediction system;the analysis system; andthe meat processing machine.

32. The method of claim 30 or 31, wherein the prediction model generated comprises at least one of:predicted underlying tissue structures of the meat product to be processed; andpredicted optimized or near-optimized processed specifications of a predicted final product obtainable from the meat product to be processed.

33. The method of any one of claims 30 to 32, wherein the second measurement system, prediction system and analysis system are configured to cooperate to enable omitting the first measurement system from the meat processing system.

34. The method of any one of claims 30 to 32, wherein the second measurement system, prediction system and analysis system are configured to replace the first measurement system from the meat processing system, wherein the prediction models and / or the final products are added to the database as new models to be accessed for subsequent training steps.

35. The method of any one of claims 30 to 34, wherein, in the training step, at least one of the models of meat products acquired by the first measurement system corresponds to the meat product to be processed in a subsequent step.

Images