Intelligent quantification cutting method and system of boneless raw meat with ultrasonic assistance

Abstract

The application provides an ultrasonic-assisted intelligent quantitative cutting method and system for boneless raw meat, and relates to the technical field of meat food processing machinery. The visual parameters and ingredient parameters of the boneless raw meat are obtained, the boneless raw meat is first shaped, and the first shaping parameters are obtained. The three-dimensional profile information of the boneless raw meat after the first shaping is obtained, and the cutting path is planned based on the three-dimensional profile information. The pre-stored meat quality structure data and the first shaping parameters are input into the rebound prediction model to predict the profile rebound amount. The boneless raw meat is second shaped based on the second shaping control parameters, the visual parameters and ingredient parameters are input into the intelligent decision model, and the ultrasonic cutting parameters are output. The ultrasonic cutting parameters are used to drive the ultrasonic cutter, the ultrasonic cutter is controlled to intermittently feed along the cutting path, quantitative cutting is performed, high-precision and low-damage quantitative cutting of the boneless raw meat is realized, and the production efficiency and the standardization level are greatly improved.

Description

Technical Field

[0001] This invention relates to the field of meat processing machinery technology, and in particular to an ultrasonic-assisted intelligent quantitative cutting method and system for deboned raw meat. Background Technology

[0002] In recent years, the development of new food industries such as industrialized cuisine has placed higher demands on the standardized pre-processing of meat, emphasizing not only the uniformity of product specifications but also the maintenance of high product integrity. Meat cutting, as a crucial link in the meat processing industry chain, directly impacts product quality, resource utilization, and industrial economic benefits through its processing precision and efficiency. Currently, most mainstream quantitative meat cutting equipment both domestically and internationally employs traditional mechanical cutting methods, such as separating meat chunks through high-speed mechanical motion using disc blades or reciprocating blades. However, such equipment commonly suffers from a "blade sticking" phenomenon in practical applications. During the cutting process, meat tissue easily adheres to the blade surface due to friction and stress, leading to increased cutting resistance, uneven cuts, and even damage to the meat's structural structure. This problem not only affects cutting precision and product appearance regularity but may also lead to decreased yield, increased waste, and impacts production efficiency and hygiene safety in continuous operations. Among related technologies, the introduction of ultrasonic scalpels as a novel cutting tool provides an effective solution to the pain points of traditional mechanical blade cutting. Ultrasonic scalpels achieve precise separation of meat tissue through high-frequency vibration, effectively reducing the adhesion between the scalpel and the meat tissue, significantly decreasing adhesion, ensuring a smooth cut surface, and reducing product waste. However, the initial contour of boneless raw meat is irregular, and it is prone to springback after shaping, resulting in a mismatch between the scanned contour and the actual contour, making it impossible to plan the path for quantitative slicing. The force generated by the ultrasonic scalpel during slicing causes plastic deformation of the boneless raw meat, making it difficult to slicing precisely, ultimately leading to uneven product quality and inconsistent specifications, which cannot meet the needs of industrialized cuisine. Furthermore, different types of meat vary greatly, and ultrasonic processing parameters cannot be adapted to different types, affecting the quality of meat slicing. Summary of the Invention

[0003] This invention provides an ultrasound-assisted intelligent quantitative cutting method and system for boneless raw meat, which solves the defects of existing cutting methods, such as the inability to perform quantitative cutting path planning, uneven product quality, and inconsistent specifications, which affect the quality of meat cutting.

[0004] This invention provides an ultrasound-assisted intelligent quantitative cutting method for boneless raw meat, comprising: Obtain visual and compositional parameters of boneless raw meat; Based on the visual and component parameters, the boneless raw meat is first shaped to obtain the first shaping parameters; Obtain the three-dimensional contour information of the boneless raw meat after the first shaping, and plan the cutting path based on the three-dimensional contour information; Input the pre-stored meat texture data and the first shaping parameters into the springback prediction model to predict the contour springback amount; The secondary shaping control parameters are determined based on the contour rebound amount and the interval between the first and second shaping, and the boneless raw meat is then subjected to secondary shaping based on the secondary shaping control parameters. The visual parameters and component parameters are input into the intelligent decision-making model, which outputs ultrasonic cutting parameters. The ultrasonic cutting tool is driven by the ultrasonic cutting parameters, and the ultrasonic cutting tool is controlled to feed intermittently along the cutting path to perform quantitative cutting.

[0005] According to the ultrasound-assisted intelligent quantitative cutting method for boneless raw meat provided by the present invention, the training method of the intelligent decision model includes: Construct a sample set, which includes boneless raw meat samples of various types and parts, as well as visual parameters, compositional parameters and optimal ultrasonic cutting parameters determined experimentally for each sample; The initial deep learning model is trained using the sample set. The initial deep learning model has a first branch for processing visual parameters and a second branch for processing component parameters. The outputs of the two branches are fused by a feature fusion module to map to ultrasonic cutting parameters. The parameters of the initial deep learning model are optimized according to the loss function corresponding to the ultrasonic cutting parameters. After the iteration termination condition is met, the trained deep learning model is used as the intelligent decision model.

[0006] According to the ultrasound-assisted intelligent quantitative cutting method for boneless raw meat provided by the present invention, the step of fusing the outputs of the two branches through a feature fusion module includes: The visual feature vector output from the first branch and the component feature vector output from the second branch are concatenated or weighted and summed. The attention mechanism is used to calculate the fusion vector after concatenation or weighted summation, and a weight vector reflecting the importance of different feature dimensions is generated. The visual feature vector and the component feature vector are reweighted using the weight vector to obtain a multimodal fusion feature vector for mapping to ultrasonic cutting parameters.

[0007] According to the ultrasound-assisted intelligent quantitative cutting method for boneless raw meat provided by the present invention, the step of inputting pre-stored meat texture data and the first shaping parameters into a springback prediction model to predict the contour springback amount includes: The viscoelastic parameters of the boneless raw meat are obtained based on the visual parameters and component parameters. Based on the shaping parameters and the three-dimensional contour information, a state vector describing the shaping process is constructed. The state vector includes at least the displacement of the shaping plate, the deflection angle, and the mechanical parameters fed back by the pressure sensor. The viscoelastic parameters, the state vector, and the three-dimensional contour information are input into the rebound prediction model, and the estimated contour of the boneless raw meat after the shaping force is removed is output as the contour rebound amount. The rebound prediction model adopts a dual-branch parallel architecture, including a first branch for processing three-dimensional geometric features and a second branch for processing physical property and process features, and dynamically weights and integrates the features of the two branches through an attention fusion layer.

[0008] According to the ultrasonic-assisted intelligent quantitative cutting method for boneless raw meat provided by the present invention, the secondary shaping of the boneless raw meat based on the secondary shaping control parameters includes: Based on the difference between the contour springback amount output by the springback prediction model and the three-dimensional contour information, the initial target control parameters of the secondary shaping fixture are determined. Based on the initial target control parameters and the interval between the first and second shaping, the second shaping control parameters are determined. The secondary shaping fixture is driven to move according to the secondary shaping control parameters, and the position and posture data and mechanical feedback data of the shaping plate are collected in real time during the movement. The real-time collected data is compared with the initial target control parameters. When a deviation between the actual contour and the target contour is detected, a fine-tuning increment is generated based on the residual learning model. The secondary shaping control parameters are dynamically adjusted based on the fine-tuning increment to compensate for the shape deviation during the meat rebound process in real time.

[0009] According to the ultrasonic-assisted intelligent quantitative cutting method for deboned raw meat provided by the present invention, the step of controlling the ultrasonic cutter to feed intermittently along the cutting path includes: The conveying device carrying the boneless raw meat is controlled to move in a stepping motion such that during each cutting action, only the slice area to be cut on the cutting path is exposed outside the constraint of the secondary shaping fixture, while the main body of the boneless raw meat remains constrained.

[0010] The present invention also provides an ultrasound-assisted quantitative cutting system for boneless raw meat, comprising: The conveying module is used to intermittently convey workstation trays carrying boneless raw meat along a preset path; The first scanning unit is used to acquire the visual parameters and composition parameters of the boneless raw meat. A primary shaping module is used to perform an initial shaping on the boneless raw meat based on the visual parameters and component parameters, and to record the primary shaping parameters. The second scanning unit is used to acquire high-precision three-dimensional contour information of the boneless raw meat. The secondary shaping module is used to perform secondary shaping on the boneless raw meat based on secondary shaping control parameters; An ultrasonic cutting module is located at or downstream of the secondary shaping module, and includes an ultrasonic cutting tool and its driving mechanism. The control unit is communicatively connected to the conveying module, the first scanning unit, the primary shaping module, the second scanning unit, the secondary shaping module, and the ultrasonic cutting module, respectively; wherein, the control unit is configured to execute the ultrasonic-assisted intelligent quantitative cutting method for deboned raw meat as described above.

[0011] According to the ultrasonic-assisted quantitative cutting system for deboned raw meat provided by the present invention, both the primary shaping module and the secondary shaping module include: a clamping frame; Multiple independently driveable shaping plates are used to contact and constrain the boneless raw meat from different directions; Pressure sensors and / or displacement sensors installed on the shaping plate are used to provide feedback on shaping force or displacement information. The driving parameters of the secondary shaping module are set by the control unit according to the secondary shaping control parameters.

[0012] The ultrasound-assisted quantitative cutting system for boneless raw meat provided by the present invention includes a conveying module comprising: Circular guide rail; At least two slitting station trays are movably mounted on the annular guide rail; A drive mechanism is used to drive the slitting station tray to move cyclically along the annular guide rail; The positioning actuator is used to precisely position the slitting station tray at the workstation corresponding to the primary shaping module, the secondary shaping module, or the ultrasonic cutting module.

[0013] The present invention provides an ultrasound-assisted intelligent quantitative slicing method and system for boneless raw meat. The method involves: acquiring visual and compositional parameters of the boneless raw meat; performing an initial shaping of the boneless raw meat based on these parameters to obtain initial shaping parameters; acquiring the three-dimensional contour information of the boneless raw meat after the initial shaping and planning a slicing path based on this contour information; inputting pre-stored meat texture data and the initial shaping parameters into a springback prediction model to predict the contour springback amount; determining secondary shaping control parameters based on the contour springback amount and performing a secondary shaping of the boneless raw meat based on these secondary shaping control parameters; and inputting the visual and compositional parameters into an intelligent decision-making model to output... Ultrasonic cutting parameters; the ultrasonic cutting tool is driven by the ultrasonic cutting parameters and controlled to feed intermittently along the cutting path to perform quantitative cutting. This invention achieves high-precision, low-damage quantitative cutting of boneless raw meat through a closed-loop strategy of "scanning-contour matching-secondary shaping" and adaptive decision-making of ultrasonic parameters driven by multimodal data. It effectively solves the contour mismatch problem caused by the springback of fresh meat shaping and ensures a high degree of consistency between the cutting path and the actual shape of the meat. Combined with ultrasonic cutting technology, it significantly reduces the "sticking" phenomenon and improves the quality of the cut surface and the integrity of the product. The entire process is automated, reducing manual intervention and greatly improving production efficiency and standardization. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0015] Figure 1 This is a flowchart of the ultrasound-assisted intelligent quantitative cutting method for boneless raw meat provided in this embodiment of the invention; Figure 2 This is a flowchart of the intelligent decision-making model processing provided in the embodiments of the present invention; Figure 3 This is a schematic diagram illustrating the determination of the shaping force provided in an embodiment of the present invention; Figure 4 This is a flowchart of the contour matching strategy provided in an embodiment of the present invention; Figure 5 This is a flowchart of the deep learning algorithm image processing provided in an embodiment of the present invention; Figure 6 This is a functional structure diagram of the ultrasound-assisted quantitative cutting system for deboned raw meat provided in an embodiment of the present invention; Figure 7 This is a structural diagram of the boneless raw meat visualization imaging module provided in an embodiment of the present invention; Figure 8 This is a structural diagram of the boneless raw meat adaptive shaping fixture provided in an embodiment of the present invention; Figure 9 This is a structural diagram of the multi-station annular guide rail conveying module provided in an embodiment of the present invention; Figure 10 This is a structural diagram of the ultrasonic cutting module provided in an embodiment of the present invention; Figure 11 This is a module connection diagram of the ultrasound-assisted quantitative cutting system for deboned raw meat provided in an embodiment of the present invention; Figure 12 This is a flowchart illustrating the specific process of segmentation provided in an embodiment of the present invention; Figure 13 This is a schematic diagram of the slicing path planning and the moving distance for each slicing provided in the embodiments of the present invention. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0017] Figure 1 The flowchart of the ultrasound-assisted intelligent quantitative cutting method for boneless raw meat provided in the embodiments of the present invention is as follows: Figure 1 As shown, the ultrasound-assisted intelligent quantitative cutting method for boneless raw meat provided in this embodiment of the invention includes: Step 101: Obtain the visual parameters and composition parameters of the boneless raw meat; Step 102: Perform the first shaping on the boneless raw meat and obtain the shaping parameters; Step 103: Obtain the three-dimensional contour information of the boneless raw meat after the first shaping, and plan the cutting path based on the three-dimensional contour information; Step 104: Input the pre-stored meat texture data and the first shaping parameters into the springback prediction model to predict the contour springback amount; Step 105: Determine the secondary shaping control parameters based on the contour rebound amount and the interval between the first and second shaping, and perform secondary shaping on the boneless raw meat based on the secondary shaping control parameters. Step 106: Input the visual parameters and component parameters into the intelligent decision-making model, and output the ultrasonic cutting parameters; In this embodiment of the invention, the visual parameters include, but are not limited to, texture, color, and size information; the component parameters include, but are not limited to, fat content, protein content, and moisture content.

[0018] Step 107: Drive the ultrasonic cutter with the ultrasonic cutting parameters and control the ultrasonic cutter to feed intermittently along the cutting path to perform quantitative cutting.

[0019] Traditional meat cutting equipment suffers from several core problems in deboned raw meat cutting, including low precision, poor production efficiency, insufficient automation, "sticking" of mechanical blades, unknown optimal ultrasonic parameters for different deboned raw meats, and low contour matching after shaping.

[0020] The ultrasound-assisted intelligent quantitative slicing method for boneless raw meat provided in this invention involves: acquiring visual and component parameters of the boneless raw meat; performing an initial shaping of the boneless raw meat based on the visual and component parameters to obtain initial shaping parameters; acquiring the three-dimensional contour information of the boneless raw meat after the initial shaping and planning a slicing path based on the three-dimensional contour information; inputting pre-stored meat texture data and the initial shaping parameters into a springback prediction model to predict the contour springback amount; determining secondary shaping control parameters based on the contour springback amount and performing a secondary shaping of the boneless raw meat based on the secondary shaping control parameters; and inputting the visual and component parameters into an intelligent decision-making model to output... Ultrasonic cutting parameters; the ultrasonic cutting tool is driven by the ultrasonic cutting parameters and controlled to feed intermittently along the cutting path to perform quantitative cutting. This invention achieves high-precision, low-damage quantitative cutting of boneless raw meat through a closed-loop strategy of "scanning-contour matching-secondary shaping" and adaptive decision-making of ultrasonic parameters driven by multimodal data. It effectively solves the contour mismatch problem caused by the springback of fresh meat shaping and ensures a high degree of consistency between the cutting path and the actual shape of the meat. Combined with ultrasonic cutting technology, it significantly reduces the "sticking" phenomenon and improves the quality of the cut surface and the integrity of the product. The entire process is automated, reducing manual intervention and greatly improving production efficiency and standardization.

[0021] In this invention, due to the significant differences in component parameters such as fat content, moisture content, protein content, and viscoelasticity, as well as visual parameters such as texture, color, and size, between different types and parts of boneless raw meat, a single ultrasonic parameter is insufficient to guarantee the quality of the cut meat after processing. Using the same ultrasonic parameter can easily lead to a decline in meat quality. To address this issue, this invention establishes an intelligent decision-making model that calculates the most suitable ultrasonic parameters for different types of boneless raw meat. These ultrasonic parameters include, but are not limited to, ultrasonic amplitude, ultrasonic frequency, and cutting speed.

[0022] Based on any of the above embodiments, the training method of the intelligent decision-making model includes: Step 201: Construct a sample set, which includes boneless raw meat samples of various types and parts, as well as visual parameters, composition parameters and optimal ultrasonic cutting parameters determined by experiments for each sample. Step 202: Train the initial deep learning model using the sample set. The initial deep learning model has a first branch for processing visual parameters and a second branch for processing component parameters. The outputs of the two branches are fused by the feature fusion module to map to ultrasonic cutting parameters. Step 203: Optimize the parameters of the initial deep learning model according to the loss function corresponding to the ultrasonic cutting parameters. After the iteration termination condition is met, use the trained deep learning model as the intelligent decision model.

[0023] In this embodiment of the invention, the intelligent decision-making model processing flow is as follows: Figure 2 As shown, 1000 boneless raw meat samples from different cuts of beef, pork, and lamb, such as tenderloin, hind leg, and pork belly, as well as salmon belly, were first collected. The ratio of training set, test set, and validation set was 7:1.5:1.5. The texture, color, and size visual parameters, as well as component parameters such as fat content, moisture content, protein content, and viscoelasticity of the training set were acquired simultaneously using a 3D laser scanner and a near-infrared spectral imager. Subsequently, the optimal ultrasonic parameters for different boneless raw meats were selected through orthogonal experiments to construct a multimodal dataset.

[0024] The model is built using the PyTorch framework and employs a dual-branch parallel structure. The visual branch uses ResNet34 as its backbone, retaining the ability to extract general features such as texture, color, and size, compressing high-dimensional image information into multi-dimensional visual feature vectors. The component channel constructs a fully connected neural network, establishing coupling relationships between fat content, protein content, water content, and viscoelasticity, outputting multi-dimensional component feature vectors. The core fusion module introduces an attention mechanism, adjusting the importance ratio of visual and component features through a weight matrix, strengthening the role of key parameters in both visual and component parameters, and generating multi-dimensional feature vectors. The parameter mapping layer sets up three parallel regression sub-networks, corresponding to the prediction of ultrasonic frequency, ultrasonic amplitude, and cutting speed, respectively. During training, overfitting is suppressed through regularization and cross-validation. Combined with feature importance analysis and parameter sensitivity experiments, the network layer dimensions and attention weight allocation are further optimized, ultimately obtaining a basic model with both accuracy and generalization ability. After model training is complete, a validation set is imported into the model. Based on the results of the validation set, the model is evaluated, and various parameters are adjusted. Once the validation set samples converge, the test set is imported into the model to perform a generalization test on the model, testing the model's prediction speed and accuracy, in preparation for the subsequent use of the ultrasonic scalpel-assisted deboned raw meat quantitative cutting device.

[0025] When boneless raw meat passes through the first visualization imaging unit, a laser scanner and spectrometer scan the meat to acquire its visual and compositional parameters. The model then feeds the scanned image into a ResNet34, where alternating convolution and pooling layers capture features such as fiber thickness, direction, color, and basic outline, generating a visually structured output vector. Subsequently, parameters such as fat content, protein content, and moisture content are extracted, viscoelasticity is predicted, and these parameters are input into the compositional branch FCN. Through nonlinear transformation, a compositional feature vector is generated.

[0026] In this embodiment of the invention, fusing the outputs of the two branches through the feature fusion module includes: Step 2021: Concatenate or weight the visual feature vector output by the first branch and the component feature vector output by the second branch; Step 2022: Calculate the fusion vector after splicing or weighted summation based on the attention mechanism to generate a weight vector that reflects the importance of different feature dimensions; Step 2023: Reweight the visual feature vector and the component feature vector using the weight vector to obtain a multimodal fusion feature vector for mapping to ultrasonic cutting parameters.

[0027] In this embodiment of the invention, after obtaining two feature vectors, the model introduces these two vectors into the attention fusion module. Based on the influence weights of each feature dimension on the cutting quality obtained from previous sample experiments, key factors affecting the cutting quality, such as fat content and protein content, are weighted, while the influence of secondary factors on the results is weakened. Finally, the two vectors are fused into a multimodal vector. This multimodal vector is then introduced into a regression network to accurately predict the most suitable ultrasonic amplitude, ultrasonic frequency, and cutting speed for this type of boneless raw meat. This data is then transmitted to the lower-level ultrasonic scalpel flexible quantitative cutting module. The ultrasonic generator adjusts the ultrasonic parameters in real time based on this data to quantitatively cut the boneless raw meat.

[0028] This invention not only achieves the matching of different boneless raw meat with optimal ultrasound parameters, but also establishes the matching relationship between different visual parameters, component parameters, and optimal ultrasound parameters. When faced with the subsequent cutting of unused boneless raw meat from the sample, this intelligent decision-making model can still evaluate the most suitable ultrasound parameters through its visual and component parameters.

[0029] Based on any of the above embodiments, the step of inputting the pre-stored meat texture data and the first shaping parameters into the rebound prediction model to predict the contour rebound amount includes: Step 301: Obtain the viscoelastic parameters of the boneless raw meat; Step 302: Based on the first shaping parameters and the three-dimensional contour information, construct a state vector describing the first shaping process. The state vector includes at least the displacement of the shaping plate, the deflection angle, and the mechanical parameters fed back by the pressure sensor. Step 303: Input the viscoelastic parameters, the state vector, and the three-dimensional contour information into the rebound prediction model, and output the estimated contour of the boneless raw meat after the shaping force is removed, as the contour rebound amount; The rebound prediction model adopts a dual-branch parallel architecture, including a first branch for processing three-dimensional geometric features and a second branch for processing physical property and process features, and dynamically weights and integrates the features of the two branches through an attention fusion layer.

[0030] In this embodiment of the invention, after the deboned meat arrives at the shaping station and is positioned, the host computer sends a signal to the adaptive shaping fixture for the deboned meat. Upon receiving the signal, the longitudinal module servo motor and the upper push rod motor first start working, driving the entire adaptive shaping fixture for the deboned meat to move downwards until the upper shaping plate contacts the deboned meat and the force sensor detects the force, at which point it stops. Subsequently, the transverse module servo motor starts working, driving the two shaping plates on both sides to move towards the center position of the fixture. At the same time, the push rod motors on both sides push the two shaping plates to deflect at a specific angle, applying a controllable shaping force to the deboned meat. When the data fed back by the force sensor reaches the preset shaping force, the motor stops working. At this time, the moving distance and angle of each shaping plate of the adaptive shaping fixture for the deboned meat are recorded, and the force sensor data is uploaded to the host computer. This method can adapt to the shaping of deboned meat with different contours and sizes.

[0031] like Figure 3 As shown, the upper shaping plate is already in contact with and locked to the meat before the shaping process begins. Therefore, during the shaping process, only the shaping force in the y-direction (i.e., the width direction of the meat) needs to be considered. The relationship between the compression amount and the proportionality coefficient k of the actual width and the deformation s is as follows: This refers to the displacement distance of the two shaping plates after sensing the force, where W is the actual width and θ is the angle between the two shaping plates and the horizontal plane. Combining equations (1) and (2), the displacement distance of the shaping plates can be obtained. Relationship with the proportionality constant k: Subsequently, based on the Hertzian contact formula, the relationship for the contact width 'a' is as follows: Where l is the total contact length between the meat and the shaping fixture, R is the radius of the roller, and E* is the equivalent elastic modulus, the value of which is: v c It is the Poisson's ratio of the roller, v m It is the Poisson's ratio of meat products. It is the elastic model of the roller. This refers to the elastic modulus of the meat product. Since the rollers are made of P210 material, their elastic modulus is much greater than that of the meat product. Therefore, the equivalent elastic modulus is... It can be approximated as the elastic modulus of meat. ,Right now: a is related to the displacement of the roller: After sorting (4), the shaping force of the shaping fixture can be obtained as follows: Combining (3), (6), (7), and (8), we get: Where n is the number of rollers, H is the distance between the upper shaping plate and the cutting station tray, and F is the shaping force applied to the meat by the shaping fixture. The shaping process needs to both shape irregularly shaped boneless raw meat into a regular shape and maintain the meat's characteristics, avoiding damage to its internal structure. Testing showed that when k is 5%, the meat achieves the best shaping effect.

[0032] Because this invention is applied to the slicing of fresh boneless raw meat, the meat will spring back after the first shaping. Therefore, scanning and imaging are required immediately after the first shaping to obtain the 3D contour and point cloud information of the boneless raw meat after the first shaping. Since fresh meat has significant viscoelastic properties, it is prone to elastic rebound after the first shaping, causing a deviation between the actual contour of the material and the initial shaped form. Directly using the parameters from the first shaping will affect the adaptability of the slicing path. To ensure accurate contour matching after two shaping processes, such as... Figure 4 As shown, the embodiments of the present invention employ the following rebound prediction model and contour matching strategy: First, various cuts of beef, pork, and lamb, such as tenderloin, hind leg, and belly, as well as different types and cuts of boneless raw meat like salmon belly, were selected and divided into several groups based on their size. For example, the ratio of the training set, test set, and validation set was 7:1.5:1.5. Subsequently, the viscoelasticity, three-dimensional profile, first-stage shaping parameters, and three-dimensional profile before second-stage shaping of each meat product were measured experimentally. The specific experimental procedures are as follows: The meat is scanned by a first laser scanner to obtain its viscoelasticity. Then, based on the size of the meat, the shaping parameters are calculated using equation (9), and a first shaping is performed. Immediately after the first shaping, a three-dimensional laser scanner is triggered to obtain the three-dimensional information of the meat. Finally, after the time required for the second scan and the second shaping, the meat is scanned a third time to obtain the outline of the meat before the second shaping. Based on the final outline, rebound prediction is performed for different types of meat. The model built on the PyTorch framework adopts a multi-branch parallel model structure. It will predict rebound and adjust the parameters of the second shaping based on the obtained viscoelasticity, three-dimensional information of the meat, and the first shaping parameters.

[0033] First, the meat sample passes through the first scanning imaging unit. A 3D laser scanner and a near-infrared spectrometer collect parameters such as viscoelasticity, providing crucial physical property input for subsequent springback modeling. Then, the meat sample moves to a shaping fixture, where it undergoes a shaping process, and the fixture's baseline data is recorded, including the movement positions δ of each baffle. 11 δ 12 Deflection angle θ 11 θ 12 and the real-time mechanical feedback parameter F collected by the pressure sensor 11 F 12 F 13 These data together constitute a state vector describing a shaping process. After a shaping process is completed, the meat product will be scanned by a second imaging unit to obtain the three-dimensional contour information of the shaped meat product. The three-dimensional convolutional neural network will output precise geometric features including the overall deformation trend of the meat product, such as its actual length, width, height, angle, and curvature.

[0034] The data are then fused to form a complete input feature set, which is then imported into a multimodal contour rebound prediction model. This model employs a dual-branch parallel architecture, where the 3D geometric feature branch is composed of a 3D convolutional neural network, and the physical property and process branch is composed of a fully connected network. The meat viscoelasticity and primary shaping parameters are dynamically weighted and integrated in an attention fusion layer, ultimately outputting an accurate prediction of the rebound at various locations on the meat, yielding the 3D contour before secondary shaping and its corresponding 2D height map.

[0035] like Figure 5As shown, the 2D height map of the meat obtained from the 3D laser scanner and the predicted 2D height map before secondary shaping are thresholded to segment the meat contour maps at different heights. These images are then binarized to separate the contours and background, extracting the contour lines at different heights. The multidimensional array format of the contour lines is then converted into an N×2 point set, including the x and y coordinates of each point. After obtaining the point set of contour points, the algorithm first separates the x and y coordinates of the contour points and performs Gaussian smoothing to further reduce the impact of local fluctuations on curvature calculation. Subsequently, the first derivatives of the smoothed x and y coordinates of each contour point are calculated. and second derivative , This is used to reflect the tangent direction and rate of change of the tangent in the contour; then, the curvature of the plane curve is calculated using the formula: Finally, the curvature value of each contour point is obtained; a larger curvature indicates a higher degree of bending at that point. Multiple sub-images are then created to display the corresponding height contour curvature maps of the meat product's 2D height image acquired by the 3D laser scanner and the predicted 2D height image before secondary shaping. Different colors are used to represent the curvature magnitude, and statistical information such as the number of contour points, the maximum, minimum, and average curvature are output to quantify the analysis results. Anomaly handling is also incorporated throughout the process to ensure the algorithm's robustness. Finally, the contour curvature maps of corresponding heights are compared to calculate the required distance δ between the shaping plates on both sides and the upper shaping plate of the secondary shaping fixture. 21 δ 22 Deflection angle θ 21 θ 22 and various pressure sensors F 21 F 22 F 23 The data.

[0036] Based on any of the above embodiments, the secondary shaping of the boneless raw meat based on the secondary shaping control parameters includes: Based on the difference between the contour springback amount output by the springback prediction model and the three-dimensional contour information, the initial target control parameters of the secondary shaping fixture are determined. The initial target control parameters are adjusted based on the interval between the first and second shaping, and the second shaping control parameters are determined. The secondary shaping fixture is driven to move according to the secondary shaping control parameters, and the position and posture data and mechanical feedback data of the shaping plate are collected in real time during the movement. The real-time collected data is compared with the initial target control parameters. When a deviation between the actual contour and the target contour is detected, a fine-tuning increment is generated based on the residual learning model. The secondary shaping control parameters are dynamically adjusted based on the aforementioned fine-tuning increment to compensate for the shape deviation during the meat's rebound process in real time.

[0037] For example, the initial target control parameters for the secondary shaping fixture were calculated as follows: inward displacement of the two shaping plates by 3.8 mm, deflection angle of 45 degrees, downward displacement of the upper shaping plate by 2.5 mm, and shaping force threshold of 52 N. However, since the meat has already rebounded by 3.5 mm in width and 2.0 mm in height within a 3-second interval, directly using the above initial parameters for secondary shaping will not be able to restore the meat to the target contour. Therefore, this embodiment of the invention performs a timing correction on the initial target control parameters based on a 3-second interval: according to the rebound rate and interval time, the displacement compensation of the two shaping plates is increased by 0.4 mm, the displacement compensation of the upper shaping plate is increased by 0.3 mm, and the shaping force threshold is correspondingly increased to 55 N. The final determined secondary shaping control parameters are: displacement of the two shaping plates by 4.2 mm, deflection angle of 45 degrees, displacement of the upper shaping plate by 2.8 mm, and shaping force threshold of 55 N. The secondary shaping fixture operates according to the modified parameters, which can compensate for the rebound that occurs during the conveying process before the secondary shaping begins, so that the actual contour of the meat product after secondary shaping is highly consistent with the target contour, with a similarity of over 98%.

[0038] During the secondary shaping process, there is a time interval between the boneless raw meat being transported from the primary shaping station to the secondary shaping station. During this period, the meat continuously rebounds due to its viscoelastic properties, causing its actual contour to change compared to its state immediately after the primary shaping. To accurately compensate for the additional rebound that occurs during this interval, this embodiment of the invention modifies the initial target control parameters based on the interval, enabling the secondary shaping fixture to begin operation with parameters that more closely match the actual shape of the current meat.

[0039] Building upon this foundation, a closed-loop correction mechanism based on real-time feedback is further introduced to ensure the accuracy of contour matching. Once the secondary shaping fixture begins operation according to the target parameters predicted by the model, such as the target movement distance, deflection angle, and constraint force threshold of each baffle, the pressure and angle sensors integrated on the fixture continuously collect real-time mechanical feedback and pose data, comparing them in real-time with the algorithm's predicted values. If a deviation is detected between the actual contour and the target contour, the system immediately triggers the online correction module. A lightweight neural network based on residual learning outputs real-time fine-tuning increments, transforming the deviation into dynamic adjustment commands for the position, angle, and constraint force of the secondary shaping fixture baffles. This real-time offsets morphological shifts caused by viscoelastic nonlinearity, uneven tissue structure, or external interference during the meat's rebound process. Through this dual guarantee mechanism of time-series correction based on intervals and real-time closed-loop correction, the contours of the deboned raw meat after the two shaping processes are highly consistent, with a stable similarity exceeding 98%. This provides a core guarantee for the precise execution of subsequent cutting paths and the accuracy of quantitative cutting.

[0040] Based on any of the above embodiments, controlling the ultrasonic cutter to feed intermittently along the cutting path includes: The conveying device carrying the boneless raw meat is controlled to move in a stepping motion such that during each cutting action, only the slice area to be cut on the cutting path is exposed outside the constraint of the secondary shaping fixture, while the main body of the boneless raw meat remains constrained.

[0041] The ultrasonic-assisted intelligent quantitative cutting method for boneless raw meat provided in this invention establishes an intelligent decision-making model for ultrasonic parameters of boneless raw meat based on visual and component parameters of different types and parts of boneless raw meat, combined with artificial intelligence and machine learning. This model outputs the optimal ultrasonic parameters for different types of boneless raw meat, maximizing the quality of different types of boneless raw meat. By integrating multi-source data such as primary shaping parameters, meat viscoelasticity, and meat contour, the method uses deep learning algorithms to predict meat rebound and contour. Through curvature analysis and real-time feedback closed-loop adjustment of secondary shaping parameters, the consistency of meat contour after two shaping processes is ensured. This solves the industry pain points of traditional mechanical knives, guarantees the quality of boneless raw meat after quantitative cutting, and improves the level of intelligent standardization in meat pre-processing.

[0042] The ultrasonic-assisted quantitative cutting system for deboned raw meat provided by the present invention will be described below. The ultrasonic-assisted quantitative cutting system for deboned raw meat described below can be referred to in correspondence with the ultrasonic-assisted intelligent quantitative cutting method for deboned raw meat described above.

[0043] Figure 6 This is a functional structure diagram of the ultrasound-assisted quantitative cutting system for boneless raw meat provided in an embodiment of the present invention, as shown below. Figure 6As shown, the ultrasound-assisted quantitative cutting system for boneless raw meat provided in this embodiment of the invention includes: The conveying module 601 is used to intermittently convey workstation trays carrying deboned raw meat along a preset path. The first scanning unit 602 is used to acquire the visual parameters and composition parameters of the boneless raw meat. A primary shaping module 603 is used to perform an initial shaping on the boneless raw meat based on the visual parameters and component parameters, and to record the primary shaping parameters. The second scanning unit 604 is used to acquire high-precision three-dimensional contour information of the boneless raw meat. The secondary shaping module 605 is used to perform secondary shaping on the boneless raw meat based on secondary shaping control parameters. The ultrasonic cutting module 606 is located at or downstream of the secondary shaping module and includes an ultrasonic cutter and its driving mechanism. The control unit 607 is communicatively connected to the conveying module, the first scanning unit, the primary shaping module, the second scanning unit, the secondary shaping module, and the ultrasonic cutting module, respectively; wherein, the control unit is configured to execute the ultrasonic-assisted intelligent quantitative cutting method for deboned raw meat as described in any of the above embodiments.

[0044] In an embodiment of the present invention, The first scanning unit and the second scanning unit are composed of (as shown in the example) Figure 7 The system (as shown) consists of a scanning imaging enclosure, photoelectric switches, a 3D laser scanner, and a host computer. The photoelectric switch triggers the 3D laser scanner upon sensing the arrival of the cutting station tray, ensuring accurate imaging. The first 3D laser scanner, along with a near-infrared spectral imager, scans the boneless raw meat after triggering, acquiring component parameters such as moisture content, protein content, and fat content, as well as visual parameters such as texture, color, and initial contour. This information is then transmitted to the host computer. The second 3D laser scanner scans the boneless raw meat, acquiring high-precision 3D contours and point cloud information, and transmits this data to the host computer. The host computer processes the received information, using a built-in deep learning intelligent decision-making model for boneless raw meat and a quantitative meat cutting intelligent algorithm to calculate the optimal ultrasonic parameters and positioning parameters for each cut. This information is then sent to the lower-level computer, providing technical support for subsequent quantitative cutting.

[0045] In this embodiment of the invention, both the primary shaping module and the secondary shaping module include: a clamping frame; Multiple independently driveable shaping plates are used to contact and constrain the boneless raw meat from different directions; Pressure sensors and / or displacement sensors installed on the shaping plate are used to provide feedback on shaping force or displacement information. The driving parameters of the secondary shaping module are set by the control unit according to the secondary shaping control parameters.

[0046] In this embodiment of the invention, the adaptive shaping fixture for deboned raw meat primarily shapes the initially irregular shape of the meat into a regular shape, reducing obstruction and improving the accuracy of the 3D laser scanner, thereby ensuring the precision of subsequent quantitative cutting. The adaptive shaping fixture consists of two sets of identical shaping fixtures: a primary shaping fixture and a secondary shaping fixture, to ensure the consistency of the deboned raw meat contour during scanning and cutting. The shaping fixture comprises a transverse module, a longitudinal module, two side shaping plates, two side push rod motors, an upper shaping plate, an upper push rod motor, an angle sensor, and a pressure sensor (e.g., ...). Figure 8 (As shown). The horizontal and vertical modules enable the movement of the side shaping plates and the upper baffle to accommodate boneless raw meat of different shapes and sizes. The side push rod motors can drive the side shaping plates to adjust within an angle range of 30°-75° with the horizontal plane, achieving adaptive shaping. The upper push rod motor can drive the upper shaping plate to move and press on the meat to prevent the rear end of the meat from curling up due to ultrasonic knife cutting. Angle and pressure sensors are installed on the side shaping plates to record and provide feedback on the deflection angle of the shaping plates and the force applied to the meat, providing guidance for secondary shaping.

[0047] In this embodiment of the invention, the conveying module includes: Circular guide rail; At least two slitting station trays are movably mounted on the annular guide rail; A drive mechanism is used to drive the slitting station tray to move cyclically along the annular guide rail; The positioning actuator is used to precisely position the slitting station tray at the workstation corresponding to the primary shaping module, the secondary shaping module, or the ultrasonic cutting module.

[0048] This module mainly consists of a circular guide rail, a slitting station tray, a positioning cylinder, and a photoelectric switch (such as...). Figure 9 (As shown). The slitting station tray is the carrying unit for boneless raw meat, responsible for transporting the boneless raw meat to the designated working position to complete the corresponding process. A baffle is installed at the rear of the tray to prevent the meat from shifting due to cutting pressure during ultrasonic slitting. Subsequent positioning is also based on this baffle to precisely control the movement of the entire annular guide rail. The photoelectric switch and secondary positioning cylinder installed on the annular guide rail provide precise positioning, ensuring the movement accuracy of the annular guide rail.

[0049] In this embodiment of the invention, the ultrasonic cutting module mainly consists of a four-axis module, an ultrasonic generator, and an ultrasonic cutter (e.g., Figure 10 (As shown). The four-axis module includes an X-axis (parallel to the direction of movement of the ring guide rail), a Y-axis (horizontal direction perpendicular to the direction of movement of the ring guide rail), a Z-axis (vertical direction), and an R-axis (rotation direction around the ultrasonic generator). It can drive the ultrasonic scalpel to the required position and rotation angle at a speed of 0-100 mm / s. The ultrasonic generator is the core of the ultrasonic scalpel flexible quantitative slitting module, providing high-frequency ultrasonic vibration to the ultrasonic scalpel to achieve slitting. This ultrasonic generator can provide the ultrasonic scalpel with a vibration frequency of 20k-40kHz and an ultrasonic amplitude range of 0-50%, which can adjust the cutting parameters in real time according to different boneless raw meat, ensuring that the scalpel efficiently cuts muscle fibers and connective tissue while improving meat quality and reducing meat loss.

[0050] The overall structure of the ultrasound-assisted quantitative cutting system for boneless raw meat provided in this embodiment of the invention is as follows: Figure 11 As shown, the system includes, but is not limited to, a multi-station ring rail conveying module, a deboned meat visualization imaging module, a deboned meat adaptive shaping module, and an ultrasonic scalpel flexible cutting module. The ultrasonic-assisted quantitative cutting system for deboned raw meat provided in this embodiment of the invention can achieve fully automated operation of the entire process, including feeding, shaping, scanning, cutting, and sorting. Through efficient collaboration between processes, it significantly optimizes overall processing efficiency and accuracy, forming a closed-loop continuous processing system.

[0051] like Figure 12 As shown, the specific slicing process provided in this embodiment of the invention includes: First, the entire machine is initialized. The first slitting station tray will move precisely to the deboned raw meat loading station. The operator places a whole piece of deboned raw meat in the center of the slitting station tray and makes the tail of the meat contact the baffle at the rear of the slitting station tray.

[0052] After the loading process is completed, the operator presses a button to send a movement signal to the host computer. Upon receiving the signal, the multi-station circular guide rail conveyor module will move the slitting station tray forward to the primary shaping station (which is also the end point of the previous slitting operation). The tray first passes through the first scanning unit, where a laser scanner and a near-infrared spectral imager acquire visual parameters such as texture, color, and initial contour of the meat, as well as component parameters such as fat content, moisture content, viscoelasticity, and protein content. This information is then transmitted to the host computer. Through the ultrasonic parameter intelligent decision-making model based on multimodal deep learning of deboned raw meat, the optimal ultrasonic parameters such as ultrasonic frequency, ultrasonic amplitude, and cutting speed are calculated for the meat. This information is then transmitted to the ultrasonic knife flexible quantitative slitting module. Subsequently, when the slitting station tray moves to the point where its rear baffle is flush with the rear end of the primary shaping fixture, the secondary positioning cylinder of the multi-station circular guide rail conveyor module is triggered, allowing the tray to stop precisely at the primary shaping station. At this point, the deboned raw meat is completely constrained within the primary shaping fixture. After the slitting station tray stops, it sends a signal to the host computer. Upon receiving the signal, the primary shaping fixture moves inward according to the set control method. Under the real-time mechanical feedback of the pressure sensor, it applies a controllable constraint force to the irregularly shaped boneless raw meat, initially shaping it into a relatively regular geometric shape. The mechanical and deformation data collected during this process are recorded and transmitted to the system simultaneously, providing guidance for subsequent secondary shaping. This ensures that the material shape tends to be consistent after two consecutive shapings, laying the foundation for subsequent high-precision visual recognition and slitting path planning.

[0053] After one shaping cycle, the multi-station circular guide rail conveyor module drives the slitting station tray forward into the boneless raw meat visualization imaging module. When the front end of the tray reaches the preset photoelectric sensor detection area, the sensor is triggered and generates a trigger signal, simultaneously activating the high-precision 3D laser scanner. The scanner then performs comprehensive, multi-angle acquisition of surface contour and depth information of the moving boneless raw meat, obtaining its complete contour and point cloud data. The 3D laser scanner can upload the acquired high-resolution depth image data to the host computer in real time. On the host computer, after receiving the depth image data, the system immediately calls the built-in image processing algorithm sequence to perform a series of operations on the raw data, including denoising, registration, feature extraction, and 3D point cloud reconstruction, generating a 3D meat quality model that can be used for path planning. Based on this model, the system automatically generates the optimal slitting path that meets the set weight requirements and shape constraints through an intelligent planning module integrating deep learning algorithms. The planning results are compiled into control instructions and sent to the lower-level PLC and motion controller.

[0054] After the slitting station tray completes scanning, it moves forward with the multi-station circular guide rail conveyor module according to a preset rhythm, arriving at the secondary shaping station (the loading position of the next station). Using the rear baffle of the slitting station tray as a reference, it aligns with the rear end of the secondary shaping fixture, at which point the boneless raw meat is completely constrained within the secondary shaping fixture. Due to factors such as the viscoelasticity of the meat, it may spring back after the second shaping, making it difficult to match the initial contour. Therefore, the secondary shaping fixture uses the movement positions and angles of each baffle recorded by the primary shaping fixture, as well as the mechanical parameters of the pressure sensors, as a basis. Based on the three-dimensional information of the meat obtained by the scanner, such as length, width, height, and meat angles, it processes this data using deep learning algorithms, combined with the texture and viscoelasticity of the meat. This results in the distance and angle that each baffle of the secondary shaping fixture needs to move. Real-time adjustments and corrections are made based on data from the pressure sensors to ensure that the contour of the boneless raw meat is consistent after both shaping processes, guaranteeing the feasibility of path planning and the accuracy of quantitative slitting. At the same time, the second slitting station pallet arrives at the loading position, where the operator loads the second pallet.

[0055] After receiving the optimal ultrasonic parameters for deboned meat from the host computer, the ultrasonic scalpel flexible quantitative slitting module adjusts the ultrasonic amplitude and frequency of the ultrasonic scalpel in real time via the ultrasonic generator, and adjusts the movement speed of the four-axis module and the cutting speed of the Z-axis via the PLC. During slitting, the force generated by the ultrasonic scalpel causes severe deformation of the deboned meat, which seriously interferes with the slitting path generated by the host computer's path planning algorithm. Therefore, this invention integrates the secondary shaping station and the slitting station in the same location. After the first tray completes secondary shaping and the second tray completes loading, the secondary shaping fixture sends a signal to the host computer. The multi-station circular guide rail conveying module moves to the state where the secondary shaping fixture is just exposed at the first cutting position, i.e., moves a distance X, according to the slitting path calculated by the path planning algorithm. Where L is the total length of the tray at the slitting station. It refers to the length of the meat. This is the distance from the first cutting position to the head of the meat. At this point, only the first slice of meat to be cut is exposed to the outside of the secondary shaping fixture; the main body of the boneless raw meat remains within the secondary shaping fixture and is subjected to the shaping force applied during the secondary shaping process. Subsequently, the multi-station circular guide rail conveyor module moves intermittently, cooperating with the ultrasonic scalpel flexible cutting device. Following this, the nth cut only requires moving X units of space. n The distance is sufficient; this slicing strategy can effectively avoid inaccurate slicing due to deformation during the slicing process (e.g., ...). Figure 13(As shown). When the first pallet completes its cutting process, the second pallet arrives at the primary shaping station, and the rear baffle of the second pallet aligns with the rear of the primary shaping station, sending a shaping request instruction to the host computer. Upon receiving the instruction, the primary shaping fixture begins operation, shaping the boneless raw meat in the second pallet. This system achieves synchronization of actions at each station through ingenious timing design.

[0056] The system determines that the deboned meat cutting instruction on the first cutting station tray is completely completed, and if the second station has also completed shaping, the host computer will immediately issue an instruction to drive the multi-station circular guide rail conveyor module to advance by a fixed cycle length. This allows the second cutting station tray to accurately enter the secondary shaping position to begin operation. The first cutting station tray will move to the arc at the end of the circular guide rail, where gravity will sort the already cut deboned meat. The subsequent third cutting station tray will move to the loading station to receive new deboned meat, thus starting a new processing cycle. Throughout the process, the encoder continuously feeds real-time position information to the PLC, forming a closed-loop, self-driven automated cycle. Under this cyclical logic, high-precision, high-consistency, and low-loss quantitative cutting of deboned meat is ultimately achieved.

[0057] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0058] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of software products. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0059] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An ultrasound-assisted intelligent quantitative cutting method for boneless raw meat, characterized in that, include: Obtain visual and compositional parameters of boneless raw meat; Based on the visual and component parameters, the boneless raw meat is first shaped to obtain the first shaping parameters; Obtain the three-dimensional contour information of the boneless raw meat after the first shaping, and plan the cutting path based on the three-dimensional contour information; Input the pre-stored meat texture data and the first shaping parameters into the springback prediction model to predict the contour springback amount; The secondary shaping control parameters are determined based on the contour rebound amount and the interval between the first and second shaping, and the boneless raw meat is then subjected to secondary shaping based on the secondary shaping control parameters. The visual parameters and component parameters are input into the intelligent decision-making model, which outputs ultrasonic cutting parameters. The ultrasonic cutting tool is driven by the ultrasonic cutting parameters, and the ultrasonic cutting tool is controlled to feed intermittently along the cutting path to perform quantitative cutting.

2. The ultrasound-assisted intelligent quantitative cutting method for boneless raw meat according to claim 1, characterized in that, The training method for the intelligent decision-making model includes: Construct a sample set, which includes boneless raw meat samples of various types and parts, as well as visual parameters, compositional parameters and optimal ultrasonic cutting parameters determined experimentally for each sample; The initial deep learning model is trained using the sample set. The initial deep learning model has a first branch for processing visual parameters and a second branch for processing component parameters. The outputs of the two branches are fused by a feature fusion module to map to ultrasonic cutting parameters. The parameters of the initial deep learning model are optimized according to the loss function corresponding to the ultrasonic cutting parameters. After the iteration termination condition is met, the trained deep learning model is used as the intelligent decision model.

3. The ultrasound-assisted intelligent quantitative cutting method for boneless raw meat according to claim 2, characterized in that, The process of fusing the outputs of the two branches through the feature fusion module includes: The visual feature vector output from the first branch and the component feature vector output from the second branch are concatenated or weighted and summed. The attention mechanism is used to calculate the fusion vector after concatenation or weighted summation, and a weight vector reflecting the importance of different feature dimensions is generated. The visual feature vector and the component feature vector are reweighted using the weight vector to obtain a multimodal fusion feature vector for mapping to ultrasonic cutting parameters.

4. The ultrasound-assisted intelligent quantitative cutting method for boneless raw meat according to claim 1, characterized in that, The step of inputting pre-stored meat texture data and the first shaping parameters into the springback prediction model to predict the contour springback amount includes: The viscoelastic parameters of the boneless raw meat are obtained based on the visual parameters and component parameters. Based on the shaping parameters and the three-dimensional contour information, a state vector describing the shaping process is constructed. The state vector includes at least the displacement of the shaping plate, the deflection angle, and the mechanical parameters fed back by the pressure sensor. The viscoelastic parameters, the state vector, and the three-dimensional contour information are input into the rebound prediction model, and the estimated contour of the boneless raw meat after the shaping force is removed is output as the contour rebound amount. The rebound prediction model adopts a dual-branch parallel architecture, including a first branch for processing three-dimensional geometric features and a second branch for processing physical property and process features, and dynamically weights and integrates the features of the two branches through an attention fusion layer.

5. The ultrasound-assisted intelligent quantitative cutting method for boneless raw meat according to claim 1 or 4, characterized in that, The secondary shaping of the boneless raw meat based on the secondary shaping control parameters includes: Based on the difference between the contour springback amount output by the springback prediction model and the three-dimensional contour information, the initial target control parameters of the secondary shaping fixture are determined. The initial target control parameters are adjusted based on the interval between the first and second shaping, and the second shaping control parameters are determined. The secondary shaping fixture is driven to move according to the secondary shaping control parameters, and the position and posture data and mechanical feedback data of the shaping plate are collected in real time during the movement. The real-time collected data is compared with the initial target control parameters. When a deviation between the actual contour and the target contour is detected, a fine-tuning increment is generated based on the residual learning model. The secondary shaping control parameters are dynamically adjusted based on the aforementioned fine-tuning increment to compensate for the shape deviation during the meat's rebound process in real time.

6. The ultrasound-assisted intelligent quantitative cutting method for boneless raw meat according to claim 1, characterized in that, The control of the ultrasonic cutter to intermittently feed along the cutting path includes: The conveying device carrying the boneless raw meat is controlled to move in a stepping motion such that during each cutting action, only the slice area to be cut on the cutting path is exposed outside the constraint of the secondary shaping fixture, while the main body of the boneless raw meat remains constrained.

7. An ultrasound-assisted quantitative cutting system for boneless raw meat, characterized in that, include: The conveying module is used to intermittently convey workstation trays carrying boneless raw meat along a preset path; The first scanning unit is used to acquire the visual parameters and composition parameters of the boneless raw meat. A primary shaping module is located on the path of the conveying module and is used to perform the first shaping on the boneless raw meat based on the visual parameters and component parameters and record the primary shaping parameters. The second scanning unit is used to acquire high-precision three-dimensional contour information of the boneless raw meat. The secondary shaping module is used to perform secondary shaping on the boneless raw meat based on secondary shaping control parameters; An ultrasonic cutting module is located at or downstream of the secondary shaping module, and includes an ultrasonic cutting tool and its driving mechanism. The control unit is communicatively connected to the conveying module, the first scanning unit, the primary shaping module, the second scanning unit, the secondary shaping module, and the ultrasonic cutting module, respectively; wherein, the control unit is configured to perform the ultrasonic-assisted intelligent quantitative cutting method for deboned raw meat according to any one of claims 1 to 6.

8. The ultrasound-assisted quantitative cutting system for boneless raw meat according to claim 7, characterized in that, Both the primary shaping module and the secondary shaping module include: Fixture frame; Multiple independently driveable shaping plates are used to contact and constrain the boneless raw meat from different directions; Pressure sensors and / or displacement sensors installed on the shaping plate are used to provide feedback on shaping force or displacement information. The driving parameters of the secondary shaping module are set by the control unit according to the secondary shaping control parameters.

9. The ultrasound-assisted quantitative cutting system for boneless raw meat according to claim 7, characterized in that, The conveying module includes: Circular guide rail; At least two slitting station trays are movably mounted on the annular guide rail; A drive mechanism is used to drive the slitting station tray to move cyclically along the annular guide rail; The positioning actuator is used to precisely position the slitting station tray at the workstation corresponding to the primary shaping module, the secondary shaping module, or the ultrasonic cutting module.

Images