Bone meat three-dimensional contour intelligent identification method and system based on image identification

Abstract

The invention is suitable for the field of meat processing and production, and provides an intelligent recognition method and system for the three-dimensional contour of boned meat based on image recognition, and the method comprises the steps: carrying out the multi-modal image collection of frozen boned meat, and capturing a surface crack pattern; analyzing the crack pattern to construct a topological network model; reconstructing a bone-meat three-dimensional holographic model based on the topological network model, and marking joint, marrow cavity and bone mineral density abnormal areas; identifying potential skeleton weak points; generating a double-path cutting guide line; cooperative operation and visual presentation of the guide lines are controlled through dynamic compensation. Precise recognition and intelligent cutting guide of the internal skeleton structure of the boned meat are achieved, the segmentation precision and efficiency are improved, and the method is suitable for the field of meat processing and the like.

Description

Technical Field

[0001] This invention belongs to the field of meat processing and production, and particularly relates to a method and system for intelligent recognition of the three-dimensional contour of bone-in meat based on image recognition. Background Technology

[0002] In meat processing, the food industry, and related production sectors, precise cutting of bone-in meat is crucial for improving product quality and production efficiency. With the development of image recognition and 3D reconstruction technologies, machine vision-based bone-in meat contour recognition methods are gradually replacing traditional manual judgment, becoming an important direction for industry technological upgrading. Currently, related technologies mostly acquire surface information through visible light imaging and depth sensing, combining this with 3D modeling algorithms to achieve contour recognition. However, their application is limited by interference from soft tissues such as fat and muscle covering the bone-in meat surface, and the accuracy of recognizing the internal skeletal structure still needs improvement. Simultaneously, the demand for intelligent cutting guidance technology is increasing, requiring recognition systems not only to present contours but also to provide depth information such as skeletal weak points and cutting path optimization to meet the high-precision requirements of automated production.

[0003] However, existing technologies struggle to penetrate soft tissues such as fat and muscle, making it difficult to accurately reconstruct the three-dimensional contours of the encased bones. This results in significant errors in labeling joint positions, medullary cavity orientation, and areas of abnormal bone density, and makes it impossible to effectively identify potential weak points in the bones beneath the muscle layer. Furthermore, the cutting guide lines are generated solely based on surface morphology, which can easily lead to bone fragmentation and muscle tissue damage during the cutting process. Moreover, it is difficult to address pose deviations during the cutting process, affecting segmentation accuracy and efficiency. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent recognition method for the three-dimensional contour of bone-in meat based on image recognition, aiming to solve the technical problems existing in the prior art as identified in the background art.

[0005] This invention is implemented as follows: an intelligent recognition method for three-dimensional contours of bone-in meat based on image recognition, the method comprising: Multimodal image acquisition was performed on frozen bone-in meat to capture surface crack patterns formed by tissue shrinkage; The surface crack pattern is analyzed to determine its bifurcation direction, extension speed, and spatial distribution, and a topological network model of the surface crack pattern is constructed. Based on the mapping relationship between the constructed topological network model and the three-dimensional contour of the skeleton, a three-dimensional holographic model of bone and flesh wrapped in fat and muscle is reconstructed, and the joint positions, medullary cavity orientation and abnormal bone density areas are marked. Based on the stress propagation pattern of the surface crack pattern, potential weak points of fragile bones in the muscle layer corresponding to the three-dimensional holographic model of bone and flesh are identified. Based on the three-dimensional model of the skeleton and the weak points of the skeleton, a dual-path cutting guide line is generated, including a muscle fiber-oriented cutting guide line for muscle tissue and a mechanically optimized cutting guide line for the bone structure. The collaborative operation of the dual-path cutting guide lines is controlled by dynamic compensation of three-dimensional spatial coordinates, and the bone-and-flesh holographic three-dimensional model and the dual-path cutting guide lines are presented simultaneously on the visualization interface.

[0006] As a further embodiment of the present invention, the multimodal image acquisition specifically includes: For bone-in meat during gradient freezing, multispectral imaging is used to scan the reflectance spectral characteristics of surface crack patterns, generate high-dimensional feature vectors, and reconstruct the three-dimensional geometric morphology of surface crack patterns. Record the dynamic formation sequence of surface crack patterns during freezing, generate a time-series image sequence, and mark the crack initiation time nodes and propagation paths.

[0007] As a further aspect of the present invention, the analysis of the surface crack pattern specifically includes: The U-Net convolutional neural network is used to segment the main trunk and branches of the surface crack pattern, extract the bifurcation angle features of the surface crack pattern, and generate an angle distribution histogram. Based on the time-series image sequence, the propagation velocity field of the surface crack pattern is calculated: ; in, Indicates the crack propagation rate. This indicates the displacement of the crack tip. For time intervals, for Coordinates of the crack tip at any given moment. For the first Frame timestamp; Based on the node coordinates of the surface crack pattern, a topological network model including vertex connection relationships is constructed, and a heatmap of node degree distribution and spatial density is output.

[0008] As a further aspect of the present invention, the reconstruction of the bone-and-flesh three-dimensional holographic model encased in fat and muscle specifically includes: The node coordinates of the topological network model are non-rigidly registered with the pre-stored skeletal CT template, and the spatial mapping function is calculated using the thin plate spline algorithm. Based on the spatial mapping function, the topological network model is mapped to a three-dimensional skeletal point cloud, and the orientation of the medullary cavity is reconstructed by combining depth sensing data. Based on multispectral imaging data, fat / muscle regions are segmented to generate a soft tissue layer that encapsulates the bone, forming a three-dimensional whole-system model of bone and muscle. In the aforementioned three-dimensional holographic model of flesh and bone: Extract the extreme points of bone curvature, match them with anatomical atlases to determine the coordinates of the joint space, and identify and label the joint positions; Calculate the standard deviation of local voxel gray values, mark areas exceeding the threshold range as abnormal areas, and identify areas with abnormal bone density. ; in, This represents the standard deviation of local voxel gray levels. The voxel grayscale value is obtained from a pre-stored skeletal CT template. For the number of voxels in a local region, This represents the average gray value of a local area. ; Abnormal area determination criteria: ; in, The standard deviation threshold, The osteoporosis threshold, This is the threshold for osteosclerosis.

[0009] As a further aspect of the present invention, the identification of potential fragile bone weak points in the muscle layer corresponding to the three-dimensional holographic model of bone and muscle specifically includes: Loading biomechanical parameters onto a three-dimensional holographic model of meat with bone; Finite element stress propagation simulation based on the propagation path of surface crack patterns; Identify the overlapping regions of regions with stress >150MPa and bone porosity >30% in the bone-bone 3D holographic model, and mark the overlapping regions as potential weak points in fragile bones.

[0010] As a further embodiment of the present invention, the generation of the dual-path cutting guide line specifically includes: Extract the texture direction features of the muscle layer in the bone-in 3D holographic model, and generate the cutting guide line along the main direction of the muscle fibers; On the bone-and-flesh three-dimensional holographic model, an obstacle area is set to avoid the weak points of the bone, and the mechanically optimized cutting guide line is generated along the bone suture interface; Establish a spatial co-constraint equation between the suture fiber cutting guide line and the mechanically optimized cutting guide line, and optimize the relative position of the paths.

[0011] As a further aspect of the present invention, the method of dynamically compensating for three-dimensional spatial coordinates to control the coordinated operation of the dual-path cutting guide lines specifically includes: The pose deviation of the cutting tool relative to the bone-in meat three-dimensional holographic model is acquired in real time; The dynamic compensation vector of the dual-path cutting guide line is calculated based on the pose deviation. The three-dimensional holographic model of bone and flesh is overlaid on the visualization interface; The execution status of the dual-path cutting guide line is rendered synchronously on the visualization interface.

[0012] Another object of the present invention is to provide an intelligent recognition system for three-dimensional contours of bone-in meat based on image recognition, the system comprising: The multimodal image acquisition module is used to acquire multimodal images of frozen bone-in meat and capture surface crack patterns formed by tissue shrinkage. The crack pattern analysis module is used to analyze the surface crack pattern, analyze the bifurcation direction, extension speed and spatial distribution law, and construct the topological network model of the surface crack pattern. The 3D holographic model reconstruction module is used to reconstruct a 3D holographic model of bone and flesh wrapped in fat and muscle based on the mapping relationship between the constructed topological network model and the 3D contour of the skeleton, and to annotate the joint positions, the direction of the medullary cavity and the areas of abnormal bone density. The fragile bone weak point identification module is used to identify potential fragile bone weak points in the muscle layer corresponding to the three-dimensional holographic model of bone and flesh based on the stress propagation pattern of the surface crack pattern. The cutting guide line generation module is used to generate dual-path cutting guide lines based on the three-dimensional model of the skeleton and the weak points of the skeleton, including a muscle fiber-oriented cutting guide line for muscle tissue and a mechanically optimized cutting guide line for the bone structure. The dynamic compensation control module is used to dynamically compensate and control the collaborative operation of the dual-path cutting guide lines through three-dimensional spatial coordinates, and simultaneously present the bone and flesh holographic three-dimensional model and the dual-path cutting guide lines on the visualization interface.

[0013] The beneficial effects of this invention are: This invention captures surface crack patterns of frozen bone-in meat through multimodal image acquisition. Combined with topological network model analysis and 3D holographic reconstruction, it achieves precise restoration of the 3D contours of bones encased in fat and muscle. It accurately marks joint positions, medullary cavity orientation, and areas of abnormal bone density, overcoming the limitations of traditional techniques in identifying internal structures. By analyzing the stress propagation patterns of surface cracks, potential bone weak points can be accurately identified, providing crucial information for cutting path planning. The generated dual-path cutting guide line considers both the muscle fiber characteristics of muscle tissue and the optimization needs of bone mechanics. Combined with a dynamic compensation mechanism for 3D spatial coordinates, it ensures the synergy and precision of the cutting process, reducing bone fragmentation and muscle tissue damage. Simultaneously, the synchronized visualization interface enhances the intuitiveness and controllability of the operation, comprehensively improving the accuracy, efficiency, and product quality of bone-in meat segmentation, making it suitable for practical needs in automated meat processing and other scenarios. Attached Figure Description

[0014] Figure 1 A flowchart of an image recognition-based intelligent recognition method for three-dimensional contours of bone-in meat provided in an embodiment of the present invention; Figure 2 A flowchart of multimodal image acquisition provided in an embodiment of the present invention; Figure 3 This is a flowchart for analyzing the surface crack pattern provided in an embodiment of the present invention; Figure 4 A flowchart for reconstructing a three-dimensional holographic model of bone-in flesh and muscle encased in fat and muscle, as provided in an embodiment of the present invention; Figure 5 A flowchart for identifying potential fragile bone weak points in the muscle layer corresponding to a three-dimensional holographic model of bone and flesh, provided in an embodiment of the present invention; Figure 6 This is a flowchart for generating dual-path cutting guide lines provided in an embodiment of the present invention; Figure 7 A flowchart illustrating the coordinated operation of dual-path cutting guide lines through dynamic compensation of three-dimensional spatial coordinates, provided in an embodiment of the present invention. Figure 8 The diagram shows the structure of the intelligent image recognition system for three-dimensional contour recognition of bone-in meat, as provided in an embodiment of the present invention. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0016] Figure 1 A flowchart of the intelligent recognition method for three-dimensional contours of bone-in meat based on image recognition provided in this embodiment of the invention is shown below. Figure 1 As shown, the method includes: S100 performs multimodal image acquisition on frozen bone-in meat to capture surface crack patterns formed by tissue shrinkage. Gradient freezing treatment causes uneven contraction of different tissues such as muscle, fat, and bone in bone-in meat due to differences in their coefficients of thermal expansion and contraction by gradually lowering the temperature. This contraction forms specific crack patterns on the surface. The fibrous structure of muscle tissue causes the cracks to extend along the fiber direction, while the brittleness of the fat layer may lead to more branching cracks. The junction between the bone and the surrounding soft tissue will form a unique crack intersection pattern due to the difference in rigidity. The distribution and direction of these cracks are implicitly related to the three-dimensional contour of the internal bone.

[0017] Based on this, multimodal image acquisition achieves precise capture of surface crack patterns through multispectral imaging technology. Multispectral imaging covers visible light and multiple infrared and ultraviolet bands. Different tissues (such as muscle, fat, and crack areas) have significant differences in reflectivity to different wavelengths of light. By scanning these reflectance spectral features to generate high-dimensional feature vectors, it is possible to effectively distinguish real cracks from surface stains, textures, and other interfering information, thereby more accurately reconstructing the three-dimensional geometric morphology of cracks and clearly presenting the depth, width, and spatial distribution of cracks.

[0018] Simultaneously, the dynamic formation sequence of surface crack patterns during freezing is recorded. The generated time-series image sequence can completely capture the entire process of crack initiation and propagation, including the appearance time of the initial microcracks and the changes in the propagation path at different stages. This dynamic information contains the mechanical mechanism of crack formation. At a certain temperature node during gradient cooling, the crack may accelerate its extension due to the sudden release of tissue shrinkage stress. This dynamic characteristic provides a direct basis for subsequent analysis of crack propagation speed and bifurcation law.

[0019] In actual meat processing scenarios, such as when processing frozen beef, lamb and other bone-in meats, gradient freezing can create cracks related to the orientation of the internal skeleton at different cooling stages. Multispectral imaging can clearly distinguish between cracks between muscle fibers and cracks in the fat layer, while time-series images can record the changes of these cracks with freezing time, laying the foundation for subsequent analysis of the relationship between crack patterns and bone structure.

[0020] Traditional single-modal image acquisition (such as using only visible light imaging) often struggles to distinguish real cracks from interfering factors such as surface textures and stains, and fails to capture the dynamic process of crack formation, resulting in insufficient information for subsequent analysis. In contrast, multimodal image acquisition, through the acquisition of high-dimensional feature vectors using multispectral technology, can significantly improve the accuracy and robustness of crack identification, providing high-quality raw data for subsequent construction of topological network models. The time-series image sequence, by recording the dynamic formation process of cracks, reveals the correlation between cracks and internal tissues (especially bones) under stress, making the surface crack pattern no longer isolated static information, but a dynamic signal that can reflect the internal structural characteristics.

[0021] Furthermore, the surface cracks induced by gradient freezing have an inherent mechanical relationship with the distribution of bones, muscles, and fat inside the bone-in meat. This relationship makes it possible to indirectly infer the internal three-dimensional contour through surface cracks, effectively solving the technical problem that bones in bone-in meat are difficult to identify directly because they are wrapped in soft tissue.

[0022] like Figure 2 As shown, the multimodal image acquisition specifically includes: S110, for bone-in meat in the gradient freezing process, uses multispectral imaging to scan the reflection spectral characteristics of the surface crack pattern, generates a high-dimensional feature vector, and reconstructs the three-dimensional geometric morphology of the surface crack pattern. S120 records the dynamic formation sequence of surface crack patterns during freezing, generates a time-series image sequence, and marks the crack initiation time nodes and propagation paths.

[0023] S200, the surface crack pattern is analyzed to determine the bifurcation direction, extension speed and spatial distribution law, and a topological network model of the surface crack pattern is constructed. The U-Net convolutional neural network is used to segment the main trunk and branches of cracks. By leveraging its ability to capture subtle features, it accurately distinguishes the main trunk (which often relates to the main direction of the internal skeleton) and the branches (which reflect local tissue differences, corresponding to the junction of muscle and bone or the distribution of fat layer) in cracks. The extracted bifurcation angle features and the generated angle distribution histogram can quantify the pattern of crack bifurcation. For example, the branch angle of cracks near the skeleton is usually smaller because the skeleton restricts the direction of tissue contraction, while the branch angle of cracks in the fat layer may be more dispersed. These angle features provide morphological basis for subsequent inference of the distribution of internal tissues.

[0024] By calculating the crack propagation velocity field based on time-series image sequences and tracking the coordinate changes of the crack tip at different times, the dynamic propagation pattern of the crack during freezing can be captured. This dynamic information contains the differences in the mechanical properties of the internal tissues: in areas near the bone, the tissue contraction is constrained due to the rigidity of the bone, resulting in a slower and more stable crack propagation velocity; while in areas with loose fat or muscle fibers, the tissue contraction is more free, resulting in a faster and more fluctuating crack propagation velocity. These velocity differences provide quantitative clues for analyzing the elasticity, brittleness, and other properties of the internal tissues.

[0025] A topological network model of surface crack patterns is constructed, integrating nodes such as crack bifurcation points and endpoints through connection relationships. The resulting network structure has an implicit correspondence with the spatial connection patterns of internal skeletons and soft tissues. Regions with high node degree (i.e., multiple bifurcation intersections) often correspond to the attachment points of bones and muscles, as the tissue in these areas is prone to multidirectional cracks due to complex stress. Spatial density heatmaps can intuitively show the dense areas of cracks. These areas usually correspond to the bends or joints of internal skeletons, as the tissue morphology changes greatly in these areas, and uneven contraction leads to crack concentration.

[0026] When processing bone-in meat containing long bones, the main trunks of surface cracks mostly extend along the length of the bone, with few branches and concentrated angles, and the extension speed is stable, showing obvious linear characteristics in the topological network. However, when processing bone-in meat containing joints, the surface cracks around the joints form dense branches with a high degree of node distribution, and the extension speed varies significantly at the joint edge. The spatial density heatmap of the topological network shows a high-intensity signal in the region corresponding to the joint. These analytical results can accurately reflect the structural characteristics of the internal skeleton.

[0027] This step ensures the objectivity of crack feature extraction through precise segmentation of U-Net, while the calculation of the extended velocity field introduces dynamic information in the time dimension, upgrading crack analysis from snapshot-based to process-based. The topological network model integrates scattered crack features into structured data associated with the internal structure, significantly improving the reliability and efficiency of inferring the internal structure from surface features.

[0028] like Figure 3 As shown, the analysis of the surface crack pattern specifically includes: S210, U-Net convolutional neural network is used to segment the main trunk and branches of the surface crack pattern, extract the bifurcation angle features of the surface crack pattern and generate an angle distribution histogram; S220, Based on the time-series image sequence, calculate the propagation velocity field of the surface crack pattern: ; in, Indicates the crack propagation rate. This indicates the displacement of the crack tip. For time intervals, for The coordinates of the crack tip at any given moment. For the first Frame timestamp; S230, Based on the node coordinates of the surface crack pattern, construct a topological network model including vertex connection relationships, and output a heat map of node degree distribution and spatial density.

[0029] S300, based on the mapping relationship between the constructed topological network model and the three-dimensional contour of the skeleton, reconstructs a three-dimensional holographic model of bone and flesh wrapped in fat and muscle, and marks the joint position, medullary cavity direction and abnormal bone density areas. Non-rigid registration of the node coordinates of the topological network model with the pre-stored skeletal CT template is not a simple coordinate matching, but rather a flexible spatial mapping function constructed through the thin plate spline algorithm. This function can adapt to the skeletal morphology variations caused by differences in growth and development among different bone-bearing individuals. For example, bone-bearing individuals of different ages or in different parts of the body have differences in the degree of bone curvature and thickness distribution. Non-rigid registration allows for local fine-tuning between the topological network model and the template, ensuring the accuracy of the mapping.

[0030] After the topological network model is transformed into a three-dimensional skeletal point cloud based on the mapping function, the orientation of the medullary cavity is reconstructed by combining depth sensing data. The depth sensing data can capture the subtle depressions and bending features of the internal cavity of the bone, allowing the medullary cavity structure, which is originally difficult to observe directly, to be presented in a three-dimensional form. For example, the medullary cavity in the middle of the bone may have a spiral orientation due to the need for nutrient delivery, and the depth data can accurately restore this shape.

[0031] Meanwhile, by analyzing the reflection characteristics of different wavelengths, multispectral imaging data can clearly separate fat and muscle regions. Fat tissue has a high reflectivity to light in specific wavelength bands, while muscle has a lower reflectivity due to its higher water content. Based on this, the generated soft tissue layer can realistically wrap around the bone, so that the three-dimensional holographic model not only includes hard tissue, but also reflects the distribution range and thickness of different soft tissues. For example, the muscle layer is thicker in the attachment area at both ends of the bone, while the fat layer may accumulate in the gaps between bones.

[0032] This step extracts the extreme points of bone curvature and matches them with anatomical atlases to determine joint locations. These extreme points of curvature often correspond to the turning points of joints. These locations are key nodes in bone movement, and accurate labeling can provide a basis for subsequent cutting to avoid damaging the joint structure. The standard deviation of local voxel gray values ​​is calculated to identify areas of abnormal bone density. The dispersion of gray value distribution can identify porous or sclerotic areas in the bone. These areas differ from normal bone in mechanical properties, and labeling them can guide the cutting process to avoid or handle them with caution.

[0033] When processing bone and muscle containing complex joint structures, it can accurately reconstruct the gaps and connections of the joint cavity and clearly mark the position of the joint in three-dimensional space; when processing bone and muscle with local bone density abnormalities, it can accurately identify the range and distribution of these abnormal areas, so that the entire three-dimensional holographic model includes both macroscopic bone and soft tissue morphology and microscopic structural abnormality information.

[0034] This step, through the fusion of topological network models and multimodal data, constructs a complete 3D holographic model containing bones, fat, and muscles without complex invasive detection. The labeled joint locations, medullary cavity orientation, and areas of abnormal bone density provide precise anatomical references for subsequent cutting guidance, ensuring that the cutting process avoids critical structures while adapting to the mechanical properties of the bone. Simultaneously, the stereoscopic presentation of the 3D holographic model allows operators to intuitively grasp the internal structural relationships of bone and flesh, solving the problems of fragmented information and unclear spatial relationships in traditional 2D images. This lays a high-precision, comprehensive structural cognitive foundation for the entire recognition and processing process.

[0035] like Figure 4 As shown, the reconstruction of the bone-and-flesh three-dimensional holographic model encased in fat and muscle specifically includes: S310, Non-rigid registration is performed between the node coordinates of the topological network model and the pre-stored skeletal CT template, and the spatial mapping function is calculated using the thin plate spline algorithm; S320, based on a spatial mapping function, maps a topological network model into a three-dimensional skeletal point cloud, and reconstructs the orientation of the medullary cavity by combining depth sensing data; S330, based on multispectral imaging data, segments the fat / muscle region, generates a soft tissue layer that wraps around the bone, and forms a three-dimensional whole-system model with bone and muscle. In the aforementioned three-dimensional holographic model of flesh and bone: Extract the extreme points of bone curvature, match them with anatomical atlases to determine the coordinates of the joint space, and identify and label the joint positions; Calculate the standard deviation of local voxel gray values, mark areas exceeding the threshold range as abnormal areas, and identify areas with abnormal bone density. ; in, This represents the standard deviation of local voxel gray levels. The voxel grayscale value is obtained from a pre-stored skeletal CT template. For the number of voxels in a local region, This represents the average gray value of a local area. ; Abnormal area determination criteria: ; in, The standard deviation threshold, The osteoporosis threshold, This is the threshold for osteosclerosis.

[0036] S400 identifies potential weak points in the muscle layer that correspond to the three-dimensional holographic model of bone and flesh based on the stress propagation pattern of the surface crack pattern. Biomechanical parameters are loaded onto a 3D holographic model of bone and flesh. These parameters cover the elastic modulus, Poisson's ratio, and shear modulus of different tissues such as bone, muscle, and fat. They together form the basis for simulating the real mechanical environment. The differences in the mechanical properties of different tissues will directly affect the stress transmission path and distribution state in the model. For example, the high elastic modulus of bone makes it the main carrier of stress transmission, while the toughness of muscle will buffer some stress. Loading these parameters can make the subsequent stress simulation more in line with the physical state of bone and flesh under actual stress.

[0037] Based on this, finite element stress propagation simulation is performed based on the propagation path of surface crack patterns. The propagation direction and speed of surface cracks are a direct reflection of stress action. Crack bifurcation points are often stress concentration nodes, while the areas where cracks extend rapidly correspond to the directions of continuous stress action. By discretizing the model into numerous elements, the finite element simulation calculates the deformation and stress of each element under stress, thereby tracing the entire process of stress transmission from surface cracks to the internal skeleton and clearly presenting the areas of stress accumulation and diffusion trends within the skeleton.

[0038] By identifying the overlap between areas where stress exceeds a certain threshold and areas with high bone porosity, potential weak points in fragile bones can be identified. High bone porosity means that the internal structure of the bone is loose and its mechanical strength is low, while high-stress areas are where the force is concentrated. When both conditions are met, the area is very prone to fracture when subjected to external forces such as cutting. This dual-condition screening can accurately locate those weak points hidden under the muscle layer that are difficult to detect by appearance alone, providing clear guidance for avoiding these risky areas in the subsequent cutting process.

[0039] When dealing with bone-in flesh containing multiple bone segments, surface cracks may exhibit multi-directional branching and extension. Stress propagation simulation can reveal stress concentrations at bone connection points, and if these points happen to have high bone porosity, they will be accurately marked as weak points. When dealing with bone-in flesh where the muscle layer is thick, resulting in a greater distance between surface cracks and internal bones, the loading of biomechanical parameters and finite element simulation can overcome the buffering effect of the muscle layer, accurately track the transmission of stress to the internal bones, and find those weak areas that are difficult to observe directly due to muscle coverage.

[0040] This step combines the stress propagation mode of surface cracks with the structural characteristics of the internal skeleton. With the help of finite element simulation and multi-parameter analysis, it can accurately identify potential weak points from two dimensions: mechanical mechanism and structural features. The identification of these weak points is not only based on surface phenomena, but also delves into the internal structure and stress nature of the skeleton, making the identification results highly reliable and targeted. This provides a key basis for the subsequent generation of scientific and reasonable cutting guide lines, effectively reducing the risk of bone fracture during cutting and improving the integrity and efficiency of bone-in meat processing.

[0041] like Figure 5 As shown, the identification of potential fragile bone weaknesses in the muscle layer corresponding to the three-dimensional holographic model of bone and muscle specifically includes: S410, loading biomechanical parameters onto a three-dimensional holographic model of flesh and bone; S420, finite element stress propagation simulation based on the propagation path of surface crack patterns; S430, identify the overlapping areas of regions with stress >150MPa and bone porosity >30% in the bone-and-flesh three-dimensional holographic model, and mark the overlapping areas as potential weak points in fragile bones.

[0042] S500 generates dual-path cutting guide lines based on a three-dimensional bone model and bone weak points, including a muscle fiber-oriented cutting guide line for muscle tissue and a mechanically optimized cutting guide line for bone structure. When generating guide lines for cutting along muscle fibers, extracting the texture direction features of the muscle layer in the 3D holographic model of bone and muscle is crucial. The orientation of muscle fibers directly affects the resistance during cutting and the integrity of muscle tissue. Cutting along the main direction of muscle fibers can minimize damage to muscle fibers and preserve the structural integrity of the muscle. This is because the connections between muscle fibers are relatively weak, and cutting along the direction can reduce mechanical damage during the cutting process and avoid muscle tissue fragmentation or fiber breakage.

[0043] When generating mechanically optimized cutting guide lines, setting obstacle areas to avoid weak points in the skeleton is to prevent the skeleton from fracturing due to external forces during the cutting process. These weak points have poor mechanical properties, and if the cutting path passes through them, it is very easy to cause the skeleton to break, affecting the shape of the product after cutting. Cutting along the bone seam interface takes advantage of the structural characteristics of the skeleton itself. The bone seam is the natural boundary between bone growth and connection. The mechanical strength of the skeleton is relatively low at this point. Cutting along this path can reduce cutting resistance, reduce energy consumption, and at the same time ensure the flatness of the bone cutting surface and reduce the generation of bone fragments. In addition, combined with areas of abnormal bone density, these areas are avoided during cutting to reduce the generation of bone fragments.

[0044] Establishing a spatial coordination constraint equation between the muscle fiber cutting guide line and the mechanically optimized cutting guide line is intended to coordinate the relative positions of the two paths in three-dimensional space, avoid cutting interference caused by path intersection or excessive distance, and ensure that the two guide lines perform their respective functions and cooperate with each other during the cutting process, so that muscle cutting and bone cutting can be carried out synchronously and efficiently, ensuring both the integrity of muscle tissue and the precision of bone cutting.

[0045] When processing bone-in meat with clear and consistent muscle texture, the fiber-guided cutting line can precisely extend along the fiber direction, ensuring that the cut muscle maintains a good block shape. When processing bone-in meat with many bone branches and complex bone suture distribution, the biomechanically optimized cutting line can cleverly avoid the weak points of each bone and plan the path naturally along the bone suture, ensuring that the bone shape is intact and there is no fragmentation after cutting.

[0046] Traditional cutting methods often plan the path from a single perspective. Focusing only on muscle cutting may result in irregular bone cuts, while considering only bone structure may damage muscle tissue. This step, however, utilizes the coordinated operation of dual-path guide lines to ensure both the integrity of muscle tissue and the mechanical rationality of bone cutting, reducing ineffective operations and material waste during the cutting process. Simultaneously, the establishment of spatial coordination constraint equations allows the two paths to form an organic whole in three-dimensional space, avoiding path conflicts and improving the smoothness and precision of the cutting. This significantly enhances the appearance, texture, and utilization rate of the cut bone-in meat.

[0047] like Figure 6 As shown, the generation of the dual-path cutting guide line specifically includes: S510, Extract the texture direction features of the muscle layer in the bone-and-flesh three-dimensional holographic model, and generate the cutting guide line along the main direction of the muscle fibers; S520, On the bone-and-flesh three-dimensional holographic model, an obstacle area is set to avoid the weak points of the bone, and the mechanically optimized cutting guide line is generated along the bone suture interface; S530, establish the spatial collaborative constraint equation between the suture fiber cutting guide line and the mechanically optimized cutting guide line, and optimize the relative position of the path.

[0048] Extract the feature vector of the Hessian matrix of the muscle layer, and generate a cubic B-spline curve with continuous curvature along the main direction to ensure that the angle between the path and the muscle fiber is ≤15°. The improved A* algorithm is applied to the skeletal model, with weak points set as obstacles (penalty coefficient ≥ 1000), and the bone suture interface (width > 0.2 mm) is preferentially selected as path nodes; Establish a dual-path spatial cooperative equation: set a minimum spacing constraint (≥5mm), and optimize the relative positions of the paths using the Lagrange multiplier method.

[0049] For the muscle layer cutting path, extracting the Hessian matrix feature vector can accurately capture the direction trend of muscle fibers. The cubic B-spline curve generated based on this ensures the continuity of the path curvature, avoiding tearing of muscle tissue due to abrupt path changes. Setting the angle between the path and muscle fibers to ≤15° is because muscle fibers have a significant directionality. A smaller angle can maximize the conformity to the natural distribution of muscle fibers, reducing lateral severing of muscle fibers during cutting, thereby maintaining the integrity of muscle tissue and avoiding muscle fragmentation or fiber breakage after cutting, ensuring better morphology of the cut muscle. This angular constraint allows the muscle cutting path to organically fit the tissue texture, reducing the impact of mechanical damage on muscle quality.

[0050] In the path planning of the skeletal model, the improved A* algorithm sets the penalty coefficient for weak point regions to ≥1000. Essentially, by significantly increasing the cost of the path passing through weak points, the algorithm is forced to actively avoid these areas with poor mechanical properties during planning, preventing bone fragmentation at weak points due to external forces during cutting and ensuring the integrity of the skeletal structure. Prioritizing bone suture interfaces with a width >0.2mm as path nodes is because bone sutures are natural weak connecting lines in the skeleton; sufficient width means clear structural boundaries and lower mechanical strength. Cutting along this path reduces resistance, resulting in a smoother bone cut surface and less bone fragmentation. If the bone suture width is too narrow, it is not only difficult to accurately identify, but it is also easy to deviate from the preset path during cutting, leading to bone morphological damage.

[0051] For spatial coordination of dual paths, setting a minimum spacing of ≥5mm is to avoid cutting interference caused by the two paths being too close in three-dimensional space. This ensures that cutting operations targeting muscle tissue and bone structures can be performed independently, reserving sufficient space for tool movement and preventing a decrease in cutting accuracy due to path intersections or overlaps. Based on this, the Lagrange multiplier method optimizes the relative positions of the paths. While satisfying the spacing constraints, it makes the two paths more closely conform to the overall distribution characteristics of muscles and bones, avoiding path redundancy caused by overemphasizing spacing and improving cutting efficiency.

[0052] The S600 uses dynamic compensation of three-dimensional spatial coordinates to control the coordinated operation of dual-path cutting guide lines, and simultaneously presents the bone and flesh holographic three-dimensional model and dual-path cutting guide lines on the visualization interface.

[0053] The dynamic compensation vector based on pose deviation is essentially a correction parameter that can be directly applied to the guide line by using spatial coordinate transformation algorithms (such as Euler angle transformation and homogeneous coordinate transformation). For example, when the tool deviates along the X-axis, the compensation vector will adjust the X coordinate of the guide line accordingly to ensure that the actual movement trajectory of the tool always fits the compensated guide line path. This real-time correction mechanism effectively offsets the accuracy loss caused by external interference.

[0054] The visual interface overlays a holographic 3D model of the bone and flesh, breaking the limitations of traditional cutting that relies on 2D drawings or experience. Operators can intuitively observe the skeletal structure, joint positions, weak points, and their relative spatial relationship with the cutting tool within the bone and flesh, avoiding miscuts caused by opaque information. The interface simultaneously renders the execution status of the dual-path cutting guide line, using different colors to distinguish between completed and uncut segments, and updates the tool's current position projection on the guide line in real time. This allows operators to clearly grasp the cutting progress. When the deviation between the guide line and the actual trajectory exceeds the limit, the interface provides an immediate warning, offering a basis for manual intervention.

[0055] like Figure 7 As shown, the coordinated operation of the dual-path cutting guide lines controlled by dynamic compensation of three-dimensional spatial coordinates specifically includes: S610, Real-time acquisition of the pose deviation of the cutting tool relative to the bone-in flesh three-dimensional holographic model; S620, calculate the dynamic compensation vector of the dual-path cutting guide line based on the pose deviation; S630, the three-dimensional holographic model with bone and flesh is overlaid on the visualization interface; S640, the execution status of the dual-path cutting guide line is simultaneously rendered on the visualization interface.

[0056] Figure 8 The structural block diagram of the intelligent recognition system for three-dimensional contours of bone-in meat based on image recognition provided in the embodiments of the present invention is as follows: Figure 8 As shown, the system includes: The multimodal image acquisition module 100 is used to acquire multimodal images of frozen bone-in meat and capture surface crack patterns formed by tissue shrinkage. The crack pattern analysis module 200 is used to analyze the surface crack pattern, analyze the bifurcation direction, extension speed and spatial distribution law, and construct the topological network model of the surface crack pattern. The 3D holographic model reconstruction module 300 is used to reconstruct a 3D holographic model of bone and flesh wrapped in fat and muscle based on the mapping relationship model between the constructed topological network model and the 3D contour of the skeleton, and to annotate the joint position, the direction of the medullary cavity and the abnormal bone density areas. The fragile bone weak point identification module 400 is used to identify potential fragile bone weak points in the muscle layer corresponding to the three-dimensional holographic model of bone and flesh based on the stress propagation pattern of the surface crack pattern. The cutting guide line generation module 500 is used to generate dual-path cutting guide lines based on the three-dimensional model of the skeleton and the weak points of the skeleton, including a muscle fiber-oriented cutting guide line for muscle tissue and a mechanically optimized cutting guide line for the bone structure. The dynamic compensation control module 600 is used to control the coordinated operation of the dual-path cutting guide lines through dynamic compensation of three-dimensional spatial coordinates, and to simultaneously present the bone and flesh holographic three-dimensional model and the dual-path cutting guide lines on the visualization interface.

[0057] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0058] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

[0059] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent recognition of three-dimensional contours of bone-in meat based on image recognition, characterized in that, The method includes: Multimodal image acquisition was performed on frozen bone-in meat to capture surface crack patterns formed by tissue shrinkage; The surface crack pattern is analyzed to determine its bifurcation direction, extension speed, and spatial distribution, and a topological network model of the surface crack pattern is constructed. Based on the mapping relationship between the constructed topological network model and the three-dimensional contour of the skeleton, a three-dimensional holographic model of bone and flesh wrapped in fat and muscle is reconstructed, and the joint positions, medullary cavity orientation and abnormal bone density areas are marked. Based on the stress propagation pattern of the surface crack pattern, potential weak points of fragile bones in the muscle layer corresponding to the three-dimensional holographic model of bone and flesh are identified. Based on the three-dimensional model of the skeleton and the weak points of the skeleton, a dual-path cutting guide line is generated, including a muscle fiber-oriented cutting guide line for muscle tissue and a mechanically optimized cutting guide line for the bone structure. The collaborative operation of the dual-path cutting guide lines is controlled by dynamic compensation of three-dimensional spatial coordinates, and the bone-and-flesh holographic three-dimensional model and the dual-path cutting guide lines are presented simultaneously on the visualization interface.

2. The method according to claim 1, characterized in that, The multimodal image acquisition specifically includes: For bone-in meat during gradient freezing, multispectral imaging is used to scan the reflectance spectral characteristics of surface crack patterns, generate high-dimensional feature vectors, and reconstruct the three-dimensional geometric morphology of surface crack patterns. Record the dynamic formation sequence of surface crack patterns during freezing, generate a time-series image sequence, and mark the crack initiation time nodes and propagation paths.

3. The method according to claim 2, characterized in that, The analysis of the surface crack pattern specifically includes: The U-Net convolutional neural network is used to segment the main trunk and branches of the surface crack pattern, extract the bifurcation angle features of the surface crack pattern, and generate an angle distribution histogram. Based on the time-series image sequence, the propagation velocity field of the surface crack pattern is calculated: ； in, Indicates the crack propagation rate. This indicates the displacement of the crack tip. For time intervals, for Coordinates of the crack tip at any given moment. For the first Frame timestamp; Based on the node coordinates of the surface crack pattern, a topological network model including vertex connection relationships is constructed, and a node degree distribution and spatial density heatmap are output.

4. The method according to claim 3, characterized in that, The reconstructed three-dimensional holographic model of bone and flesh encased in fat and muscle specifically includes: The node coordinates of the topological network model are non-rigidly registered with the pre-stored skeletal CT template, and the spatial mapping function is calculated using the thin plate spline algorithm. Based on the spatial mapping function, the topological network model is mapped to a three-dimensional skeletal point cloud, and the orientation of the medullary cavity is reconstructed by combining depth sensing data. Based on multispectral imaging data, fat / muscle regions are segmented to generate a soft tissue layer that encloses the bone, forming a three-dimensional whole-system model of bone and muscle. In the aforementioned three-dimensional holographic model of flesh and bone: Extract the extreme points of bone curvature, match them with anatomical atlases to determine the coordinates of the joint space, and identify and label the joint positions; Calculate the standard deviation of local voxel gray values, mark areas exceeding the threshold range as abnormal areas, and identify areas with abnormal bone density. ； in, This represents the standard deviation of local voxel gray levels. The voxel grayscale value is obtained from a pre-stored skeletal CT template. For the number of voxels in a local region, This represents the average gray value of a local area. ; Abnormal area determination criteria: ； in, The standard deviation threshold, The osteoporosis threshold, This is the threshold for osteosclerosis.

5. The method according to claim 4, characterized in that, The identification of potential fragile bone weaknesses in the muscle layer corresponding to the three-dimensional holographic model of bone and flesh specifically includes: Biomechanical parameters were loaded onto a three-dimensional holographic model of meat with bone; Finite element stress propagation simulation based on the propagation path of surface crack patterns; Identify the overlapping regions of regions with stress >150MPa and bone porosity >30% in the bone-bone 3D holographic model, and mark the overlapping regions as potential weak points in fragile bones.

6. The method according to claim 5, characterized in that, The generation of the dual-path cutting guide line specifically includes: Extract the texture direction features of the muscle layer in the bone-in 3D holographic model, and generate the cutting guide line along the main direction of the muscle fibers; On the bone-and-flesh three-dimensional holographic model, an obstacle area is set to avoid the weak points of the bone, and the mechanically optimized cutting guide line is generated along the bone suture interface; Establish a spatial co-constraint equation between the suture fiber cutting guide line and the mechanically optimized cutting guide line, and optimize the relative position of the paths.

7. The method according to claim 6, characterized in that, The coordinated operation of the dual-path cutting guide lines controlled by dynamic compensation of three-dimensional spatial coordinates specifically includes: The pose deviation of the cutting tool relative to the bone-in meat three-dimensional holographic model is acquired in real time; The dynamic compensation vector of the dual-path cutting guide line is calculated based on the pose deviation. The three-dimensional holographic model of bone and flesh is overlaid on the visualization interface; The execution status of the dual-path cutting guide line is rendered synchronously on the visualization interface.

8. A three-dimensional contour intelligent recognition system for bone-in meat based on image recognition, characterized in that, The system includes: The multimodal image acquisition module is used to acquire multimodal images of frozen bone-in meat and capture surface crack patterns formed by tissue shrinkage. The crack pattern analysis module is used to analyze the surface crack pattern, analyze the bifurcation direction, extension speed and spatial distribution law, and construct the topological network model of the surface crack pattern. The 3D holographic model reconstruction module is used to reconstruct a 3D holographic model of bone and flesh wrapped in fat and muscle based on the mapping relationship between the constructed topological network model and the 3D contour of the skeleton, and to annotate the joint positions, the direction of the medullary cavity and the areas of abnormal bone density. The fragile bone weak point identification module is used to identify potential fragile bone weak points in the muscle layer corresponding to the three-dimensional holographic model of bone and flesh based on the stress propagation pattern of the surface crack pattern. The cutting guide line generation module is used to generate dual-path cutting guide lines based on the three-dimensional model of the skeleton and the weak points of the skeleton, including a muscle fiber-oriented cutting guide line for muscle tissue and a mechanically optimized cutting guide line for the bone structure. The dynamic compensation control module is used to dynamically compensate and control the collaborative operation of the dual-path cutting guide lines through three-dimensional spatial coordinates, and simultaneously present the bone and flesh holographic three-dimensional model and the dual-path cutting guide lines on the visualization interface.

Images