Intelligent hierarchical simplification method and device for three-dimensional model

Abstract

This invention relates to the field of 3D model processing technology, specifically to an intelligent hierarchical simplification method and device for 3D models. The method involves acquiring 3D device model data and converting it into a mesh model; performing region growth and segmentation based on patch normals and curvature to obtain the original component regions; employing standard part matching, rule base matching, and deep learning recognition for three-level semantic annotation and importance grading; generating multi-precision mesh models based on the importance level using a feature-weighted point cloud non-uniform simplification algorithm; and finally assembling 3D device model data with different simplification precipitates based on the region connectivity graph. This application achieves intelligent and differentiated hierarchical simplification of 3D models, balancing core feature preservation with data lightweighting, adapting to the needs of multiple application scenarios, and improving the efficiency and flexibility of model processing.

Description

Technical Field

[0001] This invention relates to the field of 3D model processing technology, specifically to an intelligent hierarchical simplification method and device for 3D models. Background Technology

[0002] With the widespread application of 3D modeling technology in industrial production, digital twins, virtual reality, and other fields, the accuracy requirements of 3D models vary significantly across different application scenarios. For example, animation rendering and high-precision simulation require high-precision models that retain a large amount of detail to ensure the realism of the effects; digital twin systems need to strike a balance between accuracy and computational efficiency; while web page display and quick preview require lightweight, low-precision models to achieve fast loading and transmission.

[0003] Existing 3D model simplification techniques often employ a uniform simplification strategy, simplifying all components to the same degree without considering the differences in component functional importance. This approach easily leads to the loss of key features of core functional components, affecting the model's usability. Simultaneously, redundant data from non-core components is not effectively streamlined, resulting in excessively large model data volumes and reduced processing and transmission efficiency. Furthermore, existing methods lack a systematic component semantic recognition and hierarchical mechanism; the simplification process relies on manual intervention, exhibiting low levels of intelligence and difficulty adapting to the batch processing needs of complex industrial scenarios, and also failing to achieve flexible adaptation of simplification accuracy. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a method and device for intelligent hierarchical simplification of three-dimensional models, so as to overcome the problems existing in the current prior art.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] On the one hand, this application provides an intelligent hierarchical simplification method for three-dimensional models, including: Step S1: Obtain 3D device model data and convert the 3D device model data into a mesh model; Step S2: Obtain the normal and curvature of each facet in the mesh model, and divide the mesh model into multiple original component regions based on the normal and curvature of each facet using a region growing mesh segmentation algorithm. Step S3: Calculate the first matching degree between the original component region and each model in the standard parts template library, and assign the semantic label and importance level of the model in the standard parts template library with the highest first matching degree and higher than the first preset threshold to the original component region; Step S4: Perform multi-level feature recognition on the original component region where the first matching degree with each model in the standard parts template library is lower than the first preset threshold. Calculate the second matching degree with each rule in the rule-based logistics equipment knowledge base based on the multi-level feature recognition results of the original component region. Assign the semantic label and importance level of the rule in the logistics equipment knowledge base with the highest second matching degree and higher than the second preset threshold to the original component region. Step S5: Convert the original component regions with a second matching degree lower than the second preset threshold into a multi-view two-dimensional image set, identify the multi-view two-dimensional image set through a pre-trained deep convolutional neural network, and assign semantic labels and importance levels to the original component regions corresponding to the multi-view two-dimensional image set according to the identification results. Mark the original component regions that the deep convolutional neural network fails to identify as unknown components and assign a preset importance level. Step S6: Based on the importance level of each original component region, the original component regions are simplified using a feature-weighted point cloud non-uniform simplification algorithm under different simplification accuracies to generate mesh models under different simplification accuracies. Step S7: Assemble the mesh models with different simplification accuracies to generate 3D device model data with different simplification accuracies.

[0007] Furthermore, in the method described above, step S1 includes: Acquire 3D device model data and perform automated format parsing and data cleaning on the 3D device model data; wherein, the 3D device model data includes: CAD design files, 3D scan point clouds, and mesh models; The processed data is converted into a grid model.

[0008] Furthermore, in the method described above, step S2 includes: Step S21: Calculate the normal and curvature of each facet in the mesh model; Step S22: Select a seed facet in the mesh model according to the curvature of the facet; Step S23: Calculate the dihedral angle between the seed facet and its adjacent facets; Step S24: If the dihedral angle is less than the first preset angle, and the angle between the normal direction of the seed facet and the normal direction of the corresponding adjacent facet is less than the second preset angle, and the difference between the curvature of the seed facet and the curvature of the corresponding adjacent facet is less than the third threshold, then the corresponding adjacent facet is assigned to the region of the seed facet, and the average normal and average curvature of the region of the seed facet are calculated. Steps S23 and S24 are executed again until all adjacent facets of the region of the seed facet cannot be assigned to the region of the seed facet. Step S25: Repeat steps S22 to S24 until each facet in the mesh model is assigned to its corresponding region; Step S26: Calculate the bounding boxes of all regions. For any two regions, if their bounding boxes intersect or the distance between their bounding boxes is less than a fourth preset threshold, calculate the nearest point pair distance between the two region meshes. If the nearest point pair distance is less than a fifth preset threshold, create an edge in the region connection graph to connect the two regions, thus completing the division of the mesh model into multiple original component regions.

[0009] Furthermore, in the method described above, step S3 includes: Step S31: For each point on the original component area, find the nearest corresponding point on the target model in the standard parts template library; Step S32: Calculate the rotation and translation transformations that minimize the overall distance between all corresponding points, and perform corresponding rotation and translation transformations on the original component region based on the calculation results to complete the registration of the original component region; Step S33: Obtain valid point pairs between the registered original component region and the corresponding points on the target model in the standard part template library, and obtain the distance between each valid point pair; Step S34: Calculate the root mean square error of the distance between all the effective point pairs, and calculate the first matching degree between the original component region and the target model in the standard part template library based on the root mean square error; Step S35: Select a target model again in the standard parts template library, and execute steps S31 to S34 again to calculate the first matching degree between the original part region and the reselected target model until the first matching degree between the original part region and all models in the standard parts template library is obtained. Step S36: Assign the semantic labels and importance levels of the models in the standard parts template library with the highest first matching degree and higher than the first preset threshold to the original part region; Step S37: Reselect the original component area and repeat steps S31 to S36 until all the original component areas are matched.

[0010] Furthermore, in the method described above, step S4 includes: Step S41: Perform multi-level feature recognition on the original component regions where the first matching degree with each model in the standard component template library is lower than the first preset threshold, and obtain the macroscopic morphological features, microscopic geometric features, and topological set features of the original component regions; wherein, the macroscopic morphological features include: axial bounding box size ratio, compactness, and principal component ratio of moment of inertia; the microscopic geometric features include: mean curvature, standard deviation of curvature, and boundary edge ratio; and the topological set features include: motion attributes and number of holes. Step S42: Select one of the original component regions; Step S43: Calculate the confidence level between the macroscopic morphological features, microscopic geometric features, and topological set features of the selected original component region and the feature details corresponding to the target rule in the rule-based logistics equipment knowledge base; Step S44: Calculate the second matching degree between the original component region and the target rule based on the confidence level of all the described feature details; Step S45: Select a target rule again in the rule-based logistics equipment knowledge base, and repeat steps S43 to S44 until the second matching degree between the original component region and all rules in the rule-based logistics equipment knowledge base is obtained. Assign the semantic label and importance level of the rule in the logistics equipment knowledge base with the highest second matching degree and higher than the second preset threshold to the original component region. Step S46: Select a new original component region and repeat steps S43 to S45 until all original component regions are matched. Step S47: Record the matching process of all the original component regions, and update and adjust the rule-based logistics equipment knowledge base according to the human intervention rate and matching failure characteristics in the matching process, as well as the second matching degree of the corresponding original component regions.

[0011] Furthermore, in the method described above, step S5 includes: Step S51: Select one of the original component regions from all the original component regions where the second matching degree is lower than the second preset threshold; Step S52: Place the selected original component region at the center of a virtual sphere; Step S53: Render a two-dimensional grayscale image or depth image of the original component area from multiple predefined perspectives on the surface of the virtual sphere; Step S54: Input all the two-dimensional grayscale images or the depth images into a pre-trained deep convolutional neural network to identify the original component regions, and assign semantic labels and importance levels to the original component regions according to the identification results; wherein, the deep convolutional neural network is trained with data from a historical project component model library that has been accurately labeled with semantic tags. Step S55: Select a new original component region and repeat steps S52 to S54 until all original component regions have been identified. Step S56: The original component regions that the deep convolutional neural network failed to recognize are labeled as unknown components and a preset importance level is assigned.

[0012] Furthermore, in the method described above, step S5 further includes: Obtain semantic labels and importance level labels from the user for the original component regions that the deep convolutional neural network failed to recognize; The historical project component model library is updated based on the semantic tags and importance levels annotated by the user and the corresponding original component regions, and the deep convolutional neural network is retrained.

[0013] Furthermore, in the method described above, the importance levels include: L0, L1, and L2 levels, and step S6 includes: All the original component regions are simplified with high precision using a feature-weighted point cloud non-uniform simplification algorithm to generate corresponding high precision point cloud data. Adjust the target sampling density of the non-uniform point cloud simplification algorithm to perform intermediate-precision simplification on the original component regions of L1 and L2 levels, and generate corresponding intermediate-precision point cloud data. The target sampling density of the non-uniform point cloud simplification algorithm is adjusted again to perform low-precision simplification on the original component region at level L2, generating corresponding low-precision point cloud data. The high-precision point cloud data, the medium-precision point cloud data, and the low-precision point cloud data are reconstructed using a grid reconstruction algorithm to generate a high-precision grid model, a medium-precision grid model, and a low-precision grid model.

[0014] Furthermore, in the method described above, step S7 includes: For advanced simplification precision, advanced precision mesh models corresponding to the original component regions at levels L0, L1, and L2 are obtained. Based on the topological relationships and component spatial transformation information recorded in the region connection diagram of the original component regions, the advanced precision mesh models are assembled to generate advanced simplification precision 3D device model data. For intermediate simplification accuracy, obtain the high-precision mesh model corresponding to the original component region at level L0, and the intermediate-precision mesh models corresponding to the original component regions at levels L1 and L2. Based on the topological relationships and component spatial transformation information recorded in the region connection diagram of the original component region, assemble the high-precision mesh model and the intermediate-precision mesh model to generate 3D device model data with intermediate simplification accuracy. For low-level simplified precision, a high-precision mesh model corresponding to the original component region at level L0, a medium-precision mesh model corresponding to the original component region at level L1, and a low-precision mesh model corresponding to the original component region at level L2 are obtained. Based on the topological relationships and component spatial transformation information recorded in the region connection diagram of the original component region, the high-precision mesh model, the medium-precision mesh model, and the low-precision mesh model are assembled to generate low-level simplified precision 3D device model data.

[0015] On the other hand, this application provides an intelligent hierarchical simplification device for three-dimensional models, including a processor and a memory, wherein the processor is connected to the memory: The processor is used to call and execute the program stored in the memory; The memory is used to store the program, which is used to execute at least any of the above-described intelligent hierarchical simplification methods for three-dimensional models.

[0016] The beneficial effects of this invention are as follows: High level of intelligence: Through automated data preprocessing, region segmentation, and three-level semantic recognition (standard part matching, rule base matching, and deep learning recognition), the end-to-end hierarchical simplification of the 3D model is achieved, greatly reducing manual intervention and improving processing efficiency; Differentiated simplification strategy: Differentiated simplification is performed based on the importance level of components. Core functional components retain the highest accuracy, while non-core components are deeply simplified, balancing the integrity of core features with the need for lightweight data. Strong scene adaptability: Generates high, medium and low precision models to adapt to different application scenarios such as animation rendering, digital twins, and web page display, thereby improving the practicality and flexibility of the models; Outstanding self-optimization capability: It supports the iterative update of the logistics equipment knowledge base and deep learning model, and continuously optimizes through accumulated matching data and user-annotated data to improve the accuracy of subsequent identification and simplification; Precise preservation of topological relationships: The topological relationships between components are recorded through a region connection diagram, ensuring that the assembled multi-precision model is consistent with the original model in terms of assembly relationships, thus guaranteeing the reliability and usability of the model. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present 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 only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart provided by an embodiment of the intelligent hierarchical simplification method for three-dimensional models according to the present invention; Figure 2 This is a schematic diagram of the structure of an embodiment of the intelligent hierarchical simplification device for three-dimensional models provided by the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be described in detail below. Obviously, the described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other implementation methods obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0020] Figure 1 This is a flowchart illustrating an embodiment of an intelligent hierarchical simplification method for three-dimensional models according to the present invention. Please refer to [link / reference]. Figure 1 This embodiment may include the following steps: Step S1: Obtain the 3D device model data and convert the 3D device model data into a mesh model; Step S2: Obtain the normal and curvature of each facet in the mesh model, and divide the mesh model into multiple original component regions based on the normal and curvature of each facet using a region growing mesh segmentation algorithm. Step S3: Calculate the first matching degree between the original component region and each model in the standard parts template library, and assign the semantic label and importance level of the model in the standard parts template library with the highest first matching degree and higher than the first preset threshold to the original component region. Step S4: Perform multi-level feature recognition on the original component regions where the first matching degree with each model in the standard parts template library is lower than the first preset threshold. Calculate the second matching degree with each rule in the rule-based logistics equipment knowledge base based on the multi-level feature recognition results of the original component regions. Assign the semantic label and importance level of the rule in the logistics equipment knowledge base with the highest second matching degree and higher than the second preset threshold to the original component regions. Step S5: Convert the original component regions with a second matching degree lower than the second preset threshold into a multi-view two-dimensional image set. Use a pre-trained deep convolutional neural network to identify the multi-view two-dimensional image set. Assign semantic labels and assign importance levels to the original component regions corresponding to the multi-view two-dimensional image set based on the identification results. Label the original component regions that the deep convolutional neural network fails to identify as unknown components and assign preset importance levels. Step S6: Based on the importance level of each original component region, simplify the original component region using a feature-weighted point cloud non-uniform simplification algorithm under different simplification accuracies to generate mesh models under different simplification accuracies. Step S7: Assemble the mesh models with different simplification accuracies to generate 3D device model data with different simplification accuracies.

[0021] It is understood that in this embodiment, three-dimensional device model data is acquired and converted into a mesh model. The original component region is obtained by region growth segmentation based on the surface normal and curvature. Three-level semantic annotation and importance classification are performed through standard part matching, rule base matching, and deep learning recognition. Based on the importance level, a feature-weighted point cloud non-uniform simplification algorithm is used to generate multi-precision mesh models. Finally, three-dimensional device model data with different simplification accuracies are assembled based on the region connection graph.

[0022] Preferably, step S1 includes: Acquire 3D device model data and perform automated format parsing and data cleaning on the 3D device model data; the 3D device model data includes: CAD design files, 3D scanned point clouds, and mesh models; The processed data is converted into a grid model.

[0023] It is understood that this embodiment will use automated format parsing and data cleaning to convert heterogeneous raw data from different sources, including but not limited to CAD design files, 3D scan point clouds and mesh models, into a unified mesh model, laying the data foundation for subsequent component segmentation, feature extraction and semantic recognition.

[0024] Preferably, step S2 includes: Step S21: Calculate the normal and curvature of each facet in the mesh model; Step S22: Select a seed facet in the mesh model based on the curvature of the facet; Step S23: Calculate the dihedral angle between the seed facet and its adjacent facets; Step S24: If the dihedral angle is less than the first preset angle, and the angle between the normal direction of the seed face and the normal direction of the corresponding adjacent face is less than the second preset angle, and the difference between the curvature of the seed face and the curvature of the corresponding adjacent face is less than the third threshold, then the corresponding adjacent face is assigned to the region of the seed face, and the average normal and average curvature of the region of the seed face are calculated. Steps S23 and S24 are executed again until all adjacent faces of the region of the seed face cannot be assigned to the region of the seed face. Step S25: Repeat steps S22 to S24 until each facet in the mesh model is assigned to its corresponding region; Step S26: Calculate the bounding boxes of all regions. For any two regions, if their bounding boxes intersect or the distance between their bounding boxes is less than the fourth preset threshold, calculate the distance between the nearest point pairs of the two region meshes. If the distance between the nearest point pairs is less than the fifth preset threshold, create an edge in the region connection graph to connect the two regions, thus completing the division of the mesh model into multiple original component regions.

[0025] Understandably, after importing the original mesh model, the first step should be to decompose the mesh into components. This application employs a region-growing mesh segmentation algorithm based on concavity and convexity. This algorithm adaptively selects growth seeds by calculating the curvature of facets and uses the dihedral angle features between adjacent facets as the segmentation boundary, thereby automatically decomposing the complex assembly mesh into sub-mesh regions that conform to the physical component division.

[0026] During the execution of the growth algorithm, each facet in the model is in one and only one of the following two states at any given time: "unassigned" or "assigned". "Assigned" means that a triangular facet has been successfully assigned to a specific, unique region by the algorithm. There is a clear and unchangeable membership relationship between the facet and this region. "Unassigned" means that a triangular facet has not yet been assigned to any region; it is in a "free" state and is an object to be processed by the algorithm.

[0027] Before performing region growth, the system first calculates two key geometric properties for each facet (or vertex) of the mesh model: facet normal and curvature. These properties are used to determine the orientation of the facet and the properties of each vertex, thereby determining the approximate shape of the model.

[0028] The normal to a face is a unit vector perpendicular to that face. For a triangular face with vertices v0, v1, and v2, first calculate the vectors e1 and e2 of the two edges: e1=v1-v0 e2=v2-v0 The normal direction is obtained by calculating the cross product n of these two vectors, where n = e. 1× e2, and finally normalize the resulting vector into a unit vector.

[0029] Curvature describes the degree of bending of a surface at a given point. In a model, regions with high curvature typically correspond to the edges, corners, or feature lines of the model, and are potential segmentation boundaries. Regions with low curvature typically correspond to the largest and most central geometric features in the model. For a vertex, its curvature can be calculated by the change in its normal relative to all its adjacent vertices.

[0030] Where, k i For vertex v i curvature, and These are the vertices v i and its neighboring vertices v j The unit normal, N(i) is the vertex v i The set of adjacent vertices.

[0031] After calculating the normals and curvature of the face, it is necessary to select the region where the algorithm will begin to "grow," namely the "seed face." Starting from the seed face, the algorithm traverses adjacent faces, assigning faces with similar geometric features to a single region, until it encounters a "boundary" with significant feature differences and stops growing.

[0032] Low-curvature regions are typically the largest and most central geometric features of a component. Starting growth from these low-curvature regions, the algorithm can quickly incorporate a large number of adjacent flat surfaces with similar normal directions, avoiding the generation of numerous meaningless, fragmented small regions. This results in a macroscopically coherent and physically plausible large region, consistent with our macroscopic understanding of a "component." Therefore, the surface with the smallest curvature among the unassigned surfaces is selected as the seed surface.

[0033] In areas with low curvature, small geometric noise regions, or in thin-walled or narrow gaps, to avoid meaningless small areas and loss of component integrity, a surface with a larger area and relatively smaller curvature is selected as the seed surface.

[0034] Starting with a seed facet, iteratively incorporate its neighboring faces into the current region.

[0035] The most crucial criterion for determining whether adjacent faces are included is the dihedral angle between them. The dihedral angle is the angle between the normals to the common edge of two adjacent faces, visually reflecting the concavity / convexity and smoothness relationship between the two faces. For two adjacent faces A and B that share an edge, their unit normal vectors are respectively and The formula for calculating the dihedral angle θ is:

[0036] in, for and The dot product.

[0037] A dihedral angle θ ≈ 180° indicates that the two faces are almost on the same plane and the connection is smooth. A dihedral angle θ < 180° indicates that the connection between the two faces is a convex edge. A dihedral angle θ > 180° indicates that the connection between the two faces is a concave edge.

[0038] The algorithm presets an angle threshold α. t For each adjacent unassigned face to be examined, check its shared edges with the faces in the current region. If it is determined that θ>α... t If θ < α, then it is considered a strong feature edge, i.e., a boundary of the region, and region growth stops in this direction. t If the condition is met, other criteria are further examined. If satisfied, the patch is included in the current region and used as a new growth point. The adjacent unassigned patches are then traversed again. This continues until strong feature edges form a closed region, completing the partitioning of a region.

[0039] In addition, to ensure that patches within a region have consistent geometric properties and avoid overgrowth, one or two additional criteria are typically introduced. Only patches that simultaneously satisfy non-concave edge connection and the following criteria are included.

[0040] Rule 1: Normal Consistency. The angle between the normal direction of the patch to be added and the seed patch (or the average normal of the region) must be less than a preset threshold. This ensures that the patches within the region are "smooth".

[0041] Rule 2: Curvature consistency. The difference between the curvature value of the patch to be added and the average curvature of the region must be less than a threshold. This ensures that patches within the region have a similar degree of curvature.

[0042] Following this rule, after each growth cycle, a new seed facet is selected to begin growth again. After several rounds of growth, each facet will be assigned to a region, completing the division of the model's regions. Each region is then treated as a raw component region.

[0043] After performing region growth, a region connection graph needs to be introduced to organize the original component region relationships.

[0044] A region connectivity graph is a mathematical model that represents the result of 3D mesh segmentation as a graph structure. Nodes represent continuous sub-mesh regions that have been segmented, and edges represent shared boundaries or spatial contact relationships between adjacent regions, thus intuitively describing the spatial adjacency and connection topology between components.

[0045] Calculate the bounding boxes of all segmented regions. For any two regions, if their bounding boxes intersect or the distance is less than a preset threshold, the system will further calculate the nearest point pair distance between the two region grids, which is the minimum Euclidean distance between all possible point pairs between the two point sets.

[0046]

[0047] Where, d min (A,B) represents the distance between the closest point pairs in region A and region B, where a is a point in region A and b is a point in region B.

[0048] If the distance between the nearest point pairs is less than a preset threshold, the two regions are considered to be adjacent, and an edge is created in the region adjacency graph to connect them.

[0049] Preferably, step S3 includes: Step S31: For each point in the original part area, find the nearest corresponding point on the target model in the standard part template library; Step S32: Calculate the rotation and translation transformations that minimize the overall distance between all corresponding points, and perform corresponding rotation and translation transformations on the original component region based on the calculation results to complete the registration of the original component region. Step S33: Obtain valid point pairs between the registered original part area and the corresponding points on the target model in the standard part template library, and obtain the distance between each valid point pair; Step S34: Calculate the root mean square error of the distance between all valid point pairs, and calculate the first matching degree between the original part area and the target model in the standard part template library based on the root mean square error. Step S35: Reselect the target model in the standard parts template library, and execute steps S31 to S34 again to calculate the first matching degree between the original part area and the reselected target model until the first matching degree between the original part area and all models in the standard parts template library is obtained. Step S36: Assign the semantic labels and importance levels of the models in the standard parts template library with the highest matching degree and higher than the first preset threshold to the original part region; Step S37: Reselect the original component area and repeat steps S31 to S36 until all original component areas are matched.

[0050] Understandably, using the original component region obtained through region growing as the input unit, a rapid iterative nearest-point matching process is performed between the original component region and a predefined standard part template library using 3D point cloud or model automatic alignment techniques. First, for each point in the original component region, the nearest corresponding point in the standard part template library model is found. Then, the rotation and translation transformations that minimize the overall distance between all corresponding points are calculated, and finally, this transformation is applied. This process is repeated until the transformation parameters stabilize, ultimately achieving high-precision spatial registration between the two models. After the output transformation parameters stabilize, the matching degree between the original component region and the standard part template library is calculated.

[0051] Suppose there are N valid point pairs after registration, and the distance between each point pair is d. i (Euclidean distance between the source point and the corresponding target point), calculate the root mean square error E R :

[0052] Let L be the diagonal length of the model's bounding box (or the model's feature size). The formula for calculating the matching score S is:

[0053] Where k is a preset proportional coefficient (taken from 0.05 to 0.15), used to control the tolerance of error.

[0054] If the matching degree S reaches the threshold, the original component region is considered a standard component. The name of this standard component is used as the semantic label for the original component region. If multiple candidate standard components with matching degrees reaching the threshold exist, the candidate standard component with the highest matching degree calculation result is selected first.

[0055] Preferably, step S4 includes: Step S41: Perform multi-level feature recognition on the original component regions whose first matching degree with each model in the standard parts template library is lower than the first preset threshold to obtain the macroscopic morphological features, microscopic geometric features, and topological set features of the original component regions; wherein, the macroscopic morphological features include: axial bounding box size ratio, compactness, and principal component ratio of moment of inertia; the microscopic geometric features include: mean curvature, standard deviation of curvature, and boundary edge ratio; and the topological set features include: motion attributes and number of holes. Step S42: Select an original component area; Step S43: Calculate the confidence level between the macroscopic morphological features, microscopic geometric features, and topological set features of the selected original component region and the feature details corresponding to the target rule in the rule-based logistics equipment knowledge base; Step S44: Calculate the second matching degree between the original component region and the target rule based on the confidence of all feature details; Step S45: Select a target rule again in the rule-based logistics equipment knowledge base, and repeat steps S43 to S44 until the second matching degree between the original component region and all rules in the rule-based logistics equipment knowledge base is obtained, and assign the semantic label and importance level of the rule in the logistics equipment knowledge base with the highest second matching degree and higher than the second preset threshold to the original component region. Step S46: Select a new original part area and repeat steps S43 to S45 until all original part areas are matched. Step S47: Record the matching process of all original component areas. Based on the human intervention rate and matching failure characteristics in the matching process, as well as the second matching degree of the corresponding original component areas, update and adjust the rule-based logistics equipment knowledge base.

[0056] Understandably, for original component areas that fail to match standard parts, multi-level feature recognition is employed to assign corresponding semantic labels. Based on human cognitive logic for understanding three-dimensional objects and engineering practicality, feature recognition is performed at three levels: macroscopic morphology, microscopic geometry, and topological structure.

[0057] Macroscopic morphological characteristics describe the shape, size, and orientation of the original component region as a whole, enabling the fastest and coarsest classification of components. Specifically, these characteristics include: axial bounding box size ratio, compactness, and principal component ratio of moment of inertia.

[0058] For the axial bounding box size ratio, calculate the length (L) of the original part region along the three principal axes of X, Y, and Z. x ,L y ,L z ), and calculate the ratios R1, R2 and R3 between each pair.

[0059]

[0060] The definition of firmness T is:

[0061] Where S is the surface area of ​​the original component region and V is the volume of the original component region.

[0062] A higher compactness value indicates a looser and more irregular shape, such as a tree-like structure; a lower compactness value indicates a tighter and more regular shape, such as a sphere.

[0063] Regarding the principal component ratio of the moment of inertia, in physics, the moment of inertia describes the ease with which an object rotates about an axis. The farther the mass distribution of an object is from the axis, the larger the moment of inertia. In macroscopic morphological feature extraction, a component can be approximated as being made of the same homogeneous material, ignoring its mass and focusing on the spatial distribution of its vertices and triangles. This quantifies whether its macroscopic shape leans towards a rod-like, plate-like, or block-like form.

[0064] Principal component analysis is performed on the coordinates of all vertices in each segmented region: first, the covariance matrix of the vertex set is calculated. Then, eigenvalue decomposition is performed on this matrix to obtain three eigenvalues ​​arranged in descending order. These eigenvalues ​​characterize the dispersion of the vertices along the three principal directions.

[0065] By using principal component ratio analysis of moment of inertia, the aspect ratio of an object can be estimated without being affected by rotation. At the same time, the density distribution of the vertices of the original component region and the internal solid and hollow state and holes can be preliminarily analyzed.

[0066] Microscopic geometric features describe the detailed properties of the original component region's surface, focusing on the degree of curvature and complexity of the surface. Building upon macroscopic classification, this further distinguishes original component regions with similar shapes but different details. The focus is on the smoothness, curvature, and characteristic texture of the original component region's surface. Microscopic geometric features include: mean curvature, standard deviation of curvature, and boundary edge ratio.

[0067] For the mean curvature and standard deviation of curvature, calculate the average curvature or Gaussian curvature of each vertex in the original part region, and then calculate the mean curvature and standard deviation of the entire original part region.

[0068] Curvature is estimated by calculating the average difference between the vertex normal and the normals of its neighbors; the greater the difference, the higher the curvature.

[0069] To obtain the curvature values ​​{K1, K2, ..., K} of all N vertices in the original part region. N}back.

[0070] mean curvature μ K for:

[0071] Standard deviation of curvature σ K for:

[0072] A higher mean curvature indicates a greater degree of curvature on the surface of the region, primarily composed of convex or concave curved surfaces; conversely, a lower mean curvature indicates a very flat surface. This component is mainly composed of large flat areas.

[0073] A larger standard deviation of curvature indicates that the surface geometry of the original part region is drastic and uneven, containing both flat and highly curvature areas. This suggests rich detail and complex features, such as sharp edges, corners, holes, or serrated structures. Conversely, a smaller standard deviation indicates a smoother, more gradual surface geometry.

[0074] For the boundary edge ratio, the boundary edge ratio R b This represents the proportion of edges marked as strong feature edges to the total number of edges in the original component region.

[0075]

[0076] Where, N b N represents the number of boundary edges. t This represents the total number of edges in the original component region.

[0077] Boundary ratio is used to quantify the degree to which a grid component functions as a "shell" or "open structure." A low ratio indicates that the original component area is likely a physical functional component (such as a shaft or rod), which typically directly participates in the core operation of the equipment, realizing energy conversion or motion transmission, and is of higher importance. Conversely, a high ratio indicates that the original component area is likely a structural component with installation and fixing functions (such as perforated flanges or mounting plates), providing support, positioning, connection, or protection for the main functional components, and is of lower importance.

[0078] Topological set features directly map to the function and assembly relationships of components. They focus on the structural integrity of a component and its relationship with other components to infer its function. Topological set features include: motion attribute identifiers and the number of holes.

[0079] For motion attribute identification, the connection relationship between the original component area and other original component areas (whether there are geometric connection features of kinematic pairs such as hinges and slide rails) is analyzed to determine whether it is a moving part, and to infer its motion trajectory and measure the complexity of the motion. It includes three sub-attributes: whether it is moving, motion trajectory, and motion degrees of freedom.

[0080] The number of holes is approximated by calculating the number of ring boundaries in the mesh topology, which represents the number of perforations on the component. This helps distinguish between simple covers and shells, and complex mounting plates and brackets.

[0081] After feature recognition and extraction are completed, a component feature instance table is constructed to store the feature data of each original component region in order to facilitate subsequent analysis.

[0082] Table 1 Examples of Component Features

[0083] Based on the features extracted, all original component regions that failed the standard component matching are quickly and accurately matched. Before starting the matching, a logistics equipment knowledge base needs to be constructed based on logistics industry knowledge, expert experience, and historical data from a hierarchical simplified model. This knowledge base stores data in two tables: a general rule table and a detailed rule condition table.

[0084] The rule table includes all known components and their corresponding matching rule IDs, semantic tags, confidence thresholds for matching, and importance levels.

[0085] Table 2. Rule Summary Table

[0086] Component importance is categorized into three levels: L0, L1, and L2. L0 represents critical functional components that directly participate in core operations, possess motion characteristics, and are the visual focus. L1 represents major structural components, serving as important supports, moving structures, or key components that do not directly perform functions. L2 represents highly repetitive, non-core components that do not participate in major functions.

[0087] The data on component importance can be generated based on logistics industry knowledge and expert experience, according to the following rules: If a component meets the following rules, it is classified as L0: Precision moving parts are moving parts with ≥2 degrees of freedom of motion.

[0088] High-value moving parts participate in core operations, or components are located at the end of the motion chain.

[0089] High-value precision sensors are responsible for the detection, measurement, and feedback of key processes.

[0090] Direct-operation actuators, which come into direct contact with the goods / tools.

[0091] If a component meets the following rules, it is classified as L1: Motion transmission and conversion components do not directly participate in core operations; they transmit power from drive components to actuators.

[0092] Simple motion component, a motion part with only 1 degree of freedom.

[0093] Important supports and mounting components used to install L0-level components or supports that have a significant impact on accuracy.

[0094] Components that do not meet the above rules are classified as L2.

[0095] The rule condition details table records the feature details corresponding to each rule ID.

[0096] Table 3. Details of Rules and Conditions

[0097] For conditional expressions of Boolean and string types, the matching degree can only be 1 if it matches, and 0 if it does not match.

[0098] For motion attributes, the matching degree is 1 / 2 if only one condition is met, otherwise it is 0 or 1.

[0099] For numerical conditional expressions, an appropriate tolerance range is set. When the feature value of the original component region is within the tolerance range, the value between 0 and 1 is calculated as the confidence level of the feature match based on the statistical distribution law of the feature value (e.g., uniform distribution, normal distribution).

[0100] Suppose a rule has n feature details, and the weight corresponding to the i-th detail is w. i The confidence level is u i The formula for calculating the total confidence level u is as follows:

[0101] The original component regions that did not pass the fast iterative nearest point matching are used as input units. Matching is performed one by one according to the rule table. If multiple rule IDs can be matched, the one with the highest confidence is selected, and the corresponding semantic label and importance level are assigned to that rule ID. The component with the assigned semantic label is then output.

[0102] As the matching process proceeds, the matching process for each rule ID is recorded. Based on the statistical results, the rule base rules are updated and optimized after the end of this round of model process.

[0103] Table 4 Matching Process Table

[0104] Preferably, step S5 includes: Step S51: Select one original component region from all original component regions where the second matching degree is lower than the second preset threshold; Step S52: Place the selected original component area at the center of a virtual sphere; Step S53: Render a two-dimensional grayscale image or depth image of the original component area from multiple predefined perspectives on the surface of the virtual sphere; Step S54: Input all two-dimensional grayscale images or depth images into a pre-trained deep convolutional neural network to identify the original part regions, and assign semantic labels and importance levels to the original part regions based on the identification results; wherein, the deep convolutional neural network is trained with data from a historical project part model library that has been accurately labeled with semantic tags. Step S55: Select a new original component area and repeat steps S52 to S54 until all original component areas have been identified. Step S56: Label the original component regions that the deep convolutional neural network failed to recognize as unknown components and assign a preset importance level.

[0105] Preferably, step S5 further includes: Obtain semantic labels and importance level annotations for original component regions that were not recognized by the deep convolutional neural network; Based on the semantic labels and importance levels annotated by users, as well as the corresponding original component regions, the historical project component model library is updated, and the deep convolutional neural network is retrained.

[0106] Understandably, for complex or non-standard components that cannot be covered by the rules, machine learning-based methods are used as supplementary identification tools. Using original component regions that are not matched by the knowledge base system as input units, all original component regions are placed at the center of a virtual sphere, and two-dimensional grayscale or depth images of the model are rendered from multiple predefined perspectives on the sphere. A pre-trained deep convolutional neural network is then used for classification, with training data derived from a historical project component model library that has been accurately semantically labeled by modelers.

[0107] For a few special components that still cannot be matched, they can be semantically labeled as "unknown components" and assigned an importance level of L2. After the current model reduction process is completed, these components will be examined and added to the rule base to achieve continuous optimization and updating of the rule base.

[0108] Preferably, the importance levels include: L0, L1, and L2. Step S6 includes: All original component regions are simplified with high precision using a feature-weighted point cloud non-uniform simplification algorithm to generate corresponding high-precision point cloud data. Adjust the target sampling density of the non-uniform point cloud simplification algorithm to perform intermediate-precision simplification on the original component regions of L1 and L2 levels, and generate corresponding intermediate-precision point cloud data. The target sampling density of the non-uniform point cloud simplification algorithm is adjusted again to perform low-precision simplification on the original L2-level component region, generating corresponding low-precision point cloud data. The high-precision point cloud data, medium-precision point cloud data, and low-precision point cloud data are reconstructed using a grid reconstruction algorithm to generate high-precision grid models, medium-precision grid models, and low-precision grid models.

[0109] It is understandable that the segmented, individually labeled and importance-leveled component meshes are used as input units for differential processing.

[0110] Before simplification begins, the specific simplification logic and rules need to be preset in the system rule base. The simplification logic in this application is shown in the table below: Table 5 Simplified Rules Table

[0111] When users require high-precision models, all levels use high-precision models. When users require medium-precision models, only L1 and L2 level components are simplified to medium-precision levels. When users require low-precision models, L2 level components are simplified to low-precision levels based on the medium-precision component simplification. Therefore, for L0 level components, a high-precision model is required. For L1 level components, both high-precision and medium-precision models are required. For L2 level components, three precision models—high-precision, medium-precision, and low-precision—are required.

[0112] To address the need for rapidly generating models of varying precision, this application employs a feature-weighted non-uniform point cloud simplification algorithm. By adjusting the target sampling density, it generates point cloud data with different levels of detail.

[0113] Based on the calculated vertex curvature features, non-uniform density is applied, automatically increasing sampling points at edges and corners with significant curvature changes, while decreasing sampling points in flat areas. Simultaneously, the average sampling density of each local area is ensured to be close to the target sampling density. This generates a multi-resolution point cloud dataset that reflects the functional importance of components while preserving their key geometric features.

[0114] After the multi-precision point cloud data of all components has been generated, the simplified point cloud is input into a unified mesh reconstruction algorithm to generate the final multi-precision mesh model.

[0115] Preferably, step S7 includes: For advanced simplification precision, advanced precision mesh models corresponding to the original component regions at levels L0, L1, and L2 are obtained. Based on the topological relationships and component spatial transformation information recorded in the region connection diagram of the original component regions, the advanced precision mesh models are assembled to generate advanced simplification precision 3D device model data. For intermediate simplification accuracy, the high-precision mesh model corresponding to the original component region at level L0, and the intermediate-precision mesh models corresponding to the original component regions at levels L1 and L2 are obtained. Based on the topological relationships and component spatial transformation information recorded in the region connection diagram of the original component region, the high-precision mesh model and the intermediate-precision mesh model are assembled to generate 3D device model data with intermediate simplification accuracy. For low-level simplified precision, obtain the high-precision mesh model corresponding to the original component region at level L0, the medium-precision mesh model corresponding to the original component region at level L1, and the low-precision mesh model corresponding to the original component region at level L2. Based on the topological relationships and component spatial transformation information recorded in the region connection diagram of the original component region, assemble the high-precision mesh model, the medium-precision mesh model, and the low-precision mesh model to generate low-level simplified precision 3D device model data.

[0116] Understandably, based on the topological relationships and component spatial transformation information recorded in the region connectivity map, multi-precision mesh models with different levels of detail are used as input units. These models are associated according to their unique component identifiers, and corresponding spatial transformations are applied, automatically assembling them into multiple complete 3D device model data sets that are strictly consistent in component composition and assembly relationships. Finally, three models with different levels of detail are output to meet the application needs of animation, digital twins, and web page displays, respectively.

[0117] The present invention also provides an intelligent hierarchical simplification device for three-dimensional models, used to implement the above-described method embodiments. Figure 2 This is a schematic diagram of the structure of an embodiment of the intelligent hierarchical simplification device for three-dimensional models according to the present invention. Figure 2 As shown, the intelligent hierarchical simplification device for three-dimensional models in this embodiment includes a processor 21 and a memory 22, with the processor 21 connected to the memory 22. The processor 21 is used to call and execute a program stored in the memory 22; the memory 22 is used to store the program, which is at least used to execute the intelligent hierarchical simplification method for three-dimensional models in the above embodiments.

[0118] The specific implementation scheme of the intelligent hierarchical simplification device for three-dimensional models provided in this application embodiment can refer to the implementation scheme of the intelligent hierarchical simplification method for three-dimensional models in any of the above embodiments, and will not be repeated here.

[0119] It is understood that the same or similar parts in the above embodiments can be referred to each other, and the contents not described in detail in some embodiments can be referred to the same or similar contents in other embodiments.

[0120] It should be noted that in the description of this invention, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance. Furthermore, in the description of this invention, unless otherwise stated, "a plurality of" means at least two.

[0121] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as will be understood by those skilled in the art to which embodiments of the invention pertain.

[0122] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0123] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0124] Furthermore, the functional units in the various embodiments of the present invention can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0125] The storage media mentioned above can be read-only memory, disk, or optical disk, etc.

[0126] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0127] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for intelligent hierarchical simplification of three-dimensional models, characterized in that, include: Step S1: Obtain 3D device model data and convert the 3D device model data into a mesh model; Step S2: Obtain the normal and curvature of each facet in the mesh model, and divide the mesh model into multiple original component regions based on the normal and curvature of each facet using a region growing mesh segmentation algorithm. Step S3: Calculate the first matching degree between the original component region and each model in the standard parts template library, and assign the semantic label and importance level of the model in the standard parts template library with the highest first matching degree and higher than the first preset threshold to the original component region; Step S4: Perform multi-level feature recognition on the original component region where the first matching degree with each model in the standard parts template library is lower than the first preset threshold. Calculate the second matching degree with each rule in the rule-based logistics equipment knowledge base based on the multi-level feature recognition results of the original component region. Assign the semantic label and importance level of the rule in the logistics equipment knowledge base with the highest second matching degree and higher than the second preset threshold to the original component region. Step S5: Convert the original component regions with a second matching degree lower than the second preset threshold into a multi-view two-dimensional image set, identify the multi-view two-dimensional image set through a pre-trained deep convolutional neural network, and assign semantic labels and importance levels to the original component regions corresponding to the multi-view two-dimensional image set according to the identification results. Mark the original component regions that the deep convolutional neural network fails to identify as unknown components and assign a preset importance level. Step S6: Based on the importance level of each original component region, the original component regions are simplified using a feature-weighted point cloud non-uniform simplification algorithm under different simplification accuracies to generate mesh models under different simplification accuracies. Step S7: Assemble the mesh models with different simplification accuracies to generate 3D device model data with different simplification accuracies.

2. The method according to claim 1, characterized in that, Step S1 includes: Acquire 3D device model data and perform automated format parsing and data cleaning on the 3D device model data; wherein, the 3D device model data includes: CAD design files, 3D scan point clouds, and mesh models; The processed data is converted into a grid model.

3. The method according to claim 2, characterized in that, Step S2 includes: Step S21: Calculate the normal and curvature of each facet in the mesh model; Step S22: Select a seed facet in the mesh model according to the curvature of the facet; Step S23: Calculate the dihedral angle between the seed facet and its adjacent facets; Step S24: If the dihedral angle is less than the first preset angle, and the angle between the normal direction of the seed facet and the normal direction of the corresponding adjacent facet is less than the second preset angle, and the difference between the curvature of the seed facet and the curvature of the corresponding adjacent facet is less than the third threshold, then the corresponding adjacent facet is assigned to the region of the seed facet, and the average normal and average curvature of the region of the seed facet are calculated. Steps S23 and S24 are executed again until all adjacent facets of the region of the seed facet cannot be assigned to the region of the seed facet. Step S25: Repeat steps S22 to S24 until each facet in the mesh model is assigned to its corresponding region; Step S26: Calculate the bounding boxes of all regions. For any two regions, if their bounding boxes intersect or the distance between their bounding boxes is less than a fourth preset threshold, calculate the nearest point pair distance between the two region meshes. If the nearest point pair distance is less than a fifth preset threshold, create an edge in the region connection graph to connect the two regions, thus completing the division of the mesh model into multiple original component regions.

4. The method according to claim 3, characterized in that, Step S3 includes: Step S31: For each point on the original component area, find the nearest corresponding point on the target model in the standard parts template library; Step S32: Calculate the rotation and translation transformations that minimize the overall distance between all corresponding points, and perform corresponding rotation and translation transformations on the original component region based on the calculation results to complete the registration of the original component region; Step S33: Obtain valid point pairs between the registered original component region and the corresponding points on the target model in the standard part template library, and obtain the distance between each valid point pair; Step S34: Calculate the root mean square error of the distance between all the effective point pairs, and calculate the first matching degree between the original component region and the target model in the standard part template library based on the root mean square error; Step S35: Select a target model again in the standard parts template library, and execute steps S31 to S34 again to calculate the first matching degree between the original part region and the reselected target model until the first matching degree between the original part region and all models in the standard parts template library is obtained. Step S36: Assign the semantic labels and importance levels of the models in the standard parts template library with the highest first matching degree and higher than the first preset threshold to the original part region; Step S37: Reselect the original component area and repeat steps S31 to S36 until all the original component areas are matched.

5. The method according to claim 4, characterized in that, Step S4 includes: Step S41: Perform multi-level feature recognition on the original component regions where the first matching degree with each model in the standard component template library is lower than the first preset threshold, and obtain the macroscopic morphological features, microscopic geometric features, and topological set features of the original component regions; wherein, the macroscopic morphological features include: axial bounding box size ratio, compactness, and principal component ratio of moment of inertia; the microscopic geometric features include: mean curvature, standard deviation of curvature, and boundary edge ratio; and the topological set features include: motion attributes and number of holes. Step S42: Select one of the original component regions; Step S43: Calculate the confidence level between the macroscopic morphological features, microscopic geometric features, and topological set features of the selected original component region and the feature details corresponding to the target rule in the rule-based logistics equipment knowledge base; Step S44: Calculate the second matching degree between the original component region and the target rule based on the confidence level of all the described feature details; Step S45: Select a target rule again in the rule-based logistics equipment knowledge base, and repeat steps S43 to S44 until the second matching degree between the original component region and all rules in the rule-based logistics equipment knowledge base is obtained. Assign the semantic label and importance level of the rule in the logistics equipment knowledge base with the highest second matching degree and higher than the second preset threshold to the original component region. Step S46: Select a new original component region and repeat steps S43 to S45 until all original component regions are matched. Step S47: Record the matching process of all the original component regions, and update and adjust the rule-based logistics equipment knowledge base according to the human intervention rate and matching failure characteristics in the matching process, as well as the second matching degree of the corresponding original component regions.

6. The method according to claim 5, characterized in that, Step S5 includes: Step S51: Select one of the original component regions from all the original component regions where the second matching degree is lower than the second preset threshold; Step S52: Place the selected original component region at the center of a virtual sphere; Step S53: Render a two-dimensional grayscale image or depth image of the original component area from multiple predefined perspectives on the surface of the virtual sphere; Step S54: Input all the two-dimensional grayscale images or the depth images into a pre-trained deep convolutional neural network to identify the original component regions, and assign semantic labels and importance levels to the original component regions according to the identification results; wherein, the deep convolutional neural network is trained with data from a historical project component model library that has been accurately labeled with semantic tags. Step S55: Select a new original component region and repeat steps S52 to S54 until all original component regions have been identified. Step S56: The original component regions that the deep convolutional neural network failed to recognize are labeled as unknown components and a preset importance level is assigned.

7. The method according to claim 6, characterized in that, Step S5 further includes: Obtain semantic labels and importance level labels from the user for the original component regions that the deep convolutional neural network failed to recognize; The historical project component model library is updated based on the semantic tags and importance levels annotated by the user and the corresponding original component regions, and the deep convolutional neural network is retrained.

8. The method according to claim 7, characterized in that, The importance levels include: L0, L1, and L2. Step S6 includes: All the original component regions are simplified with high precision using a feature-weighted point cloud non-uniform simplification algorithm to generate corresponding high precision point cloud data. Adjust the target sampling density of the non-uniform point cloud simplification algorithm to perform intermediate-precision simplification on the original component regions of L1 and L2 levels, and generate corresponding intermediate-precision point cloud data. The target sampling density of the non-uniform point cloud simplification algorithm is adjusted again to perform low-precision simplification on the original component region at level L2, generating corresponding low-precision point cloud data. The high-precision point cloud data, the medium-precision point cloud data, and the low-precision point cloud data are reconstructed using a grid reconstruction algorithm to generate a high-precision grid model, a medium-precision grid model, and a low-precision grid model.

9. The method according to claim 8, characterized in that, Step S7 includes: For advanced simplification precision, advanced precision mesh models corresponding to the original component regions at levels L0, L1, and L2 are obtained. Based on the topological relationships and component spatial transformation information recorded in the region connection diagram of the original component regions, the advanced precision mesh models are assembled to generate advanced simplification precision 3D device model data. For intermediate simplification accuracy, obtain the high-precision mesh model corresponding to the original component region at level L0, and the intermediate-precision mesh models corresponding to the original component regions at levels L1 and L2. Based on the topological relationships and component spatial transformation information recorded in the region connection diagram of the original component region, assemble the high-precision mesh model and the intermediate-precision mesh model to generate 3D device model data with intermediate simplification accuracy. For low-level simplified precision, a high-precision mesh model corresponding to the original component region at level L0, a medium-precision mesh model corresponding to the original component region at level L1, and a low-precision mesh model corresponding to the original component region at level L2 are obtained. Based on the topological relationships and component spatial transformation information recorded in the region connection diagram of the original component region, the high-precision mesh model, the medium-precision mesh model, and the low-precision mesh model are assembled to generate low-level simplified precision 3D device model data.

10. A three-dimensional model intelligent hierarchical simplification device, characterized in that, It includes a processor and a memory, wherein the processor is connected to the memory: The processor is used to call and execute the program stored in the memory; The memory is used to store the program, which is at least used to execute the intelligent hierarchical simplification method for three-dimensional models according to any one of claims 1-9.

Images