A three-dimensional space-based model design method and system
By collecting and segmenting three-dimensional spatial data, constructing a topological network and optimizing node weights, the efficiency and accuracy problems of existing three-dimensional model design technologies in handling complex structures are solved, realizing efficient and accurate model generation and flexible application.
Patent Information
- Application Number
- CN202510565415.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-30
- Publication Date
- 2026-02-24
- Estimated Expiration
- 2045-04-30
AI Technical Summary
Existing 3D model design technology is inefficient and lacks accuracy when dealing with complex structures, making it difficult to meet the complex application needs of different fields. It also lacks versatility and flexibility, leading to resource waste and redundant development.
By acquiring the three-dimensional spatial data of the target object, using laser scanning equipment and sensor arrays to collect geometric and attribute data, performing spatial segmentation and feature extraction, constructing a three-dimensional spatial topology network, and adjusting node weights by optimizing the model, a three-dimensional model of the target is generated.
It enables efficient and accurate modeling of complex structures, improves the applicability and flexibility of the model, meets the application needs of different fields, and optimizes computational efficiency and resource utilization.
Smart Images
Figure CN120671318B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of three-dimensional model design technology, specifically to a model design method and system based on three-dimensional space. Background Technology
[0002] In today's digital age, 3D modeling has crucial applications in numerous fields, such as architectural design, mechanical manufacturing, virtual reality, and medical research. However, existing 3D modeling technologies still face many challenges and struggle to meet the ever-increasing demands of complex applications.
[0003] In the field of architectural design, traditional 3D modeling methods mainly rely on designers manually drawing or constructing based on simple measurement data. This approach is inefficient, and for complex building structures, it is difficult to accurately reproduce their true geometry and physical properties. For example, some modern buildings with irregular shapes, such as the Sydney Opera House, have unique shell structures that make it extremely difficult for traditional modeling methods to obtain accurate geometric data. This is not only time-consuming and labor-intensive but also prone to errors, leading to discrepancies between the design and actual needs.
[0004] The machinery manufacturing industry has extremely high requirements for the accuracy and functionality of 3D models. However, existing model design methods are inadequate when dealing with the internal structure and dynamic characteristics of complex mechanical parts. For example, the complex flow channels and dynamic changes of aero-engine components under extreme conditions such as high temperature, high pressure, and high-speed rotation are difficult to fully and accurately represent in the model using current technology. This affects product design optimization and performance improvement, and increases R&D costs and time.
[0005] In the fields of virtual reality (VR) and augmented reality (AR), highly realistic and interactive 3D models are needed to provide a more immersive and lifelike experience. However, current technologies suffer from problems such as large model data volumes and low computational efficiency when handling large-scale scenes and dynamic objects. For example, in large virtual city scenes, the inability to efficiently divide and extract features from buildings, terrain, and characters leads to slow loading speeds and stuttering in real-time rendering, severely impacting the user experience.
[0006] In medical research, three-dimensional modeling of human organs is a crucial tool for understanding human physiological structures and disease mechanisms. However, existing modeling techniques often struggle to accurately segment different tissues and organs when processing medical imaging data (such as CT and MRI), and they cannot effectively capture their dynamic changes under physiological and pathological conditions. For example, in liver disease research, the inability to accurately construct a three-dimensional model of the liver during the disease process hinders in-depth analysis of disease progression and the development of treatment plans.
[0007] Furthermore, existing 3D model design methods often lack versatility and flexibility when applied across different fields. Different industry needs and data characteristics require different modeling strategies, but existing technologies struggle to adapt quickly to these changes, leading to wasted resources and redundant development. Therefore, developing an efficient, accurate, and widely applicable 3D space-based model design method and system is urgently needed. Summary of the Invention
[0008] The purpose of this invention is to provide a three-dimensional space-based model design method and system to solve the problems mentioned in the background art.
[0009] To achieve the above objectives, the present invention provides the following technical solution: a model design method based on three-dimensional space, the method comprising:
[0010] Acquire three-dimensional spatial data of the target object, wherein the three-dimensional spatial data includes geometric structure data and attribute parameter data;
[0011] The three-dimensional spatial data is input into a preset spatial segmentation model to segment the three-dimensional spatial data into multiple spatial units, wherein each spatial unit contains corresponding geometric features and attribute features;
[0012] The spatial unit is input into a preset feature extraction model to extract the hierarchical relationship and dynamic change parameters of the spatial unit;
[0013] A three-dimensional spatial topology network is constructed based on the hierarchical relationships and dynamically changing parameters;
[0014] The parameters of the three-dimensional spatial topology network are adjusted using a preset optimization model to generate a target three-dimensional model;
[0015] The parameter adjustment process of the optimization model includes: dynamically calculating the optimization weight of each node based on the node density distribution and connection strength of the three-dimensional spatial topology network, and updating the topological relationship between nodes according to the optimization weight.
[0016] Preferably, acquiring the three-dimensional spatial data of the target object includes:
[0017] The surface geometry data of the target object is acquired by a laser scanning device, and the material property data of the target object is obtained by a sensor array.
[0018] The surface geometric data is denoised according to a preset spatial resolution threshold, and the denoised geometric data is aligned with the material property data in spatial coordinates.
[0019] Preferably, the training steps of the spatial segmentation model are as follows: An initial 3D training set is obtained, wherein all training samples in the initial 3D training set are labeled with standard segmentation boundaries; an initial segmentation network is iteratively trained using the initial 3D training set until the average matching degree between the predicted segmentation boundaries output by the initial segmentation network and the standard segmentation boundaries reaches a first preset threshold, thereby obtaining an intermediate segmentation model, wherein the number of training samples used in each iteration gradually decreases during the iteration process; the initial 3D training set is input into the intermediate segmentation model to obtain secondary predicted segmentation boundaries; if the average matching degree between the secondary predicted segmentation boundaries and the standard segmentation boundaries exceeds the first preset threshold, the intermediate segmentation model is used as the spatial segmentation model.
[0020] Preferably, dividing the three-dimensional spatial data into multiple spatial units includes:
[0021] Based on the curvature distribution in the geometric structure data and the gradient changes in the attribute parameter data, the initial segmentation region of the three-dimensional spatial data is divided.
[0022] The adjacent initially segmented regions are smoothed at the edges, and regions with similarity exceeding a third preset threshold are merged to form the spatial unit.
[0023] Preferably, extracting the hierarchical relationship of the spatial units includes:
[0024] Based on the center point coordinates and volume percentage of the spatial units, a proximity matrix between the spatial units is constructed;
[0025] Based on the proximity matrix and the similarity of the attribute features, the hierarchical weight of each spatial unit is calculated, and a multi-level spatial association tree is generated.
[0026] Preferably, constructing a three-dimensional spatial topology network includes:
[0027] Each node in the spatial association tree is mapped to a vertex of the topology network, and the connection strength between vertices is set according to the hierarchical weight.
[0028] Based on the dynamically changing parameters, the connection strength is corrected over time to generate a topological network structure with temporal attributes.
[0029] Preferably, the optimization model adjusts the parameters of the topology network as follows:
[0030] The objective function is configured and optimized according to the requirements of the target application scenario. The objective function includes node distribution uniformity, connection redundancy, and computational complexity.
[0031] The node positions and connection strengths of the topology network are iteratively adjusted using the gradient descent algorithm until the value of the objective function is less than a fourth preset threshold.
[0032] Preferably, during the training process of the spatial segmentation model, the methods for reducing the number of training samples include:
[0033] The samples in the first training subset are sorted in ascending order according to their prediction error rate.
[0034] The first N samples in the sorting results are retained as the second training subset, where N is the product of the preset decay coefficient of the current iteration round and the initial number of samples.
[0035] Preferably, the feature extraction model further includes:
[0036] The attribute features of the spatial unit are normalized, and a multidimensional feature vector is constructed.
[0037] The local correlation features and global distribution features in the multidimensional feature vector are extracted using a convolutional neural network.
[0038] Preferably, the present invention also includes a three-dimensional space-based model design system, the system comprising:
[0039] The data acquisition module is used to acquire three-dimensional spatial data of the target object, wherein the three-dimensional spatial data includes geometric structure data and attribute parameter data;
[0040] A spatial segmentation module, connected to the data acquisition module, is used to input the three-dimensional spatial data into a preset spatial segmentation model to segment the three-dimensional spatial data into multiple spatial units, wherein each spatial unit contains corresponding geometric features and attribute features;
[0041] The feature extraction module, connected to the spatial segmentation module, is used to input the spatial unit into a preset feature extraction model to extract the hierarchical relationship and dynamic change parameters of the spatial unit;
[0042] A topology network construction module, connected to the feature extraction module, is used to construct a three-dimensional spatial topology network based on the hierarchical relationship and dynamically changing parameters.
[0043] The model optimization module, connected to the topology network construction module, is used to adjust the parameters of the three-dimensional spatial topology network using a preset optimization model to generate a target three-dimensional model. The parameter adjustment process of the optimization model includes: dynamically calculating the optimization weight of each node based on the node density distribution and connection strength of the three-dimensional spatial topology network, and updating the topological relationship between nodes according to the optimization weight.
[0044] Compared with the prior art, the beneficial effects of the present invention are:
[0045] During the data acquisition phase, the combination of laser scanning equipment and sensor arrays enables comprehensive and accurate collection of surface geometry and material property data of the target object. Compared with traditional single-source data acquisition methods, this multi-source data acquisition approach greatly enriches the foundational data for model building. Taking architectural modeling as an example, laser scanning equipment can quickly acquire the precise external dimensions of a building, while the sensor array can simultaneously collect acoustic, thermal, and other property data of building materials. This not only helps to build more realistic architectural appearance models but also provides crucial data support for subsequent building performance analysis, such as sound insulation and thermal insulation effect simulation, making the model more closely resemble practical applications in terms of functional simulation.
[0046] In the spatial segmentation stage, a pre-set spatial segmentation model is trained based on specific training steps. It can rationally divide three-dimensional spatial data according to the curvature distribution in the geometric structure data and the gradient changes in the attribute parameter data. Compared to traditional segmentation methods based on simple geometric shapes or fixed rules, this segmentation method better reflects the actual structural characteristics of the target object. For example, in geological modeling, this model can accurately identify transitional regions between different geological layers and complex boundaries of geological structures, segmenting geological bodies into multiple spatial units with clear geometric and attribute characteristics, providing a more accurate model foundation for subsequent geological analysis and resource exploration.
[0047] The feature extraction model extracts the hierarchical relationships of spatial units by constructing a proximity matrix and calculating hierarchical weights, while also acquiring dynamically changing parameters. This allows the constructed 3D model to fully reflect the hierarchical structure and dynamic characteristics of the various parts within the target object. In equipment modeling on industrial production lines, this model can clearly present the assembly hierarchy between various components, as well as the dynamic changes in parameters such as temperature and pressure of each component during equipment operation. This helps engineers to promptly identify potential faults, perform preventative maintenance, and improve equipment reliability and production efficiency.
[0048] When constructing a 3D spatial topological network, nodes of the spatial association tree are mapped to vertices of the topological network, and connection strength is set according to hierarchical weights. Simultaneously, dynamic parameters are considered for time-series correction. This topological network structure can intuitively and accurately describe the spatial relationships and dynamic evolution of target objects. Taking urban traffic network modeling as an example, the topological network constructed using this method can reflect real-time information such as road congestion, vehicle flow direction, and speed changes over different time periods. This provides strong decision-making support for traffic planning and intelligent traffic management, optimizes traffic flow allocation, and alleviates traffic congestion.
[0049] The optimization model dynamically calculates node weights based on node density distribution and connection strength, adjusting parameters of the topology network. This process allows for flexible configuration of the objective function according to different target application scenarios, effectively balancing node distribution uniformity, connection redundancy, and computational complexity. In virtual reality scene modeling, optimizing the model can reduce the amount of model data, improve rendering efficiency, and reduce the computational burden on hardware devices while maintaining the scene's visual effects, enabling users to obtain a smoother and more realistic virtual reality experience. Attached Figure Description
[0050] Figure 1 This is a schematic diagram illustrating the working principle of the three-dimensional space-based model design method described in this invention.
[0051] Figure 2 A flowchart for obtaining 3D spatial data of a target object;
[0052] Figure 3 Flowchart for training a spatial segmentation model;
[0053] Figure 4 A flowchart for constructing a three-dimensional spatial topology network. Detailed Implementation
[0054] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0055] Please see Figures 1-4 This invention relates to a model design method based on three-dimensional space, and its specific implementation is described in detail below.
[0056] Acquire three-dimensional spatial data of the target object. This data includes geometric structure data and attribute parameter data, which form the basis for subsequent model construction. For example, when designing a model of a building, the geometric structure data may include the building's external dimensions and the shape of each part, while the attribute parameter data may include the building's material and the functional attributes of different areas.
[0057] The acquired 3D spatial data is input into a pre-defined spatial segmentation model. This model divides the 3D spatial data into multiple spatial units, each containing corresponding geometric and attribute features. This method allows for the initial refinement and classification of complex 3D spatial data, facilitating subsequent, more in-depth analysis.
[0058] These spatial units are input into a pre-defined feature extraction model to extract their hierarchical relationships and dynamic parameters. The hierarchical relationships reflect the hierarchical structure and interrelationships between different spatial units, while the dynamic parameters reflect how these spatial units change over time or due to other factors.
[0059] A three-dimensional spatial topology network is constructed based on the extracted hierarchical relationships and dynamically changing parameters. This topology network can intuitively display the relationships between spatial units, providing a clear framework for subsequent model optimization.
[0060] The parameters of the constructed 3D spatial topology network are adjusted using a pre-defined optimization model to generate the target 3D model. During the optimization process, the optimization weight of each node is dynamically calculated based on the node density distribution and connection strength of the 3D spatial topology network, and the topological relationships between nodes are updated according to the optimization weights, so that the final generated model better meets the actual needs.
[0061] The present invention will be further described below with reference to Examples 1 to 6:
[0062] Example 1:
[0063] In a practical application scenario, taking the digital 3D modeling of a historical building as an example, this article details the process of obtaining 3D spatial data of the target object.
[0064] High-precision laser scanning equipment was used to collect surface geometric data of the historical building. The laser scanning equipment accurately obtains the spatial coordinates of various points on the building's surface by emitting a laser beam and measuring the time delay of the reflected light. For example, using a terrestrial 3D laser scanner, scanning around the building at specific intervals, can comprehensively cover the building's facade, doors, windows, decorative components, and other parts, obtaining massive amounts of point cloud data. This point cloud data forms the basis of the building's surface geometric data.
[0065] At the same time, a sensor array composed of various types of sensors is used to acquire data on the building's material properties. For example, a spectral sensor can be used to analyze the reflection characteristics of the building's surface materials to different wavelengths of light, thereby inferring the type of material used in the building, such as stone, wood, or brick. A humidity sensor can be used to measure the humidity of different parts of the building, reflecting the degree of moisture absorption of the building materials. This data is crucial for assessing the building's preservation status.
[0066] After data acquisition, the surface geometry data needs to be denoised. The acquired point cloud data is filtered based on a preset spatial resolution threshold. For example, a distance threshold is set; if the distance between a point and its neighboring points exceeds this threshold, and the point's position is unstable across multiple scans, then that point is identified as noise and removed. This is because during actual scanning, interference from ambient light, reflections from other surrounding objects, and other factors may lead to the acquisition of erroneous point data.
[0067] After noise reduction, the denoised geometric data and material property data are aligned in spatial coordinates. This step is achieved by establishing a unified coordinate system, using a landmark corner or central axis of the building as a reference, and mapping both geometric and material property data to this unified coordinate system. In this way, each geometric location can accurately correspond to its material property information, providing an accurate data foundation for subsequent model construction.
[0068] Example 2:
[0069] Taking the spatial segmentation modeling of three-dimensional geological structure data as an example, this paper illustrates the training steps of the spatial segmentation model.
[0070] Obtain an initial 3D training set. This training set contains a large number of 3D data samples of geological structures from different regions, and each sample has been annotated with standard segmentation boundaries by professional geological surveyors. These standard segmentation boundaries are determined based on features such as geological bedding and variations in rock types, representing the true division of geological structures.
[0071] The initial segmentation network is iteratively trained using an initial 3D training set. During training, the number of training samples used in each iteration gradually decreases. Specifically, the samples in the first training subset are sorted in ascending order according to their prediction error rate. The prediction error rate is determined by calculating the difference between the segmentation boundary predicted by the model and the standard segmentation boundary, for example, using the Hausdorff distance to measure the distance between them; the larger the distance, the higher the prediction error rate. Then, the top N samples in the sorted result are retained as the second training subset, where N is the product of the preset decay coefficient for the current iteration and the initial number of samples. The preset decay coefficient is determined based on experience and previous experiments, and it controls the rate at which the number of training samples decreases. For example, in the early stages of training, the preset decay coefficient might be set to 0.9, gradually decreasing to 0.7 as the number of iterations increases. This allows for rapid learning of the model's basic features using a larger number of samples in the early stages of training, while reducing the number of samples in the later stages, improving training efficiency and preventing overfitting.
[0072] When iterative training causes the average matching degree between the predicted segmentation boundary and the standard segmentation boundary output by the initial segmentation network to reach a first preset threshold, an intermediate segmentation model is obtained. The matching degree here can be measured by the intersection-over-union (IoU), which is the area of the intersection of the predicted segmentation region and the standard segmentation region divided by the area of their union. The closer the IoU value is to 1, the higher the matching degree.
[0073] The initial 3D training set is input into the intermediate segmentation model to obtain secondary predicted segmentation boundaries. If the average matching degree between the secondary predicted segmentation boundaries and the standard segmentation boundaries exceeds a first preset threshold, the intermediate segmentation model is used as the final spatial segmentation model. Through this multi-round training and evaluation method, a high-performance spatial segmentation model can be obtained, ensuring accurate segmentation of subsequent 3D geological structure data.
[0074] Example 3:
[0075] Taking the spatial segmentation of three-dimensional medical image data of human organs (such as the liver) as an example, this paper details the steps of segmenting three-dimensional spatial data into multiple spatial units.
[0076] Based on the curvature distribution and gradient changes in attribute parameter data within the geometric structural data, initial segmentation regions of the 3D spatial data are defined. In 3D medical images of the liver, the curvature distribution of the geometric structural data reflects the changes in the degree of curvature of the liver's surface and internal structures. For example, the curvature changes are greater at the liver's periphery, while the curvature changes are smaller in the relatively smooth internal regions. The gradient changes in attribute parameter data are reflected in the density differences between different tissues. For instance, the density of structures such as blood vessels and bile ducts within the liver differs from that of the surrounding liver tissue, resulting in significant gradient changes. By analyzing these curvature distributions and gradient changes, the 3D data of the liver is initially divided into multiple initial segmentation regions, such as separating the liver periphery and the areas around internal blood vessels that exhibit significant characteristic differences.
[0077] Edge smoothing is performed on adjacent initially segmented regions. Since the boundaries of the initially segmented regions may be jagged or discontinuous, this can affect subsequent analysis and processing of spatial units. A Gaussian filtering algorithm is used to smooth the boundaries. This algorithm makes the boundaries more continuous and natural by weighted averaging of boundary pixels. For example, for the boundary of a region in a 2D slice, the weight of each pixel is calculated according to the Gaussian function, and the grayscale values of the boundary pixels are adjusted to achieve edge smoothing.
[0078] Regions with similarity exceeding a third preset threshold are merged to form the final spatial unit. Similarity is determined by comparing the geometric features (such as shape and size) and attribute features (such as tissue density and texture) of adjacent regions. Taking texture features as an example, the Gray-Level Co-Occurrence Matrix (GLCM) of adjacent regions is calculated. Texture feature parameters such as contrast, correlation, energy, and entropy are extracted through the GLCM, and then the Euclidean distance between these parameters is calculated as the similarity index. If the similarity index of adjacent regions exceeds the third preset threshold, they are merged into a single spatial unit. This reduces the fragmentation of the segmentation and makes the spatial unit more consistent with the actual anatomical structure.
[0079] Example 4:
[0080] Taking the hierarchical relationship of spatial units in a 3D model of an urban transportation area as an example, the extraction process is explained in detail.
[0081] Based on the center point coordinates and volume proportions of spatial units, a proximity matrix is constructed between spatial units. In urban transportation areas, different transportation hubs (such as train stations and bus stations), major road intersections, and large parking lots are considered as spatial units. The center point coordinates of each spatial unit are calculated, and their proximity is determined by measuring the Euclidean distance between the center points. Simultaneously, the volume proportion of each spatial unit is considered; a larger volume proportion indicates a higher relative importance of the spatial unit within the overall area. For example, a train station, as a large transportation hub, has a large volume proportion and is likely to have closer connections with other transportation units. Based on these factors, a proximity matrix is constructed, where the element values represent the proximity between two spatial units; the closer the distance and the more balanced the volume proportions, the larger the element value.
[0082] Based on the proximity matrix and the similarity of attribute features, the hierarchical weight of each spatial unit is calculated. The similarity of attribute features can be measured from multiple aspects, such as traffic flow and functional type. For traffic flow, the number of vehicles passing through each spatial unit within a certain time period is statistically analyzed to calculate the correlation of traffic flow between adjacent spatial units. For functional type, it is determined whether the spatial unit is a passenger hub, freight hub, or traffic transfer node, etc. The proximity matrix and attribute feature similarity are combined, and a weighted average method is used to calculate the hierarchical weight of each spatial unit. For example, setting the weight of proximity relationship to 0.6 and the weight of attribute feature similarity to 0.4, the weight is calculated using the formula W... i =0.6×P i +0.4×Q i Calculate the hierarchical weight W of the i-th spatial unit. i , where P iQ is the neighbor weight of the i-th spatial unit obtained from the neighbor relation matrix. i It is the attribute weight of the i-th spatial unit obtained based on the similarity of attribute features.
[0083] A multi-level spatial relationship tree is generated based on hierarchical weights. In this tree, spatial units with higher hierarchical weights are located at higher levels, reflecting their importance and influence within the urban transportation network. For example, railway stations, due to their high traffic volume and crucial function, have high hierarchical weights and are located at higher levels in the spatial relationship tree, while smaller traffic artery intersections have lower hierarchical weights and are located at lower levels. This method clearly demonstrates the hierarchical relationships between spatial units within an urban transportation area, providing strong support for transportation planning and management.
[0084] Example 5:
[0085] Taking the construction of a three-dimensional spatial topology network of a large logistics park as an example, the construction process is explained in detail.
[0086] Each node in the spatial association tree is mapped to a vertex of the topological network. In a logistics park, the spatial association tree is obtained by extracting the hierarchical relationship of spatial units in the previous embodiment. The nodes in the tree represent different functional areas, such as warehouses, sorting centers, and loading / unloading areas. These nodes are then mapped one-to-one to the vertices of the topological network, with each vertex representing a specific functional area within the logistics park.
[0087] The connection strength between vertices is set according to hierarchical weights. The higher the hierarchical weight, the stronger the connection between the corresponding vertices. For example, the hierarchical weight between the warehouse and the sorting center is high because they work closely together in the logistics process, with goods frequently flowing between them, so the connection strength between their corresponding vertices is set to be large. Conversely, the hierarchical weights of some auxiliary areas (such as employee rest areas) and the main logistics functional areas are lower, and the connection strength between their corresponding vertices is also smaller. In this way, the degree of connection between different functional areas is reflected.
[0088] By adjusting connection strength based on dynamically changing parameters over time, a topological network structure with temporal attributes is generated. In logistics parks, dynamically changing parameters include variations in cargo flow and the level of business activity at different times. For example, during peak daytime shipping periods, the cargo flow from the warehouse to the loading and unloading area is high, so the connection strength between corresponding vertices in these two areas is increased; during off-peak hours at night, the connection strength is appropriately reduced. By monitoring these dynamically changing parameters in real time and adjusting the connection strength between vertices at different points in time, the resulting topological network structure reflects the changing operational status of the logistics park at different times, which helps optimize logistics processes and resource allocation.
[0089] Example 6:
[0090] This embodiment focuses on the parameter adjustment process of the optimization model for the topology network. Taking the optimization of a 3D layout model of an industrial production line as an example, the objective function is first configured according to the requirements of the target application scenario. In the industrial production line scenario, the objective function includes node distribution uniformity, connection redundancy, and computational complexity. Node distribution uniformity is used to measure whether the distribution of nodes in the topology network is uniform, avoiding situations where nodes are overly concentrated or sparse. The formula can be expressed as:
[0091]
[0092] Where U represents the uniformity of node distribution, n is the total number of nodes, and d i It is the average distance from the i-th node to all other nodes. It is the average of the average distances from all nodes to other nodes. Connection redundancy is used to evaluate the degree of duplication and redundancy of connections in a network, and the formula is:
[0093]
[0094] Where R represents the connection redundancy. Computational complexity reflects the ease of model computation and is related to factors such as network structure and the number of nodes.
[0095] The gradient descent algorithm iteratively adjusts the node positions and connection strengths of the topology network until the objective function value is less than a fourth preset threshold. Gradient descent is a commonly used optimization algorithm that calculates the gradient of the objective function with respect to node positions and connection strengths, and then gradually adjusts the parameters in the opposite direction of the gradient to reduce the objective function value. In each iteration, the gradient of the objective function is calculated based on the current node positions and connection strengths, and then the node positions and connection strengths are updated with a certain step size. For example, the formula for updating node positions can be expressed as:
[0096]
[0097] in It is the position of the i-th node in the t-th iteration, and α is the step size. Is the objective function J about The gradient is calculated. This process is repeated until the value of the objective function is less than the fourth preset threshold. The resulting topology network is the optimized result, and the target 3D model generated based on this network can better meet the actual needs of industrial production line layout.
[0098] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.
[0099] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A model design method based on three-dimensional space, characterized in that, include: Acquire three-dimensional spatial data of the target object, wherein the three-dimensional spatial data includes geometric structure data and attribute parameter data; The three-dimensional spatial data is input into a preset spatial segmentation model to segment the three-dimensional spatial data into multiple spatial units, wherein each spatial unit contains corresponding geometric features and attribute features; The spatial unit is input into a preset feature extraction model to extract the hierarchical relationship and dynamic change parameters of the spatial unit; A three-dimensional spatial topology network is constructed based on the hierarchical relationships and dynamically changing parameters; The parameters of the three-dimensional spatial topology network are adjusted using a preset optimization model to generate a target three-dimensional model; The parameter adjustment process of the optimization model includes: dynamically calculating the optimization weight of each node based on the node density distribution and connection strength of the three-dimensional spatial topology network, and updating the topological relationship between nodes according to the optimization weight; Obtaining the three-dimensional spatial data of the target object includes: The surface geometry data of the target object is acquired by a laser scanning device, and the material property data of the target object is obtained by a sensor array. The surface geometric data is denoised according to a preset spatial resolution threshold, and the denoised geometric data is aligned with the material property data in spatial coordinates. Extracting the hierarchical relationships of the spatial units includes: Based on the center point coordinates and volume percentage of the spatial units, a proximity matrix between the spatial units is constructed; Based on the proximity matrix and the similarity of the attribute features, the hierarchical weight of each spatial unit is calculated, and a multi-level spatial association tree is generated. Constructing a three-dimensional spatial topology network includes: Each node in the spatial association tree is mapped to a vertex of the topology network, and the connection strength between vertices is set according to the hierarchical weight. Based on the dynamically changing parameters, the connection strength is corrected over time to generate a topological network structure with temporal attributes; The optimization model adjusts the parameters of the topology network as follows: The objective function is configured and optimized according to the requirements of the target application scenario. The objective function includes node distribution uniformity, connection redundancy, and computational complexity. The node positions and connection strengths of the topology network are iteratively adjusted using the gradient descent algorithm until the value of the objective function is less than a fourth preset threshold.
2. The model design method based on three-dimensional space according to claim 1, characterized in that, The training steps of the spatial segmentation model are as follows: An initial 3D training set is obtained, wherein all training samples in the initial 3D training set are labeled with standard segmentation boundaries; the initial segmentation network is iteratively trained using the initial 3D training set until the average matching degree between the predicted segmentation boundaries output by the initial segmentation network and the standard segmentation boundaries reaches a first preset threshold, thus obtaining an intermediate segmentation model, wherein the number of training samples used in each iteration gradually decreases during the iteration process; the initial 3D training set is input into the intermediate segmentation model to obtain secondary predicted segmentation boundaries; if the average matching degree between the secondary predicted segmentation boundaries and the standard segmentation boundaries exceeds the first preset threshold, the intermediate segmentation model is used as the spatial segmentation model.
3. The model design method based on three-dimensional space according to claim 2, characterized in that, Dividing the three-dimensional spatial data into multiple spatial units includes: Based on the curvature distribution in the geometric structure data and the gradient changes in the attribute parameter data, the initial segmentation region of the three-dimensional spatial data is divided. The adjacent initially segmented regions are smoothed at the edges, and regions with similarity exceeding a third preset threshold are merged to form the spatial unit.
4. The model design method based on three-dimensional space according to claim 3, characterized in that, During the training process of the spatial segmentation model, the methods for reducing the number of training samples include: The samples in the first training subset are sorted in ascending order according to their prediction error rate. The first N samples in the sorting results are retained as the second training subset, where N is the product of the preset decay coefficient of the current iteration and the number of initial samples.
5. The model design method based on three-dimensional space according to claim 4, characterized in that, The feature extraction model also includes: The attribute features of the spatial unit are normalized, and a multidimensional feature vector is constructed. The local correlation features and global distribution features in the multidimensional feature vector are extracted using a convolutional neural network.
6. A model design system based on three-dimensional space, characterized in that, A method for implementing a three-dimensional space-based model design method as described in any one of claims 1-5, comprising: The data acquisition module is used to acquire three-dimensional spatial data of the target object, wherein the three-dimensional spatial data includes geometric structure data and attribute parameter data; A spatial segmentation module, connected to the data acquisition module, is used to input the three-dimensional spatial data into a preset spatial segmentation model to segment the three-dimensional spatial data into multiple spatial units, wherein each spatial unit contains corresponding geometric features and attribute features; The feature extraction module, connected to the spatial segmentation module, is used to input the spatial unit into a preset feature extraction model to extract the hierarchical relationship and dynamic change parameters of the spatial unit; A topology network construction module, connected to the feature extraction module, is used to construct a three-dimensional spatial topology network based on the hierarchical relationship and dynamically changing parameters. The model optimization module, connected to the topology network construction module, is used to adjust the parameters of the three-dimensional spatial topology network using a preset optimization model to generate a target three-dimensional model. The parameter adjustment process of the optimization model includes: dynamically calculating the optimization weight of each node based on the node density distribution and connection strength of the three-dimensional spatial topology network, and updating the topological relationship between nodes according to the optimization weight.