Three-dimensional digital-analog retrieval and clustering method and system

Abstract

The invention discloses a three-dimensional digital-analog retrieval and clustering method and a three-dimensional digital-analog retrieval and clustering system. The method comprises the following steps: S1, selecting n representation models; s2, converting historical feature vectors; s3, splicing historical feature vectors; s4, converting a to-be-retrieved feature vector; s5, splicing the feature vectors to be retrieved; and S6, performing similarity matching and clustering analysis. According to the three-dimensional digital-analog retrieval and clustering method and system provided by the invention, the three-dimensional digital-analog can be comprehensively and effectively processed, multi-modal features are deeply fused to improve the comprehensiveness of representation, the retrieval precision and efficiency are improved through hierarchical feature processing, and the weight is dynamically adjusted to adapt to multi-scene requirements.

Description

Technical Field

[0001] The present invention relates to the field of aerospace technology, and in particular to a three-dimensional digital model retrieval and clustering method and system. Background Art

[0002] Among existing feature extraction methods, 3D feature extraction technologies, such as CGAL and Open3D, achieve model representation through geometric algorithms or topological analysis. While these methods offer the advantages of strong interpretability and high algorithmic maturity, their reliance on manually designed features limits their ability to capture complex nonlinear features. This is particularly true when dealing with high-degree-of-freedom surfaces or unstructured models, where generalization performance significantly decreases. Furthermore, the rigid design of feature dimensions makes it difficult to adapt to dynamic scenarios and lacks flexibility.

[0003] As for point cloud processing technology, PointNet, as a pioneering architecture for directly processing point cloud data, breaks through the dimensionality limitations of traditional voxelization methods by aggregating point cloud features through global pooling. However, its single-layer feature extraction mechanism ignores local geometric associations and has weak representation capabilities for models with fine local structures (such as threads of mechanical parts and blood vessel branches in biological models). At the same time, the model is highly sensitive to the density distribution of point clouds, with a mismatch rate of over 15% in sparse areas, and lacks the ability to fuse hierarchical multi-scale features.

[0004] Voxel-grid-based 3D CNN methods directly process voxelized data using three-dimensional convolution kernels, demonstrating outstanding performance in analyzing volumetric data such as medical imaging. However, their computational complexity increases cubically (e.g., a 256×256×256 voxel grid requires over 16 million convolution operations), resulting in inefficient training and inference. In practical applications, resolution is often reduced to below 64×64×64, resulting in a loss of over 30% of geometric detail. Furthermore, the voxel grid's inefficient handling of surface sparsity (over 90% of empty voxels are involved in the computation) further exacerbates resource waste.

[0005] While multi-view 2D projection can circumvent the complexity of 3D data processing, reusing mature 2D CNN technology is a core advantage of multi-view projection methods. However, the projection process results in the loss of depth information and geometric features of occluded areas. For non-rigid objects (such as deformable robotic arms) or topologically complex models (such as porous media), retrieval accuracy drops by approximately 22% compared to point cloud methods. The sensitivity of view selection strategies (e.g., a 35% increase in error when the number of viewpoints falls below six) and the difficulty of fusion consistency between viewpoints also hinder their industrial application.

[0006] While deep learning methods based on triangular mesh surface topology (such as MeshCNN) can accurately represent model details, their processing efficiency is limited by the irregular data structure of the mesh. The computational overhead of a single graph convolution operation is 3-5 times higher than that of a regular point cloud and is highly sensitive to mesh quality. Models with more than 5% holes or noise require additional preprocessing, resulting in a 40% increase in end-to-end processing latency. Furthermore, existing frameworks have limited support for large-scale mesh datasets (such as models with over a million facets), making it difficult to meet the needs of real-time retrieval.

[0007] While improving local matching accuracy through substructure decomposition offers unique value in assembly retrieval scenarios, the time complexity of decomposition algorithms (e.g., O(n²)) leads to a significant increase in processing time when processing models exceeding 10,000 components. Furthermore, the lack of global semantic features can lead to mismatches (e.g., gear tooth profiles may be locally similar but their overall dimensions may not match). While acceleration methods based on spatial indexes (KD-Tree and R-Tree) can reduce retrieval time by 60%, the cost of index construction and dynamic updates accounts for over 45% of the total system overhead, limiting their application in frequently updated industrial databases.

[0008] In summary, existing digital-analog search methods generally have the defects of single representation dimension and fixed weight. They over-rely on a single data modality and lack a collaborative optimization mechanism for cross-modal features. Moreover, the feature fusion strategy is rigid and cannot dynamically adjust the contribution weights of multi-source features according to the needs of diverse scenarios.

[0009] In view of the above problems, the existing technology is in urgent need of improvement. Summary of the Invention

[0010] In order to solve at least one of the problems existing in the prior art described above, a first aspect of the present invention provides a three-dimensional digital model retrieval and clustering method, which includes the following steps:

[0011] Step S1: selecting n representation models; wherein the n representation models are selected according to the target application scenario and the dimension of each representation vector is selected according to the performance of the representation model;

[0012] Step S2: historical feature vector conversion; wherein, the historical three-dimensional digital models are converted into n groups of historical feature vectors using the n representation models, and the historical feature vectors of the changed historical three-dimensional digital models are updated in real time;

[0013] Step S3: splicing historical feature vectors; wherein n groups of historical feature vectors of the historical three-dimensional digital model are spliced into a comprehensive historical feature vector;

[0014] Step S4: converting the feature vectors to be retrieved; wherein, the digital model to be retrieved is converted into n groups of feature vectors to be retrieved using the n representation models;

[0015] Step S5: splicing the feature vectors to be retrieved; wherein, the n groups of feature vectors to be retrieved of the digital model to be retrieved are spliced into a comprehensive feature vector to be retrieved;

[0016] Step S6: similarity matching and cluster analysis; wherein, by performing similarity matching on the feature vector to be retrieved and the historical feature vector, a similarity ranking is generated, and cluster analysis is performed on the historical feature vector to generate a category classification result.

[0017] In the three-dimensional digital model retrieval and clustering method as described above, optionally, in step S1, the n representation models include at least two of the following: a multi-view projection model, a metadata model, a text representation model, and a three-dimensional structure representation model.

[0018] In the three-dimensional digital model retrieval and clustering method as described above, optionally, step S1 includes the following steps:

[0019] Selecting the multi-view projection model to analyze the features of the digital model two-dimensional view, the dimension of the representation vector of the multi-view projection model is set to d1;

[0020] Selecting the metadata model to analyze features of the digital model metadata, the metadata features including at least one of the following: part thickness, number of mounting holes, and part width, and setting the dimension of the representation vector of the metadata model to d2 according to the number of metadata;

[0021] Selecting the text representation model to analyze the text features of the digital model, the dimension of the representation vector of the text representation model is set to d3;

[0022] The three-dimensional structure representation model is selected to analyze the topological structure characteristics of the digital model, and the dimension of the three-dimensional structure representation model is set to d4.

[0023] In the three-dimensional digital model retrieval and clustering method as described above, optionally, d1=256, d2=16, d3=256, d4=256.

[0024] In the three-dimensional digital model retrieval and clustering method as described above, optionally, by adjusting the weight ratios of different representation models to convert them into corresponding dimensional allocations, after the splicing process of step S3 and step S5, the comprehensive historical feature vector and the comprehensive feature vector to be retrieved spliced according to the weight ratio are output.

[0025] In the three-dimensional digital model retrieval and clustering method as described above, optionally, in step S6, the cosine similarity between the comprehensive feature vector to be retrieved and the comprehensive historical feature vector is calculated to generate a retrieval ranking result.

[0026] In the three-dimensional digital model retrieval and clustering method as described above, optionally, in step S6, the n groups of feature vectors to be retrieved are subjected to independent similarity matching to generate a retrieval ranking result.

[0027] To achieve the above-mentioned object, a second aspect of the present invention provides a three-dimensional digital model retrieval and clustering system, wherein the three-dimensional digital model retrieval and clustering method as described in any one of the first aspects is used, including:

[0028] A representation model selection module is configured to select n representation models according to a target application scenario and to select the dimension of each representation vector according to the performance of the representation model;

[0029] Historical feature vector conversion module: uses n representation models to convert historical three-dimensional digital models into n groups of historical feature vectors, and updates the historical feature vectors of the changed historical three-dimensional digital models in real time;

[0030] Historical feature vector splicing module: splices n groups of historical feature vectors of the historical three-dimensional digital model into a comprehensive historical feature vector;

[0031] Feature vector conversion module to be retrieved: uses n representation models to convert the digital model to be retrieved into n groups of feature vectors to be retrieved;

[0032] Feature vector splicing module to be retrieved: splicing n groups of feature vectors to be retrieved of the digital model to be retrieved into a comprehensive feature vector to be retrieved;

[0033] Similarity matching and cluster analysis module: By performing similarity matching between the feature vector to be retrieved and the historical feature vector, a similarity ranking is generated, and cluster analysis is performed on the historical feature vector to generate a category division result.

[0034] In order to achieve the above-mentioned purpose, the third aspect of the present invention provides a computer device, including a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor runs the program, the three-dimensional digital model retrieval and clustering method as described in any one of the first aspects above is implemented.

[0035] In order to achieve the above-mentioned purpose, the fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions or computer programs, and when the computer-executable instructions or computer programs are processed and executed, the three-dimensional digital model retrieval and clustering method as described in any one of the embodiments in the first aspect above is implemented.

[0036] The three-dimensional digital model retrieval and clustering method and system provided by the present invention can comprehensively and effectively process three-dimensional digital models, deeply integrate multimodal features to improve the comprehensiveness of representation, hierarchical feature processing to improve retrieval accuracy and efficiency, and dynamically adjust weights to adapt to multi-scenario requirements.

[0037] The concept, specific structure and technical effects of the present invention will be further described below in conjunction with the accompanying drawings to fully understand the purpose, characteristics and effects of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments or descriptions of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0039] Figure 1 It is a flow chart of an embodiment of a three-dimensional digital model retrieval and clustering method of the present invention. DETAILED DESCRIPTION

[0040] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0041] Terms such as “comprise” and “include” indicate that in addition to the components directly and explicitly stated in the description and claims, the technical solution of the present invention does not exclude the situation where it has other components that are not directly or explicitly stated.

[0042] like Figure 1 As shown, the three-dimensional digital model retrieval and clustering method of the present invention may specifically include the following steps:

[0043] Step S1: Select n representation models.

[0044] In step S1 , n representation models are selected according to the target application scenario and the dimension, ie, the length, of each representation vector is selected according to the performance of the representation model.

[0045] In an optional embodiment, the n representation models may include at least a multi-view projection model, a metadata model, a text representation model, and a three-dimensional structure representation model.

[0046] Optionally, step S1 may further include the following steps:

[0047] The multi-view projection model is selected to analyze the characteristics of the two-dimensional view of the digital model, and the dimension of the representation vector of the multi-view projection model is set to d1;

[0048] A metadata model is selected to analyze the features of the digital model metadata. The features of the metadata may include at least part thickness, number of mounting holes, and part width. The dimension of the representation vector of the metadata model may be set to d2 according to the number of metadata.

[0049] Select the text representation model to analyze the text features of the digital model, and the dimension of the representation vector of the text representation model is set to d3;

[0050] A three-dimensional structural representation model is selected to analyze the topological structural characteristics of the digital model, and the dimension of the three-dimensional structural representation model is set to d4.

[0051] In this embodiment, the representation models are stored or calculated in the form of column vectors, and the dimensions of each model can be set to: d1 = 256, d2 = 16, d3 = 256, and d4 = 256. This precise dimension setting ensures that each model can fully utilize its feature extraction capabilities while maintaining a reasonable size for the overall feature vector, which helps improve the efficiency and accuracy of retrieval and clustering. For example, when performing similarity matching, the distance between feature vectors can be calculated more quickly without the excessive computational burden caused by excessive dimensionality.

[0052] In this embodiment, the model category is used as the output when training the relevant model. A layer of neurons is added to the penultimate layer of the neural network, with the number of neurons corresponding to the desired characteristic length. During training, the weights of this layer of neurons are updated as the neural network trains. In actual applications, the output of this layer of neurons is used as the final model output.

[0053] In other optional embodiments, those skilled in the art may also select several representation models according to actual application scenarios, and select the dimensions of the representation vectors according to the performance of the representation models.

[0054] The multi-model combination method of this embodiment has the following advantages:

[0055] First, different types of representation models can complement each other's strengths. For example, multi-view projection models are good at capturing appearance features but may ignore internal structures; while 3D structural representation models can make up for this shortcoming and provide more comprehensive model analysis.

[0056] Secondly, by selecting at least two models, the system's adaptability and flexibility can be improved. The most appropriate model combination can be selected based on the specific application scenario and data characteristics. For example, for models with similar appearance but different internal structures, a multi-view projection model and a 3D structure representation model can be combined. For models with rich text descriptions, a text representation model can be added to improve retrieval accuracy.

[0057] In addition, multi-model fusion can improve the system's robustness to different types of features. A single model may perform poorly on certain types of data, and the combination of multiple models can reduce this risk and improve the overall performance and stability of the system.

[0058] In practical applications, different model combinations can be selected based on specific needs. For example, in an industrial parts retrieval system, a combination of a multi-view projection model, a metadata model, and a 3D structural representation model can be used. The multi-view projection model captures the part's appearance, the metadata model analyzes part attributes such as dimensions and material, and the 3D structural representation model analyzes the part's internal structure and geometric relationships. This combination allows for a comprehensive analysis of all aspects of the part, improving retrieval accuracy and efficiency.

[0059] Specifically, for example, when processing a complex mechanical part, the multi-view projection model can generate six views of the part (front view, back view, left view, right view, top view, and bottom view), and each view can be represented by a 256-dimensional feature vector. The metadata model can extract attribute information such as the thickness, width, length, weight, and material type of the part, and represent it with a 16-dimensional feature vector. The three-dimensional structure representation model can analyze the topological structure of the part and generate a 256-dimensional feature vector. These feature vectors can be combined into a comprehensive feature vector through splicing or weighted fusion for subsequent retrieval and clustering analysis.

[0060] By combining these multiple models, the system can simultaneously consider a part's appearance, attributes, and internal structure, significantly improving retrieval and clustering accuracy. For example, when searching for similar parts, the system not only matches parts with similar appearance but also considers similar attributes and internal structure, thereby finding parts with similar functions and structures.

[0061] Step S2: Historical feature vector conversion, this step is a preprocessing step before retrieval.

[0062] In step S2, n representation models are used to convert the historical three-dimensional digital models into n groups of historical feature vectors, and the historical feature vectors of the changed historical three-dimensional digital models are updated in real time.

[0063] The historical three-dimensional digital model here refers to a processed benchmark data set, a set of three-dimensional models that have been pre-collected, cleaned and feature-characterized in the system, and serves as a comparison benchmark for retrieval and clustering. It also serves as a dynamically updated knowledge base for this application. When the model data is modified (such as design iteration, parameter adjustment), the system will synchronously update its feature vector to ensure data timeliness. Each historical three-dimensional digital model extracts multidimensional features through a multi-representation model and splices them into a comprehensive feature vector. It should also be noted here that the difference between the "historical three-dimensional digital model" and the above-mentioned "digital model to be retrieved" is shown in Table 1 below:

[0064]

[0065]

[0066] Table 1

[0067] Step S3: concatenate historical feature vectors.

[0068] In step S3, n groups of historical feature vectors of the historical three-dimensional digital model are spliced into a comprehensive historical feature vector.

[0069] In step S3, the system will update the feature vectors of the historical three-dimensional digital model in real time to ensure the timeliness of the data. Then, these n groups of historical feature vectors are spliced into a longer vector, namely the comprehensive historical feature vector, according to a preset fixed order.

[0070] Step S4: Conversion of the feature vector to be retrieved. This step is completed during real-time retrieval. Since there is usually only one digital model to be retrieved, the conversion speed will be relatively fast.

[0071] In step S4, n representation models are used to convert the digital model to be retrieved into n groups of feature vectors to be retrieved.

[0072] Step S5: Summarize the feature vectors to be retrieved.

[0073] In step S5, the n groups of feature vectors to be retrieved of the digital model to be retrieved are spliced into a comprehensive feature vector to be retrieved. This process is similar to step S3, and the splicing order can be synchronized with the above step S3.

[0074] In an optional embodiment, the weight ratios of different representation models can be adjusted to convert them into corresponding dimensional allocations, and after the splicing process of step S3 and step S5, a comprehensive historical feature vector and a comprehensive feature vector to be retrieved spliced according to the weight ratio are output.

[0075] For example, if technicians want the final result to depend 70% on the three-dimensional topology structure, 20% on the three-view projection, and 10% on the text, we can set the ratio of the feature vector length (i.e., dimension) of the three-dimensional topology structure representation model, the three-view projection representation model, and the text representation model to 7:2:1.

[0076] Step S6: Similarity matching and cluster analysis.

[0077] In step S6, similarity matching is performed between the feature vector to be retrieved and the historical feature vector to generate a similarity ranking, and cluster analysis is performed on the historical feature vector to generate a classification result.

[0078] In an optional embodiment, step S6 may generate a search ranking result by calculating the cosine similarity between the comprehensive feature vector to be searched and the comprehensive historical feature vector.

[0079] These features are closely related and interactive, and can be sorted from high to low based on similarity values to generate a list of search results. The comprehensive feature vector to be retrieved and the comprehensive historical feature vector are extracted and concatenated through multiple representation models, containing comprehensive feature information of the three-dimensional digital model. Cosine similarity calculation utilizes these high-dimensional feature vectors to calculate the degree of similarity between them. This method combines the advantages of multiple representation models to more accurately capture the characteristics of the three-dimensional digital model. At the same time, the cosine similarity calculation method can effectively process high-dimensional vectors and is not affected by vector length.

[0080] In this embodiment, FAISS is used to retrieve the cosine similarity between the three-dimensional digital model feature vector to be retrieved and the historical three-dimensional digital model feature vector to generate a ranking, which is a three-dimensional digital model search function.

[0081] In another optional embodiment, step S6 may generate a search ranking result by performing independent similarity matching on n groups of feature vectors to be searched.

[0082] Through this step, technicians can perform similarity matching on the vectors of each representation model separately, and then sort or filter them according to priority. For example, technicians want to cluster texts with a similarity of more than 90%, and then cluster them according to the three-dimensional topology structure.

[0083] For the three-dimensional digital model clustering function, technicians only need to use a clustering algorithm on the historical three-dimensional digital model feature vectors according to the above method to automatically cluster them into several categories. This is not the focus of the present invention and will not be elaborated on here.

[0084] Next, an example is used to vividly illustrate the solution of this application. This application can be implemented in the 3D CAD model library management system of a large manufacturing enterprise. The specific implementation steps are as follows: First, select four representation models: multi-view projection model (dimension d1 = 256), metadata model (dimension d2 = 16), text representation model (dimension d3 = 256) and 3D structure representation model (dimension d4 = 256). For historical 3D digital models, a multi-view projection model is used to generate 2D images of 12 perspectives, and each image is extracted with a 256-dimensional feature vector through pre-trained ResNet50. The metadata model extracts 16 key parameters such as the thickness of the part and the number of mounting holes. The text representation model uses the BERT model to process the name and description of the part to generate a 256-dimensional vector. The 3D structure representation model uses the improved PointNet++ to process point cloud data to generate a 256-dimensional vector. These four sets of feature vectors are spliced into a 784-dimensional comprehensive historical feature vector. The system automatically checks and updates the feature vectors of the historical 3D digital models that have changed at 3 am every day. When a designer needs to retrieve similar parts, the three-dimensional digital model to be retrieved undergoes the same processing flow to generate a final 784-dimensional feature vector to be retrieved. The system uses the cosine similarity algorithm to calculate the similarity between the vector to be retrieved and all historical vectors, and arranges them in descending order of similarity. At the same time, the K-means algorithm (K=50) is used to perform cluster analysis on the historical feature vectors. Finally, the system shows the designer the top 20 results with the highest similarity and marks the cluster category to which each result belongs. Designers can quickly find the most suitable reference model based on this information, greatly improving design efficiency and accuracy.

[0085] The above examples are for illustration only and do not limit the scope of the patent.

[0086] In order to achieve the above-mentioned purpose, the present invention also provides a three-dimensional digital model retrieval and clustering system, which uses the three-dimensional digital model retrieval and clustering method as described in any of the above embodiments, and can include a characterization model selection module, a historical feature vector conversion module, a historical feature vector splicing module, a feature vector to be retrieved conversion module, a feature vector to be retrieved splicing module and a similarity matching and clustering analysis module.

[0087] Specifically, the representation model selection module is used to select n representation models according to the target application scenario and select the dimension of each representation vector according to the performance of the representation model. The historical feature vector conversion module uses n representation models to convert the historical three-dimensional digital model into n groups of historical feature vectors, and updates the historical feature vectors of the changed historical three-dimensional digital model in real time. The historical feature vector splicing module is used to splice the n groups of historical feature vectors of the historical three-dimensional digital model into a comprehensive historical feature vector. The feature vector conversion module to be retrieved uses n representation models to convert the digital model to be retrieved into n groups of feature vectors to be retrieved. The feature vector splicing module to be retrieved splices the n groups of feature vectors to be retrieved of the digital model to be retrieved into a comprehensive feature vector to be retrieved. The similarity matching and cluster analysis module performs similarity matching between the feature vector to be retrieved and the historical feature vector, generates a similarity ranking, performs cluster analysis on the historical feature vectors, and generates a category division result. The specific method has been described in detail above and will not be elaborated on here.

[0088] In order to achieve the above-mentioned purpose, the present invention also provides a computer device, including a memory, a processor and a program stored in the memory and runnable on the processor. When the processor runs the program, it can implement the steps of a three-dimensional digital model retrieval and clustering method as described in any of the aforementioned embodiments.

[0089] The processor and memory can be provided separately or integrated together, for example, integrated into a system-on-chip (SOC) of a terminal device. It should be understood that the processor of the embodiment of the present application may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method embodiment can be completed by an integrated logic circuit of hardware in the processor or by instructions in the form of software. The above processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, or a discrete hardware component. The various methods, steps and logic block diagrams disclosed in the embodiments of the present application can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in conjunction with the embodiments of the present application can be directly embodied as being executed by a hardware decoding processor, or can be executed by a combination of hardware and software modules in the decoding processor. The software module can be located in a storage medium well-known in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, etc. The storage medium is located in the memory, and the processor reads the information in the memory and, in conjunction with its hardware, completes the steps of the above method.

[0090] In order to achieve the above-mentioned purpose, the present invention also provides a computer-readable storage medium, which stores executable instructions or programs. When the executable instructions or programs are processed and executed, the three-dimensional digital model retrieval and clustering method as described in any of the previous embodiments is implemented.

[0091] The readable storage medium is, for example, a memory. The memory may be a volatile memory or a non-volatile memory, or the memory may include both volatile memory and non-volatile memory. The non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0092] If the integrated units in the above embodiments are implemented in the form of software functional units and sold or used as independent products, they can be stored in the above-mentioned readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the existing technology, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes a number of instructions for causing one or more devices (such as personal terminals, clients, or network devices) to execute all or part of the steps of the methods described in various embodiments of the present invention.

[0093] The above describes in detail the preferred specific embodiments of the present invention, which only express several implementation methods of the present invention, but it cannot be understood as limiting the scope of the patent. The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the various technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification. It should be understood that ordinary technology in this field can make many modifications and changes based on the concept of the present invention without creative work. Therefore, without departing from the concept of the present invention, all technical solutions that can be obtained by technicians in this technical field through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of existing technology should be within the scope of protection determined by the claims.

Claims

1. A three-dimensional digital model retrieval and clustering method, characterized in that: The steps include: Step S1: selecting n representation models; wherein the n representation models are selected according to the target application scenario and the dimension of each representation vector is selected according to the performance of the representation model; Step S2: historical feature vector conversion; wherein, the historical three-dimensional digital models are converted into n groups of historical feature vectors using the n representation models, and the historical feature vectors of the changed historical three-dimensional digital models are updated in real time; Step S3: splicing historical feature vectors; wherein n groups of historical feature vectors of the historical three-dimensional digital model are spliced into a comprehensive historical feature vector; Step S4: converting the feature vectors to be retrieved; wherein, the digital model to be retrieved is converted into n groups of feature vectors to be retrieved using the n representation models; Step S5: splicing the feature vectors to be retrieved; wherein, the n groups of feature vectors to be retrieved of the digital model to be retrieved are spliced into a comprehensive feature vector to be retrieved; Step S6: similarity matching and cluster analysis; wherein, by performing similarity matching on the feature vector to be retrieved and the historical feature vector, a similarity ranking is generated, and cluster analysis is performed on the historical feature vector to generate a category classification result.

2. The three-dimensional digital model retrieval and clustering method according to claim 1, characterized in that: In the step S1, the n representation models include at least two of the following: a multi-view projection model, a metadata model, a text representation model, and a three-dimensional structure representation model.

3. The three-dimensional digital model retrieval and clustering method according to claim 2, wherein: The step S1 includes the following steps: Selecting the multi-view projection model to analyze the features of the digital model two-dimensional view, the dimension of the representation vector of the multi-view projection model is set to d1; Selecting the metadata model to analyze features of the digital model metadata, the metadata features including at least one of the following: part thickness, number of mounting holes, and part width, and setting the dimension of the representation vector of the metadata model to d2 according to the number of metadata; Selecting the text representation model to analyze the text features of the digital model, the dimension of the representation vector of the text representation model is set to d3; The three-dimensional structure representation model is selected to analyze the topological structure characteristics of the digital model, and the dimension of the three-dimensional structure representation model is set to d4.

4. The three-dimensional digital model retrieval and clustering method according to claim 3, wherein: d1=256, d2=16, d3=256, d4=256.

5. The three-dimensional digital model retrieval and clustering method according to claim 3, wherein: By adjusting the weight ratios of different representation models to convert them into corresponding dimensional allocations, after the splicing process of step S3 and step S5, the comprehensive historical feature vector and the comprehensive feature vector to be retrieved spliced according to the weight ratio are output.

6. The three-dimensional digital model retrieval and clustering method according to claim 5, characterized in that: In step S6, the cosine similarity between the comprehensive feature vector to be retrieved and the comprehensive historical feature vector is calculated to generate a retrieval ranking result.

7. The three-dimensional digital model retrieval and clustering method according to claim 5, characterized in that: In step S6, the n groups of feature vectors to be retrieved are subjected to independent similarity matching to generate a retrieval ranking result.

8. A three-dimensional digital model retrieval and clustering system, characterized in that: The three-dimensional digital model retrieval and clustering method according to any one of claims 1 to 7 comprises: A representation model selection module is configured to select n representation models according to a target application scenario and to select the dimension of each representation vector according to the performance of the representation model; Historical feature vector conversion module: uses n representation models to convert historical three-dimensional digital models into n groups of historical feature vectors, and updates the historical feature vectors of the changed historical three-dimensional digital models in real time; Historical feature vector splicing module: splices n groups of historical feature vectors of the historical three-dimensional digital model into a comprehensive historical feature vector; Feature vector conversion module to be retrieved: uses n representation models to convert the digital model to be retrieved into n groups of feature vectors to be retrieved; Feature vector splicing module to be retrieved: splicing n groups of feature vectors to be retrieved of the digital model to be retrieved into a comprehensive feature vector to be retrieved; Similarity matching and cluster analysis module: By performing similarity matching between the feature vector to be retrieved and the historical feature vector, a similarity ranking is generated, and cluster analysis is performed on the historical feature vector to generate a category division result.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor runs the program, the three-dimensional digital model retrieval and clustering method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer-executable instructions or a computer program, and when the computer-executable instructions or the computer program are processed and executed, the three-dimensional digital model retrieval and clustering method according to any one of claims 1 to 7 is implemented.

Images