A three-dimensional weaving pattern generation method based on graph theory and neural network

Abstract

The application provides a three-dimensional weaving pattern generation method based on graph theory and neural networks, comprising: acquiring line segment data in a traditional pattern, constructing a topological graph composed of nodes and edges, determining the hierarchical order of multi-level relationships, and obtaining a hierarchical structure model; extracting three-dimensional rendering parameters from the topological graph according to the hierarchical structure model, and generating a diversified pattern skeleton set using a graph partitioning method; obtaining fusion data of the diversified pattern skeleton set and the hierarchical structure model to obtain a pattern image with visual depth; and fusing rendering parameters of the interlaced logical representation data in the pattern image and the hierarchical structure model to generate a reconstructed three-dimensional weaving pattern. Through the synergistic effect of the graph theory algorithm and the neural network, the application realizes efficient conversion from the pattern skeleton to the three-dimensional image, significantly improves the automation level and visual performance of the pattern design, and provides important technical support for the digital protection and innovative application of the traditional pattern.

Description

Technical Field

[0001] This invention relates to the field of weaving information technology, and in particular to a method for generating three-dimensional weaving patterns based on graph theory and neural networks. Background Technology

[0002] In the field of cultural and creative product design, patterns, as core elements of cultural expression and visual aesthetics, carry profound artistic value and market potential. Their design and generation technologies directly impact the product's innovation and appeal. Traditional patterns often originate from hand-weaving techniques, such as brocade patterns from some Chinese ethnic minorities or lace patterns from Europe. These patterns showcase unique three-dimensional layers and cultural connotations through the intricate interweaving of lines. However, how to achieve diversified generation while preserving the topological structure of traditional patterns has become a key issue that urgently needs to be addressed within the industry.

[0003] Existing methods often struggle to balance structural complexity and layering when generating patterns, especially woven patterns. They commonly simplify the interlacing relationships, resulting in patterns lacking depth and three-dimensionality. Many techniques merely replicate or simply deform the surface pattern, ignoring the complex interlacing logic between lines within the pattern, failing to realistically reproduce the layered visual effect of traditional weaving. For example, when generating an antique-style brocade scarf pattern, the repeated interlacing of warp and weft threads at multiple intersections in traditional techniques creates a clear contrast of light and shadow and spatial variation. However, existing methods often simplify these intersections to flat layering, resulting in a flat overall pattern that loses the texture of the fabric naturally casting shadows under light. This limitation makes the designed patterns less vivid in terms of cultural expression and artistic presentation, failing to meet the demands of modern consumers for refined and personalized cultural and creative products, such as patterns for phone cases, clothing, or home decorations. Focusing on the technical challenges, the interlacing relationships of lines in woven patterns are a core hurdle. This relationship is not merely a simple layering, but involves a multi-layered interweaving logic. For example, a line may directly contact other lines on the upper layer, be indirectly covered in the middle layer, and support the overall structure on the lower layer. This logic determines the visual three-dimensional layering of the pattern. If the direct contact relationship of each intersection point and the deeper indirect covering influence cannot be accurately captured and described, it is impossible to reproduce that well-arranged spatial sense when generating new patterns. Taking traditional Miao batik weaving patterns as an example, the batik lines need to simulate the multi-level relief effect formed by the wax layer during the weaving process. However, the current generation ignores the dynamic occlusion between these layers, resulting in inconsistent overlapping lines or gaps when the pattern is magnified. Furthermore, the complexity of this multi-layered relationship is also reflected in the coordination of the overall layering order in local areas. Without a systematic understanding of these orders, the generated patterns often exhibit chaotic lines or contradictory layers. Therefore, systematically defining and reconstructing the interwoven hierarchical structure during the pattern generation process, ensuring that the multi-level relationships at the intersections are accurately expressed, and ultimately presenting a pattern effect with visual depth and a sense of three-dimensional hierarchy, has become a key issue that this study urgently needs to address.

[0004] This issue highlights the technical contradictions in the business. On the one hand, the cultural and creative market demands diversified and innovative patterns to adapt to different product forms. On the other hand, the inherent topology and hierarchical logic of traditional woven patterns require the generation process to strictly adhere to the physical rules of handcrafts. Otherwise, it will lead to the distortion of cultural symbols and a decline in visual appeal, which restricts the efficiency of cultural and creative products from concept to market. Summary of the Invention

[0005] This invention provides a method for generating three-dimensional knitted patterns based on graph theory and neural networks, mainly including: Obtain line segment data from traditional patterns, construct a topology graph consisting of nodes and edges from the initial line segment data, identify and judge the intersection positions in the topology graph, mark the overlapping areas of multiple lines in layers, and determine the hierarchical order of multi-level relationships. Based on the hierarchical order of the multi-level relationship, the topology graph is labeled and processed in layers to obtain a hierarchical structure model. The hierarchical structure model includes enhanced interleaving logic representation data, which contains the up and down interleaving vector description of each intersection point. Based on the hierarchical structure model, the stereo rendering parameters are extracted from the topological graph, and multiple variant pattern skeletons are generated using the graph segmentation method to obtain a diverse set of pattern skeletons. The fusion data of the diverse pattern skeleton set and hierarchical structure model is obtained, and the fusion data is processed by a generative graph convolutional neural network to obtain a pattern image with visual depth. The interweaving logic representation data and hierarchical structure model in the pattern image are fused with rendering parameters to generate a reconstructed three-dimensional woven pattern.

[0006] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention discloses a method for generating 3D woven patterns based on graph theory and neural networks. Addressing the challenges of accurately modeling and rendering complex interlacing logic and multi-level intersections in traditional pattern design, this invention offers a complete solution. By extracting line data from traditional pattern databases and constructing a topological graph to analyze intersections and connections, a graph convolutional neural network is used to hierarchically label and optimize the order of intersections, solving the problem of coordinating multi-level intersections and expressing interlacing logic. Simultaneously, a graph segmentation method is used to generate diverse pattern skeletons, which are then integrated with the hierarchical structure model and rendering parameters to ultimately generate 3D woven patterns with visual depth. This invention achieves efficient conversion from pattern skeletons to 3D images through the synergistic effect of graph theory algorithms and neural networks, significantly improving the automation level and visual expressiveness of pattern design, and providing important technical support for the digital preservation and innovative application of traditional patterns. Attached Figure Description

[0007] Figure 1 This is a flowchart illustrating the overall process of a method for generating three-dimensional woven patterns based on graph theory and neural networks according to the present invention. Figure 2 This is a scene diagram illustrating a method for generating 3D woven patterns based on graph theory and neural networks according to the present invention. Figure 3 This is a schematic diagram of a method for generating three-dimensional woven patterns based on graph theory and neural networks according to the present invention.

[0008] Figure 4This is another schematic diagram of a method for generating three-dimensional weaving patterns based on graph theory and neural networks according to the present invention. Detailed Implementation

[0009] To further understand the content of this invention, a detailed description of the invention is provided in conjunction with the accompanying drawings and embodiments. The specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0010] like Figure 1 , 2 As shown, this embodiment presents a method for generating 3D woven patterns based on graph theory and neural networks, which may specifically include: S101. Obtain line segment data from traditional patterns, construct a topology graph consisting of nodes and edges from the initial line segment data, identify and judge the intersection positions in the topology graph, mark the overlapping areas of multiple lines in layers, and determine the hierarchical order of multi-level relationships. like Figure 3 As shown, it is specifically divided into: 1. Obtain line segment data from traditional patterns, and construct a topological graph consisting of nodes and edges based on the initial line segment data; 1.1 Extract initial line segment data from a pre-set traditional pattern database, and use graph theory algorithms to construct a topological graph composed of nodes and edges to obtain the connection relationship of each line segment; Specifically, line segment data that conforms to a predetermined style is selected from the preset traditional pattern database to generate an initial line segment dataset; each line segment in the initial line segment dataset is mapped to a node, and the edge relationships between nodes are constructed to form the topology graph.

[0011] 1.2 Map each line segment in the initial line segment dataset to a node, construct the edge relationship between nodes, and form the topology graph. For the connection relationship, obtain the preliminary intersection point position and determine the intersection point coordinate value. 1.3. Based on the coordinate values ​​of the intersection points, determine the matching degree of the pattern reconstruction. When the matching degree is lower than a predetermined standard, adjust the endpoint positions of the initial line segment data to form an optimized connection relationship. 1.4. By optimizing the connection relationships and integrating the pattern analysis method, the final intersection point position is determined, and the final topology map is obtained.

[0012] Initial line segments are extracted from a pre-defined traditional pattern library. A graph theory algorithm is used to construct a topological graph consisting of nodes and edges to obtain the connection relationships of each line segment. For these connection relationships, preliminary intersection point positions are obtained, and their coordinate values ​​are determined. Based on the intersection point coordinate values, the pattern reconstruction matching degree is judged. If the matching degree is lower than a preset threshold, the endpoints of the line segments are adjusted to obtain optimized connection relationships. Through these optimized connection relationships, combined with pattern analysis methods, the final intersection point positions are determined.

[0013] The step of determining the final intersection point position by optimizing the connection relationship and integrating pattern analysis includes: A. Based on the optimized connection relationship, extract the intersection features of line segments, and calculate the position distribution of intersection points by combining the pattern analysis logic; B. Based on the location distribution, adjust the coordinate values ​​of the intersection points to generate the final intersection point positions.

[0014] Specifically, in one implementation, initial line segment data is extracted from a pre-defined traditional pattern database. This process first requires querying and filtering the database. Traditional pattern databases typically store digitized images of various historical patterns, such as traditional cloud patterns or meander patterns. These images are saved in a processable format through scanning or vectorization. The extraction process involves edge detection of the image to identify the line contours that constitute the pattern, and then segmenting these contours into independent line segments. Each line segment is defined as a continuous path from one endpoint to another, avoiding the complexity of the overall image. In this way, initial line segment data is obtained, providing a foundation for subsequent topological analysis.

[0015] Furthermore, constructing a topological graph consisting of nodes and edges using graph theory algorithms is the core step in the entire process. Graph theory algorithms are used here to model the connection relationships of lines, where nodes represent the endpoints or potential intersections of line segments, and edges correspond to the line segments themselves.

[0016] For example, all extracted line segments are first traversed, the coordinates of the two endpoints of each segment are identified, and the distance between the endpoints is checked to determine whether a connection is made. If the endpoints of two line segments coincide or are close enough, they are connected as a node. This construction process ensures the undirected graph property of the topology graph, where the node degree represents the number of connected lines. By calculating all possible connections, a preliminary topology is obtained. For example, when dealing with a complex phoenix pattern, multiple line segments may form a high-density node in the central region.

[0017] It should be noted that obtaining the connectivity of each line segment involves a further graph traversal algorithm. A depth-first search or breadth-first search is used to traverse the topology graph, starting from a node and recording the connected edges and nodes, thus establishing an adjacency list between line segments.

[0018] Exemplarily, in a possible implementation, for a geometric pattern extracted from a database, the algorithm marks the adjacent segments of each line segment to generate an adjacency matrix, where the matrix elements represent whether there is a direct connection. This relationship facilitates subsequent pattern editing or deformation and avoids breaks.

[0019] In one embodiment, the determination of the preliminary intersection point positions is based on the calculation of the positions of the nodes. An intersection point is usually a point where multiple line segments meet and is represented as a node with a degree greater than 2 in the graph theory model.

[0020] Preferably, by calculating the intersection coordinates of the line segments and using geometric algorithms such as the line segment intersection formula, these positions are verified and refined.

[0021] For example, for the line data extracted from a traditional brocade pattern, the algorithm first assumes the endpoints as potential intersection points and then iteratively checks whether all line pairs intersect. If they intersect, new nodes are added and the edges are updated. This method improves the accuracy of the intersection points and effectively supports the digital reconstruction of patterns in practical applications.

[0022] It can be understood that the generality of this process is reflected in the processing of different traditional patterns.

[0023] For example, when processing the hui character pattern with more linear elements, the connection relationship is mainly a chain structure and there are fewer intersection points; while in the cloud thunder pattern, the intersection points are dense and more iterative calculations are required. Through these scenarios, the technical solution demonstrates flexible applications in the field of cultural patterns. Further, after constructing the topological graph, the obtained connection relationship and preliminary intersection point positions can be used to optimize the pattern design.

[0024] Specifically, this topological information allows users to modify the lines without destroying the overall connectivity. For example, when adjusting a line segment in a digital design software, the positions of the connected segments are automatically updated, thus maintaining the integrity of the pattern. This implementation brings an improvement in the efficiency of pattern processing and has practical value in the field of traditional art digitization.

[0025] In one embodiment, data extraction considering the influence of noise is taken into account. Traditional pattern images may have scanning noise, so before extracting the line segments, filtering processing such as median filtering is performed to smooth the edges. Then when constructing the topological graph, a tolerance threshold is introduced to allow endpoints with slight deviations to be regarded as connection points.

[0026] For example, for a flower and bird pattern in an ancient book, noise may cause line breaks, and the algorithm merges these breakpoints through the threshold to form a complete topological graph.

[0027] Preferably, the entire process can be integrated into a software system. After the user inputs a pattern image, the system automatically performs the extraction and construction steps. The resulting topology graph is output in a visual form, with nodes and edges marked with different colors to facilitate the analysis of connections and intersections. In another implementation, for large pattern databases, parallel computing is used to accelerate graph theory algorithms.

[0028] For example, line segments can be grouped for processing, with each group independently constructing a subgraph, and then merged into a complete topology graph. This approach is suitable for scenarios involving the simultaneous analysis of multiple patterns, improving efficiency. Through the above steps, this technical solution achieves accurate modeling of line structures in the traditional pattern field, supporting further applications such as pattern generation or preservation.

[0029] 2. Identify and determine the intersection positions in the topology graph, mark the overlapping areas of multiple lines in layers, and determine the hierarchical order of the multi-level relationship; Based on the intersection positions in the topology diagram, the interlacing sequence information of adjacent lines is obtained. If more than two lines overlap in the sequence, multi-level intersections are marked, and the hierarchical order of multi-level relationships is determined.

[0030] Specifically; 2.1. Based on the intersection positions in the topology diagram, obtain the interlacing sequence information of adjacent lines and determine the number of sequence superpositions; 2.2 Determine the number of sequence overlaps. If it exceeds two, mark it as a multi-level crossover to obtain a multi-level crossover identifier. 2.3 Obtain the relative positions of the overlapping lines at the multi-level intersection markers to determine the hierarchical order of the multi-level relationship; 2.4. Based on the hierarchical order, extract the overlapping segments between adjacent lines, determine the ratio of the length of the overlapping segment to the distance of the intersection point, and obtain the hierarchical order of the multi-level relationship.

[0031] The intersection points are obtained from the topology map. Interlacing sequence information is extracted from adjacent lines to determine the number of overlapping sequences. If the number of overlapping sequences exceeds two, a multi-level intersection is marked, resulting in a multi-level intersection identifier. For each multi-level intersection identifier, the relative positions of the overlapping lines are obtained to determine the hierarchical order of the multi-level relationship. Based on this hierarchical order, overlapping segments between adjacent lines are extracted, and the ratio of the overlapping segment length to the intersection point distance is determined to obtain the hierarchical order of the multi-level relationship.

[0032] Specifically, in one implementation, the topology map is first preprocessed to identify the locations of intersections.

[0033] Specifically, the coordinates of intersecting lines are detected by scanning the pixel or vector data of the topology map.

[0034] For example, geometric algorithms are used to calculate the intersections between lines, ensuring that the positional information of each intersection includes its horizontal and vertical coordinates and the relevant line identifier. This preprocessing helps in the extraction of subsequent sequence information and avoids erroneous judgments caused by positional deviations. Furthermore, based on the identified intersection positions, the interweaving sequence information of adjacent lines is obtained.

[0035] In one possible implementation, starting from the intersection point, adjacent segments are traversed along the line direction, recording the line order and interlacing pattern. The generated interlacing sequence is checked, and if three or more lines overlap at the same location or in adjacent sequences, it is marked as a multi-level intersection.

[0036] For example, in a circuit topology diagram, if multiple wires overlap at a point more than twice, the system automatically marks that point and records the number of overlaps. This judgment is based on a threshold comparison to ensure accurate differentiation between simple intersections and multi-level cases. Preferably, after marking multi-level intersections, the hierarchical order of the multi-level relationships is determined.

[0037] In one embodiment, the order is determined by analyzing the stacking order in the sequence, starting from the bottommost lines. For example, a stack structure can be used to simulate the hierarchy, where line A is below B, and B is below C, resulting in a hierarchy order of CBA. This determination process takes into account the relative positions and interlacing depth of the lines, providing a clear hierarchical representation.

[0038] For example, in the application of network topology diagrams, assuming that the intersection involves multiple interwoven data lines, the system first obtains the sequence such as line 1, line 2, and line 3 overlapping, then determines that it is multi-level, and sorts it so that line 3 is on top and line 1 is below. This approach demonstrates the versatility of the technical solution within the same domain and can handle topology structures of varying complexity.

[0039] Understandably, this method can be extended to various scenarios, such as handling signal line crossings in circuit board design. Through the above steps, precise management of multi-level relationships is achieved, improving the efficiency of topology graph analysis. In another implementation, an auxiliary data structure is introduced to optimize sequence acquisition.

[0040] For example, graph theory models can be used to represent topological graphs, where nodes are intersections and edges are line segments, thus efficiently extracting interlaced sequences. This structure helps in processing large-scale topological graphs while avoiding computational overhead.

[0041] Specifically, for multi-level overlapping labels, the system can combine threshold and pattern matching.

[0042] For example, if more than two lines overlap consecutively in a sequence, they are marked and a report is generated, detailing the hierarchical position of each line. This detailed processing ensures the reliability of the judgment. Furthermore, the hierarchical order can be determined through an iterative algorithm, traversing from the start to the end of the sequence and constructing a sequential list layer by layer.

[0043] For example, in a sequence containing four overlapping lines, the algorithm will gradually peel off the top layer to determine the order from the bottom layer to the top layer, thus forming a complete hierarchical description.

[0044] In one embodiment, the application effect of this technical solution is reflected in improving the readability of the topology diagram and the error detection capability, such as identifying potential multi-level crossover problems early in the design process and avoiding subsequent modifications.

[0045] S102. According to the hierarchical order of the multi-level relationship, the topology graph is labeled and processed in layers to obtain a hierarchical structure model. The hierarchical structure model includes enhanced interleaving logic representation data, which contains the up and down interleaving vector description of each intersection point.

[0046] like Figure 4 As shown, the process of hierarchically labeling and processing the topology graph according to the hierarchical order of multi-level relationships to obtain a hierarchical structure model includes: 1.1 According to the hierarchical order, label the hierarchical relationship of the intersection points in the topology graph, generate hierarchical labeling data, and map the hierarchical labeling data to the topology graph to obtain labeled graph data; 1.2. The labeled graph data is processed by a graph convolutional neural network to obtain interleaved logic representation data. Multi-level intersection data is extracted from the interleaved logic representation, and the coverage vector of each multi-level intersection is calculated to form a coverage vector distribution matrix. 1.3. Based on the coverage vector distribution matrix, extract the vector data of adjacent intersection points and calculate the difference between the two vectors; when the difference exceeds a predetermined standard, adjust the local hierarchical order to generate adjusted hierarchical distribution data; 1.4. Using the adjusted hierarchical distribution data, analyze the dependencies between levels and construct a hierarchical framework; 1.5 Perform connection stability analysis on the hierarchical framework. When the stability is lower than the predetermined standard, decompose the hierarchical units with higher nesting depth and generate the decomposed hierarchical data. 1.6 Based on the decomposed hierarchical data, recalculate the matching degree between the covering vector and the adjacent intersection vector, optimize the vector data distribution results, and construct the coordinated hierarchical structure model.

[0047] The hierarchical order is obtained, and the topology map is labeled hierarchically to obtain labeled map data. The labeled map data is processed using a graph convolutional neural network (PCNN), where the PCNN input is the labeled map data and the output is an interleaved logic representation, and the interleaved logic representation data is extracted. For the interleaved logic representation data, intersection vectors are determined, and the interleaved descriptions are fused to obtain a fused vector description. Based on the fused vector description, line curvature adjustment is performed, where the line curvature adjustment is obtained by calculating the ratio of line curvature to intersection distance, and a multidimensional projection mapping is obtained. The multidimensional projection mapping determines the mapping coordinates by projecting the adjustment value onto a multidimensional space, resulting in the extracted vector description. From the extracted vector description, a neural network is fused, where the neural network input is the extracted vector description and the output is an enhanced logic representation, to determine the enhanced logic representation data.

[0048] Specifically, in one implementation, the topology graph is labeled in layers based on a previously determined hierarchical order.

[0049] Specifically, the topology is first represented as a graph structure, where nodes correspond to intersections and edges correspond to line segments. Then, hierarchical labels are assigned to the lines associated with each intersection in hierarchical order.

[0050] For example, in a circuit topology diagram, if the hierarchy is line A at the bottom layer, line B in the middle layer, and line C at the top layer, the labeling process involves traversing the graph nodes and adding a numerical label to each line, such as 1 for the bottom layer, 2 for the middle layer, and 3 for the top layer. This labeling ensures the visualization of the interweaving relationships, which is helpful for subsequent processing.

[0051] It should be noted that the hierarchical annotations can be stored in matrix form, with the annotation matrix at each intersection recording the relative position of the lines, thus forming the annotated graph data. This method is also applicable to network topology graphs, for example, when dealing with interwoven data lines, the annotations highlight the order of multi-level stacking. Furthermore, the annotated graph data is processed using a graph convolutional neural network. A graph convolutional neural network is a graph neural network model used to process non-Euclidean structured data; its principle is to capture the dependencies between nodes through neighborhood aggregation.

[0052] Specifically, in this embodiment, the labeled topology graph is input into a graph convolutional neural network, and each convolutional operation aggregates the features of adjacent nodes, including hierarchical labels and positional information.

[0053] For example, the first convolutional layer of the network calculates the embedding vector of each intersection node, and the feature propagation is achieved by weighted summation of the hierarchical values ​​of neighboring nodes. This process gradually extracts high-order interleaved features, avoiding the neglect of complex topologies by traditional methods.

[0054] In one possible implementation, the graph convolutional neural network structure includes multiple convolutional layers and activation functions, such as the ReLU function, for nonlinear transformation. For applications involving circuit topology graphs, after the model receives labeled data, iterative aggregation is used to generate hidden representations for each node, capturing the vertical relationships between lines.

[0055] It should be noted that the network is trained based on supervised signals from labeled graphs, such as using cross-entropy loss to optimize parameters, ensuring that the output reflects the true interleaving logic. This detailed processing highlights the model's robustness in handling labeled data, enabling it to adapt to topologies of varying sizes.

[0056] Preferably, an enhanced interleaved logic representation is obtained, which includes an up-and-down interleaved vector description for each intersection point. The enhanced interleaved logic representation refers to the feature representation output by the graph convolutional neural network, which integrates the original topological information and hierarchical annotations to form a richer logical description.

[0057] Specifically, the vector is described as a multi-dimensional vector, with each dimension encoding the relative interweaving relationship of the lines at the intersection points. For example, the vector components of the upper layer lines are higher than those of the lower layer lines. Through the final layer of the network, each intersection point obtains a fixed-length vector, such as 128-dimensional, which records the quantized values ​​of the upper and lower order.

[0058] In one embodiment, for multi-level intersections in the network topology graph, the vector description may be represented as [0.8, 0.5, 0.2], corresponding to the intersection probability distribution of the three lines. This representation enhances the accuracy of the logical expression, facilitating downstream tasks such as path optimization.

[0059] Specifically, the generation process of the interleaved vector description involves the mapping of the network's output layer. The hidden representations are converted into vectors through fully connected layers, ensuring that the description at each intersection captures the global interleaving pattern.

[0060] In one embodiment, such vectors can be used for subsequent analysis, such as evaluating the effects of line overlap in circuit design. Furthermore, in applications of network topology graphs, enhanced representations aid in flow simulation, and vector descriptions provide a quantitative basis for interleaving relationships.

[0061] In one embodiment, the method is extended to real-time topology analysis, combining annotation and network processing to form an end-to-end process. It should be noted that this approach enables a deeper understanding of the interleaving logic, improving the overall parsing capability of the topology graph.

[0062] For each multi-level intersection point in the interleaved logic representation, a coverage vector is calculated. If the difference between the coverage vector and the adjacent intersection point vector exceeds a preset threshold, the local hierarchical order is adjusted to obtain a coordinated hierarchical structure model.

[0063] By processing the multi-level intersection data in the interleaved logic representation, a spatial vector analysis algorithm is used to calculate the coverage vector of each intersection, resulting in a distribution matrix of the coverage vectors. Based on the distribution matrix of the coverage vectors, vector data of adjacent intersections are extracted, and the difference between the two vectors is analyzed using a distance calculation method to determine the magnitude of the difference. If the difference exceeds a preset threshold, the local hierarchical order is reordered using an adjustment algorithm to obtain adjusted hierarchical distribution data. Based on the adjusted hierarchical distribution data, combined with quantitative indicators of the number of loops and nesting depth, a preliminary hierarchical structure framework is constructed using a hierarchical parsing method to obtain an initial structural model. By validating the interleaved connections in the initial structural model, a graph theory algorithm is used to analyze the dependencies between each level to determine the connection stability of the structural model. If the connection stability is lower than a preset standard, a recursive splitting method is used to decompose the loop levels with high nesting depth to obtain the decomposed hierarchical unit data. Based on the decomposed hierarchical unit data, vector fusion technology is used to recalculate the matching degree between the coverage vector and adjacent intersections, resulting in an optimized vector distribution result. By combining the optimized vector distribution results with the deep structural representation vectors of the original patterns, a coordinated hierarchical structure model is constructed using a data mapping method to determine the final structural framework. Based on the final structural framework, cultural symbol element data is integrated, and a loop reconstruction algorithm is used to generate pattern variants with consistent structure, resulting in cultural and creative pattern data with innovative expression.

[0064] Specifically, assuming a topology diagram contains multiple intersections, the coverage vector of each intersection can be constructed using the hierarchical values ​​and spatial coordinates of its adjacent lines. For example, in a circuit topology diagram, the coverage vector of an intersection might be represented as a three-dimensional vector, numerically [2.5, 1.8, 3.2], corresponding to the hierarchical influence range in three directions, respectively. This approach is helpful for subsequent analysis of the interactions between intersections.

[0065] For example, when extracting vector data from adjacent intersections and analyzing their differences, the Euclidean distance method can be used to measure the difference between the two vectors. Assuming the coverage vectors of two adjacent intersections are [2.5, 1.8, 3.2] and [2.0, 1.5, 2.8], the distance calculated is 0.7. If the preset threshold is 0.5, the local hierarchical order needs to be adjusted. The adjustment algorithm can be based on the redistribution of hierarchical labels, for example, increasing the hierarchical value of an intersection from 2 to 3 to reduce potential conflicts caused by the differences.

[0066] S103. Extract stereo rendering parameters from the topology graph according to the hierarchical structure model, and generate multiple variant pattern skeletons using graph segmentation method to obtain a diverse set of pattern skeletons.

[0067] Specifically, it includes: 1. Obtain the topology graph from the hierarchical model, extract the stereo rendering parameters, and obtain the initial pattern framework; 2. The initial pattern framework is segmented using a graph theory algorithm to generate multiple variant pattern skeletons, the skeleton variant sequence is determined, and diverse elements are integrated through adaptive adjustment of the skeleton pattern to obtain an extended pattern set; 3. If the extended pattern set meets the preset diversity threshold, its completeness is determined to obtain the optimized pattern skeleton group; 4. Integrate the optimized pattern skeleton group to determine a diverse set of pattern skeletons.

[0068] A topology graph is obtained from the coordinated hierarchical model, and 3D rendering parameters are extracted to obtain an initial pattern framework. A graph theory algorithm is used to segment the initial pattern framework, generating multiple variant pattern skeletons and determining the skeleton variant sequence. For the skeleton variant sequence, diverse elements are integrated through adaptive adjustment of the skeleton patterns to obtain an extended pattern set. If the extended pattern set meets a preset diversity threshold, its completeness is judged to obtain an optimized pattern skeleton group. Based on the optimized pattern skeleton group, relevant business attributes are integrated to determine a diverse pattern skeleton set.

[0069] Specifically, in one implementation, the coordinated hierarchical model refers to the initial hierarchical model that is adjusted by a preset algorithm to adapt to the structural characteristics of a specific topological graph.

[0070] Specifically, the model comprises multiple layers, each corresponding to a different level of abstraction in the topological graph. For example, the bottom layer represents basic nodes and edges, the middle layer represents subgraph structures, and the top layer represents the overall topological relationships. Through this coordination, the model can capture the geometric and connectivity properties of the topological graph, thus providing a foundation for subsequent parameter extraction.

[0071] It should be noted that the reconciliation process involves iterative optimization, such as using the least squares method to adjust the hierarchical weights to ensure that the matching degree between the model and the topology graph reaches a preset threshold. This method is widely used in the field of pattern design and can handle the topological representation of complex patterns. Furthermore, when extracting 3D rendering parameters from the reconciled hierarchical model, the key nodes and edges of the topology graph are first identified.

[0072] For example, the topology graph can be an undirected graph representing pattern elements, where nodes represent pattern intersections and edges represent connecting line segments. The extraction process includes traversing each level of the model and calculating the 3D coordinates and rendering attributes of the nodes, such as lighting coefficients and material parameters.

[0073] Specifically, for a given node, its stereo rendering parameters can be obtained by calculating its projected position in three-dimensional space, for example, mapping two-dimensional topological coordinates to three-dimensional space, taking into account depth values ​​and viewpoint transformations. This extraction ensures the accuracy of the parameters and supports subsequent pattern variant generation. In digital art creation scenarios, this parameter extraction enables the transformation from planar topology to stereoscopic effects, providing diverse visual expression forms.

[0074] Preferably, when generating multiple variant pattern skeletons using graph theory algorithms for graph segmentation, a spectral segmentation algorithm is used to partition the topological graph. Graph segmentation refers to dividing a graph into multiple subgraphs, each subgraph corresponding to a pattern variant. The specific process includes constructing the Laplacian matrix of the graph, then solving for its eigenvalues ​​and eigenvectors, and achieving segmentation by minimizing the cutting cost function.

[0075] For example, in one embodiment, for the input topology graph, the degree matrix and adjacency matrix of the nodes are first calculated to generate the Laplacian matrix. Then, k-means clustering is applied to group the feature vectors, forming a set of subgraphs. This method can generate multiple independent pattern skeleton variants, each retaining the core structure of the original topology but introducing local variations, such as edge length adjustments or node relocation. In pattern design, this segmentation helps create a series of patterns for fabric or decorative production, ensuring the diversity and consistency of variants. This algorithmic approach makes the generation process efficient and avoids inconsistencies caused by manual intervention.

[0076] In one possible implementation, after generating the variant pattern skeleton, the skeleton's geometric properties need further optimization. The skeleton is a simplified framework resembling a fingerprint, composed of key line segments and intersections. Optimization steps include smoothing processes, such as fitting skeleton edges with Bezier curves to improve visual fluidity.

[0077] Specifically, when determining a diverse set of pattern skeletons, the set is constructed by evaluating the similarity and dissimilarity between variants. Similarity can be calculated based on graph isomorphism checks, while dissimilarity is derived by comparing the number of edges and node distribution of the skeleton.

[0078] In one embodiment, the generated variants are first sorted and filtered according to diversity metrics such as Shannon entropy to form a final set. This determination process ensures that the set covers a wide range of pattern styles and avoids repetition. In the field of pattern design, this set can be used for product customization, providing users with multiple choices, thereby improving design flexibility and market applicability. Furthermore, in another embodiment, the coordination of the hierarchical model can be adjusted in conjunction with user input parameters.

[0079] It should be noted that the parameter extraction module and the model coordination module are tightly integrated and seamlessly connected through data flow, ensuring the continuity of the entire process.

[0080] For example, graph segmentation methods can employ a multi-level segmentation strategy, first performing coarse segmentation to generate preliminary variants, and then refining the segmentation to generate more detailed variants.

[0081] Specifically, the coarse segmentation uses the NormalizedCut algorithm, while the refinement employs a random walk method. This multi-level approach enhances the diversity of variant generation, making it suitable for generating seasonal pattern series, such as spring, summer, autumn, and winter themed patterns. Through this strategy, the number of skeleton variants not only increases but also ensures quality, meeting high-end customization needs. In one implementation, the diversification of the pattern skeleton set can be achieved by introducing random perturbations.

[0082] Preferably, the overall technical solution improves the efficiency and innovation of pattern design. By extracting parameters from the topology diagram and generating a set, the system can automatically produce diverse outputs, reducing the manual workload of designers in practical applications and providing richer creative options. Furthermore, in a specific scenario, the parameters extracted after coordinating the model of a traditional ethnic pattern topology diagram can be used for virtual reality rendering to generate an immersive pattern experience. This extension demonstrates the versatility of the solution, supporting the transformation from static design to dynamic application.

[0083] S104. Obtain the fusion data of the diverse pattern skeleton set and hierarchical structure model, and process the fusion data through a generative graph convolutional neural network to obtain a pattern image with visual depth.

[0084] Specifically, it includes: 1. Obtain the feature vector of the preset diverse pattern skeleton set, and obtain the fused data based on the feature vector and the node data of the hierarchical structure model; By using a pre-defined set of diverse pattern skeletons, feature vectors of the pattern skeleton set are obtained. These feature vectors are then combined with node data from a hierarchical model to obtain fused data. This fused data is processed using a generative graph convolutional neural network to determine pattern style transfer parameters.

[0085] 2. The fused data is processed by a generative graph convolutional neural network to obtain pattern style transfer parameters. If the node weights exceed a preset threshold, an enhanced depth feature map is obtained. The enhanced depth feature map is then mapped to the image space to obtain a pattern image with visual depth.

[0086] The node weights of the fused data are adjusted by the pattern style transfer parameters. If the node weights exceed a preset threshold, an enhanced depth feature map is obtained. This enhanced depth feature map is then mapped to the image space to obtain a pattern image with visual depth.

[0087] Specifically, in one implementation, it is first necessary to acquire fused data of a diverse set of pattern skeletons and a hierarchical structure model. The diverse set of pattern skeletons refers to the basic structural framework extracted from various pattern designs; for example, in digital art design, it refers to a set of lines and nodes extracted from geometric shapes, floral patterns, or abstract textures. These skeletons can be obtained from the original image using edge detection algorithms, ensuring the set covers a variety of styles to enhance the diversity of the generated data. The hierarchical structure model is a framework representing the hierarchical relationships within a pattern, such as dividing the pattern into foreground, middle, and background layers, each containing different density and complexity information. By fusing the skeleton set with the hierarchical model, comprehensive data can be generated for subsequent processing. This fused data can capture the multi-dimensional features of the pattern, providing a foundation for generating visual depth.

[0088] Specifically, the process of acquiring fused data includes several key steps.

[0089] For example, the pattern skeleton set is first preprocessed, such as normalized to unify the scale, and then aligned with the hierarchical structure model. The hierarchical structure model can be understood as a tree structure, where the root node represents the overall pattern and the child nodes represent the details of each layer.

[0090] In one possible implementation, skeleton data is embedded into nodes of a hierarchical model through matrix transformations to form fused data. This method is widely used in artistic pattern generation, such as for creating wallpaper patterns or textile designs, ensuring that the spatial hierarchy of the pattern is preserved after data fusion. Furthermore, the fused data is processed using a generative graph convolutional neural network. A generative graph convolutional neural network is a graph-based generative model that treats fused data as a set of graph nodes and edges, where nodes correspond to skeleton elements and edges represent hierarchical relationships. This network captures local and global features through convolutional operations, achieving generative transformations of the data. In the field of digital art, such networks are often used to enhance the depth effect of images.

[0091] It should be noted that the network processing includes an encoding stage and a decoding stage. During encoding, feature vectors are extracted from the fused data, and during decoding, new image data is generated. This process can transform planar patterns into images with a three-dimensional feel.

[0092] In one embodiment, the specific implementation of the generative graph convolutional neural network can employ a design with multiple graph convolutional layers.

[0093] Preferably, the first convolutional layer processes node features, calculates the neighborhood aggregation value for each node, and then performs a non-linear transformation using an activation function such as ReLU. Subsequent layers gradually extend to the global graph structure, ensuring that the depth information of the hierarchical model is incorporated.

[0094] For example, when processing fused data, the network iteratively optimizes its parameters, minimizing the difference in depth between the generated image and the target image using a loss function such as mean squared error. This approach excels in pattern design scenarios, such as generating decorative patterns with shadows and perspective, achieving more realistic visual effects without introducing additional complexity.

[0095] Understandably, after processing and fusing the data, a pattern image with visual depth is obtained. Visual depth refers to the three-dimensional effect created in the image through lighting, perspective, or layering effects, such as adding gradient shadows or multiple layers to simulate depth in the pattern. Based on the aforementioned network output, this image retains the skeletal diversity of the original pattern while enhancing its sense of layering. In the field of art generation, this image can be applied to texture rendering in virtual reality scenes.

[0096] In one embodiment, the process of generating visual depth is further refined. After network processing, the image is refined through post-processing steps such as depth mapping adjustment.

[0097] Specifically, the feature map output by the network is converted into a depth map, where each pixel is assigned a depth value, weighted according to the structure of the hierarchical model. This adjustment ensures that the image presents a consistent depth effect under different viewing angles.

[0098] For example, when designing mural patterns, this method can create a transition from two-dimensional to three-dimensional, enhancing the user's immersive experience. The technical goal of this process is to improve the realism and applicability of pattern images through precise depth calculations, bringing more efficient artistic creation tools to the business.

[0099] Preferably, in another embodiment, data processing for the fusion of different pattern types is considered.

[0100] For example, for geometric patterns, the skeleton set emphasizes straight lines and angles, while the hierarchical model handles symmetry layer by layer; for organic patterns such as leaf veins, the fused data focuses on curved branches. This data is processed into depth images using a generative graph convolutional neural network, which adaptively adjusts the convolutional kernel size to match the pattern complexity. This diversified processing demonstrates the versatility of the technical solution, covering multiple scenarios within the same art and design field. Furthermore, the embodiment also includes verification of the generated pattern images.

[0101] It should be noted that the accuracy of visual depth is ensured by comparing the similarity index between the original fused data and the output image.

[0102] For example, structural similarity metrics can be used to evaluate depth performance. This validation step helps optimize network parameters, providing reliable output in practical applications such as digital print design.

[0103] One possible implementation extends to dynamic pattern generation. The fused data can contain time-series information, and temporal convolutional layers are introduced during network processing to generate images with animated depth.

[0104] For example, in interactive art installations, such images can change depth in response to user input, enhancing interactivity. This approach maintains the core of the technical solution while adapting to the expanding needs of the art field. Finally, based on the above process, the resulting pattern image achieves efficient depth enhancement in art design. Through the combination of fusion and network processing, this method can handle complex patterns, provide high-quality output, and support rapid iteration from concept to finished product in business applications.

[0105] S107. The interlacing logic representation data and hierarchical structure model in the pattern image are fused with rendering parameters to generate the reconstructed three-dimensional woven pattern.

[0106] Specifically, it includes: 1. Extract the interlacing logic representation data from the pattern image, and combine it with the hierarchical structure model to generate a rendering parameter fusion result; 2. Based on the rendering parameter fusion result, adjust the matching degree between the curvature vector and the interlacing point to generate the reconstructed three-dimensional woven pattern.

[0107] The pattern curvature is corrected on the rendering parameter fusion result. By adjusting the matching degree between the curvature vector and the interlacing point, three-dimensional weaving simulation data is obtained. The continuity of the three-dimensional weaving simulation data is compared to determine the reconstruction completeness; if the reconstruction completeness meets a preset threshold, the reconstructed three-dimensional weaving pattern is generated.

[0108] First, an interlacing logic representation is obtained from the pattern image to determine the hierarchical structure model. Based on this hierarchical structure model, a graph theory algorithm is used to obtain the rendering parameter fusion result. Pattern curvature correction is then performed on the rendering parameter fusion result, and 3D weaving simulation data is obtained by adjusting the matching degree between the curvature vector and the interlacing points. Weaving patterns are extracted from the 3D weaving simulation data, and the reconstruction completeness is determined by comparing the pattern continuity. If the reconstruction completeness meets a preset threshold, a reconstructed 3D weaving pattern is generated and output.

[0109] Specifically, in one implementation, the pattern image is first preprocessed to extract the interleaved logic representation.

[0110] Specifically, pattern images are typically derived from scans or digital captures in the textile design field, such as photographs of cotton fabric patterns. Image segmentation techniques are used to break down the pattern into basic units, such as yarn intersections. The interlacing logic representation refers to the rule-based description of these intersections, such as the positional relationship of upper yarns pressing down on lower yarns. This representation can be in matrix form, where rows and columns correspond to different yarns, and element values ​​represent the interlacing type, such as "1" for upper yarn pressing down and "0" for lower yarn pressing up.

[0111] It should be noted that this logic facilitates subsequent model building, ensuring the structural accuracy of the pattern. In textile design, the goal of this step is to achieve digital reproduction of the pattern, enabling more efficient pattern modification. Furthermore, based on the interlacing logic representation, a hierarchical structure model is established. This model divides the pattern into multiple layers; for example, the bottom layer represents individual yarn paths, the middle layer represents local interlacing modules, and the top layer represents the overall pattern structure.

[0112] For example.

[0113] In one possible implementation, the bottom layer generates a path map by tracing yarn paths, the middle layer aggregates adjacent interlacing points to form modules, and the top layer integrates all modules to form a complete model. This hierarchical structure is helpful for handling complex patterns, such as the multi-layered design of jacquard fabrics. The relationships between layers are reflected through dependencies; for example, middle-layer modules depend on the continuity of the bottom-layer paths. In short, this model provides a scalable framework in the textile field, facilitating the adaptation to different patterns.

[0114] Preferably, fusing rendering parameters using graph theory algorithms is the core step. Graph theory algorithms here transform the hierarchical model into a graph structure, where nodes represent interlacing points and edges represent thread connections. Blending rendering parameters involves incorporating parameters such as lighting and materials into the graph, for example, using weighted edges to represent rendering intensity. The detailed process includes first constructing an undirected graph, then using a graph traversal algorithm such as depth-first search to traverse the nodes and calculate the rendering value for each node.

[0115] Specifically, rendering parameters such as shadow coefficients are calculated using the average value of adjacent nodes to ensure a smooth transition in the 3D effect. This fusion innovatively solves the 2D to 3D conversion problem in textile pattern reconstruction, enabling a more realistic visual output.

[0116] In one embodiment, for silk patterns, the algorithm adjusts parameters to simulate a glossy effect. The process begins with graph construction and gradually merges parameters until complete rendering data is generated. The key to this step lies in the iterative optimization of the algorithm, such as multiple iterations to refine parameter values ​​and avoid rendering distortion. In this way, the technical solution enhances the three-dimensional representation of patterns in textile design and is suitable for the reconstruction of various patterns.

[0117] For example, in another implementation, for the reconstruction of linen patterns, graph theory algorithms first map the hierarchical model into a directed graph, with directions representing yarn flow. Then, when fusing rendering parameters, texture mapping is introduced, a process that includes calculating the rendering weight of each edge based on yarn thickness. This approach can be said to extend the versatility of the solution. Based on the above steps, a reconstructed 3D woven pattern is further generated as an output. The output process involves converting the fused graph data into a 3D model, for example, using a rendering engine to generate a visualization image.

[0118] In one embodiment, the output is a digital file that can be used to 3D print textile prototypes.

[0119] Specifically, coordinate data is extracted from the graph, and a transformation matrix is ​​applied to generate a 3D view, ensuring the fidelity of the interlacing logic. In the textile industry, this output supports design verification and can reduce the trial-and-error costs of actual weaving.

[0120] In one possible implementation, for woolen patterns, the hierarchical model emphasizes the design of the insulation layer, and the graph theory algorithm prioritizes thickness rendering when fusing parameters. The process includes layering the model and then calculating volume parameters to ensure the three-dimensionality of the output pattern.

[0121] It should be noted that this implementation demonstrates the application of the solution in the design of thermal textiles. Furthermore, the embodiment can be extended to brocade patterns. In this scenario, the interlacing logic represents the capture of intricate patterns, the model is constructed at up to five layers, and the algorithm uses advanced graph theory, such as minimum spanning trees, to optimize parameter distribution during fusion. The fusion process is explained in detail: first, the graph node parameters are initialized, then the rendered values ​​are propagated through the tree structure, iteratively adjusted to match realistic lighting and shadows. This method performs exceptionally well in complex patterns, achieving high-fidelity reconstruction.

[0122] For example, another embodiment involves embroidery patterns, with an output emphasizing color rendering. In the algorithm fusion step, parameters include chromaticity vectors, which are assigned using a graph coloring algorithm to ensure color consistency in the 3D effect.

[0123] Understandably, these implementation methods are all limited to the field of textile pattern design, demonstrating the flexibility of the technical solution. Through objective description, the solution can achieve efficient 3D reconstruction. In the effect description, this technical solution provides a reliable pattern output method in textile design through graph theory fusion, supporting rapid iteration of multiple patterns.

[0124] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A method for generating three-dimensional woven patterns based on graph theory and neural networks, characterized in that, The method includes: Obtain line segment data from traditional patterns, construct a topology graph consisting of nodes and edges from the initial line segment data, identify and judge the intersection positions in the topology graph, mark the overlapping areas of multiple lines in layers, and determine the hierarchical order of multi-level relationships. Based on the hierarchical order of the multi-level relationship, the topology graph is labeled and processed in layers to obtain a hierarchical structure model. The hierarchical structure model includes enhanced interleaving logic representation data, which contains the up and down interleaving vector description of each intersection point. Based on the hierarchical structure model, the stereo rendering parameters are extracted from the topological graph, and multiple variant pattern skeletons are generated using the graph segmentation method to obtain a diverse set of pattern skeletons. The fusion data of the diverse pattern skeleton set and hierarchical structure model is obtained, and the fusion data is processed by a generative graph convolutional neural network to obtain a pattern image with visual depth. The interweaving logic representation data and hierarchical structure model in the pattern image are fused with rendering parameters to generate a reconstructed three-dimensional woven pattern.

2. The method for generating three-dimensional woven patterns based on graph theory and neural networks according to claim 1, characterized in that, The step of obtaining line segment data from traditional patterns and constructing a topological graph consisting of nodes and edges from the initial line segment data includes: Initial line segment data is extracted from a pre-set traditional pattern database, and a topological graph consisting of nodes and edges is constructed using graph theory algorithms to obtain the connection relationship of each line segment; For the aforementioned connection relationship, obtain the preliminary intersection point position and determine the intersection point coordinate value; Based on the coordinate values ​​of the intersection points, the matching degree of the pattern reconstruction is determined. When the matching degree is lower than a predetermined standard, the endpoint positions of the initial line segment data are adjusted to form an optimized connection relationship. By optimizing the connection relationships and integrating the pattern analysis method, the final intersection point position is determined, and the final topology map is obtained.

3. The method for generating three-dimensional woven patterns based on graph theory and neural networks according to claim 2, characterized in that, The step of determining the final intersection point position by optimizing the connection relationship and integrating pattern analysis includes: Based on the optimized connection relationship, the intersection features of line segments are extracted, and the position distribution of intersection points is calculated in conjunction with the pattern analysis logic. Based on the given location distribution, the coordinate values ​​of the intersection points are adjusted to generate the final intersection point locations.

4. The method for generating three-dimensional woven patterns based on graph theory and neural networks according to claim 1, characterized in that, The process of identifying and determining the intersection positions in the topology graph, marking the overlapping areas of multiple lines in layers, and determining the hierarchical order of multi-level relationships includes: Based on the intersection positions in the topology diagram, obtain the interlacing sequence information of adjacent lines and determine the number of sequence superpositions; Determine the number of sequence overlaps; if it exceeds two, mark it as a multi-level crossover to obtain a multi-level crossover identifier. Obtain the relative positions of the overlapping lines at the multi-level intersection markers to determine the hierarchical order of the multi-level relationship; Based on the hierarchical order, the overlapping segments between adjacent lines are extracted, and the ratio of the length of the overlapping segment to the distance of the intersection point is determined to obtain the hierarchical order of the multi-level relationship.

5. The method for generating three-dimensional woven patterns based on graph theory and neural networks according to claim 1, characterized in that, The process of hierarchically labeling and processing the topology graph according to the hierarchical order of multi-level relationships to obtain a hierarchical structure model includes: According to the hierarchical order, the hierarchical relationship of the intersection points in the topology graph is labeled, hierarchical labeling data is generated, and the hierarchical labeling data is mapped to the topology graph to obtain labeled graph data; The labeled graph data is processed by a graph convolutional neural network to obtain interleaved logic representation data. Multi-level intersection data is extracted from the interleaved logic representation, and the coverage vector of each multi-level intersection is calculated to form a coverage vector distribution matrix. Based on the coverage vector distribution matrix, vector data of adjacent intersection points are extracted, and the difference between the two vectors is calculated. When the difference exceeds a predetermined standard, the local hierarchical order is adjusted to generate adjusted hierarchical distribution data. By analyzing the dependencies between levels using the adjusted hierarchical distribution data, a hierarchical framework is constructed.

6. The method for generating three-dimensional knitted patterns based on graph theory and neural networks according to claim 5, characterized in that, After analyzing the dependencies between levels and constructing a hierarchical framework using the adjusted hierarchical distribution data, the process further includes: The connection stability analysis is performed on the hierarchical framework. When the stability is lower than the predetermined standard, the hierarchical units with higher nesting depth are decomposed to generate the decomposed hierarchical data. Based on the decomposed hierarchical data, the matching degree between the covering vector and the adjacent intersection vector is recalculated, the vector data distribution results are optimized, and the coordinated hierarchical structure model is constructed.

7. The method for generating three-dimensional woven patterns based on graph theory and neural networks according to claim 1, characterized in that, The step involves extracting stereo rendering parameters from the topological graph based on the hierarchical structure model, and generating multiple variant pattern skeletons using a graph segmentation method, resulting in a diverse set of pattern skeletons, including: The topology graph is obtained from the hierarchical model, the 3D rendering parameters are extracted, and the initial pattern framework is obtained. The initial pattern framework is segmented using a graph theory algorithm to generate multiple variant pattern skeletons. The sequence of skeleton variants is determined, and diverse elements are integrated through adaptive adjustment of the skeleton patterns to obtain an extended pattern set. If the extended pattern set meets the preset diversity threshold, its completeness is determined to obtain the optimized pattern skeleton group; The optimized pattern skeleton group is integrated to determine a diverse set of pattern skeletons.

8. The method for generating three-dimensional woven patterns based on graph theory and neural networks according to claim 1, characterized in that, The process of acquiring the fusion data of the diverse pattern skeleton set and hierarchical structure model, and processing the fusion data through a generative graph convolutional neural network to obtain a pattern image with visual depth includes: Obtain the feature vectors of the preset diverse pattern skeleton set, and obtain the fused data based on the feature vectors and the node data of the hierarchical structure model; The fused data is processed by a generative graph convolutional neural network to obtain pattern style transfer parameters. If the node weights exceed a preset threshold, an enhanced depth feature map is obtained. The enhanced depth feature map is then mapped to the image space to obtain a pattern image with visual depth.

9. The method for generating three-dimensional woven patterns based on graph theory and neural networks according to claim 1, characterized in that, The interlacing logic representation data and hierarchical structure model in the pattern image are fused with rendering parameters to generate a reconstructed three-dimensional woven pattern, including: The interlacing logic representation data is extracted from the pattern image, and combined with the hierarchical structure model to generate a rendering parameter fusion result; Based on the rendering parameter fusion results, the matching degree between the curvature vector and the interlacing points is adjusted to generate the reconstructed three-dimensional woven pattern.

10. The method for generating three-dimensional woven patterns based on graph theory and neural networks according to claim 9, characterized in that, Based on the rendering parameter fusion results, the matching degree between the curvature vector and the interlacing points is adjusted to generate the reconstructed 3D woven pattern output, including: The pattern curvature is corrected on the rendering parameter fusion result. By adjusting the matching degree between the curvature vector and the interlacing point, three-dimensional weaving simulation data is obtained. The continuity of the three-dimensional weaving simulation data is compared to determine the reconstruction completeness; if the reconstruction completeness meets a preset threshold, the reconstructed three-dimensional weaving pattern is generated.

Images