Building design scene automatic intelligent generation method and system

Abstract

The invention relates to the technical field of computer aided design, in particular to an automatic intelligent building design scene generation method and system.The automatic intelligent building design scene generation method comprises the following steps of analyzing a material text mapping high-dimensional semantic vector, executing dot product projection with a physical reference to generate an attribute module length, and resolving noise frequency and a normal intensity coefficient; the method comprises the following steps of: mapping a texture style description text to a high-dimensional vector space, projecting the texture style description text to a physical attribute reference dimension, constructing a height field matrix by driving gradient interpolation and fractal superposition, generating a normal and roughness map based on differential operation, covering the surface of a three-dimensional model, and outputting a design scene. According to the method, a quantitative conversion path from semantic features to geometric generation control variables is established, noise frequency and normal intensity generation of programmed textures are driven, normal and roughness maps meeting physical rendering standards are generated, automatic rendering of material details is achieved, and unification of microscopic geometric textures and macroscopic visual styles of a design scene is achieved.

Description

Technical Field

[0001] This invention relates to the field of computer-aided design technology, and in particular to an automated intelligent generation method and system for architectural design scenarios. Background Technology

[0002] Computer-aided design (CAD) technology involves using computer hardware and software systems to assist engineers in conceiving, analyzing, modifying, and optimizing design schemes. Its core aspects include geometric modeling, graphic display, engineering data management, and human-computer interaction interface design. The aim is to transform the geometric shape, physical properties, and spatial relationships of products into computer-recognizable and processable data models through digital means, thereby achieving precision and visualization in the design process. Traditional architectural design scene generation methods involve designers manually drawing basic geometric elements such as points, lines, and surfaces in the CAD software interface using mouse clicks and keyboard input to construct the outlines of building walls and floors. Based on design requirements, they then search for and import 3D models of standard components such as doors, windows, furniture, and furnishings from a local material library. They then manually adjust each model to its specific position within the virtual coordinate system using movement, rotation, and scaling tools, and set the material textures and ambient lighting attributes for each component surface to complete the scene construction.

[0003] Current methods for generating architectural design scenes rely on manually retrieving and adjusting components from local libraries. This manual, discrete approach severs the connection between abstract concepts and physical properties. Designers must spend time on repetitive primitive drawing and spatial transformation. Local material libraries limit the diversity of material representation. When faced with complex weathering or uneven texture requirements, textures can only be mechanically superimposed, lacking an automatic evolution mechanism for micro-features. As a result, scene construction is limited by operational precision and material matching, making it difficult to accurately present differentiated physical details. This leads to a lengthy cycle in transforming design intent into visual results. Summary of the Invention

[0004] To address the technical problems existing in the prior art, this invention provides an automated intelligent generation method for architectural design scenes, comprising the following steps: To achieve the above objectives, the present invention adopts the following technical solution: an automated intelligent generation method for architectural design scenes, comprising the following steps: S1: Obtain the material style description text, analyze the contextual association and semantic proximity of the character sequence in the text, map the character sequence to a high-dimensional vector space, generate the material semantic feature vector, and calculate the sparse distribution characteristics of the material semantic feature vector in the multi-dimensional space. S2: Obtain the physical attribute reference vector that characterizes color tendency, surface weathering degree and geometric roughness, calculate the vector dot product of the material semantic feature vector and the physical attribute reference vector, analyze the projection component characteristics and direction consistency characteristics of the vector in each physical attribute subspace, and generate attribute projection modulus parameters. S3: Extract the attribute projection modulus parameter, perform normalization and interval scaling operations, match the programmed input range, generate texture synthesis control parameters and parse component dimensions, and output the noise frequency coefficient that controls the noise distribution density and the normal intensity coefficient that controls the surface undulation. S4: Extract the noise frequency coefficients, input them into the mesh gradient interpolation logic and perform pseudo-random gradient calculations to generate a texture noise layer and calculate the superposition weights. Perform fractal superposition calculations based on the normal intensity coefficients to construct a height field data matrix. S5: Based on the height field data matrix, perform differential operations to obtain the slope vectors of the pixels in the horizontal and vertical directions, transform the coordinate space to generate a normal channel image, analyze the local distribution density characteristics of the slope vectors and perform reverse mapping processing to generate a roughness channel image, and map it onto the surface of the three-dimensional model of the building component to output the design scene generation result.

[0005] As a further aspect of the present invention, the material semantic feature vector includes semantic dimension values, context association weights, and abstract attribute feature values; the attribute projection modulus parameters include color tendency projection values, surface weathering projection values, and geometric bump projection values; the texture synthesis control parameters include noise distribution density coefficients, surface undulation intensity coefficients, and signal amplitude constraint values; the height field data matrix includes continuous fractal geometric data, multi-layer noise superposition values, and mesh vertex height information; and the design scene generation results include physically rendered standard normal maps, adaptive roughness maps, and materialized 3D component models.

[0006] As a further aspect of the present invention, the step of obtaining the material semantic feature vector specifically includes: S101: Obtain the material style description text describing the material style of the target building component, analyze the arrangement and combination of character sequences, calculate the semantic proximity of adjacent character nodes, identify the core keywords in the material description, weight the character sequence according to the contextual association strength, aggregate the semantic logic information in the text, and generate semantic sequence association features. S102: Based on the semantic sequence association features, determine the mapping position of character nodes in the high-dimensional feature space, calculate the spatial distance between semantic feature points and quantify the similarity, adjust the distribution and clustering state of feature points in the coordinate system according to the similarity, convert unstructured text into a set of spatial coordinate points, and generate high-dimensional spatial mapping coordinate data. S103: For the high-dimensional spatial mapping coordinate data, perform vectorization synthesis, construct a multi-dimensional feature array, calculate the sparse distribution density in each dimension, identify redundant dimensions and separate attribute components to generate a material semantic feature vector.

[0007] As a further aspect of the present invention, the process of adjusting the distribution and clustering state of feature points in the coordinate system based on similarity specifically includes: Based on the semantic sequence association features, a semantic similarity measurement matrix is ​​constructed to quantify the semantic association strength between character nodes, and a semantic clustering judgment threshold for classifying semantic closeness and distance and a maximum number of iterations are set to limit the iteration operation cycle. For any two feature points in the semantic similarity measurement matrix, extract the corresponding semantic similarity value, and compare the semantic similarity value with the semantic clustering judgment threshold. If the semantic similarity value is greater than the semantic clustering determination threshold, then the two feature points are determined to belong to the same semantic cluster with strong contextual association. The Euclidean distance between the two feature points in the current high-dimensional coordinate system is calculated, a gravity vector pointing to the geometric center of the feature points is generated, and the vector magnitude is adjusted according to the Euclidean distance to drive the feature points to perform coordinate displacement along the direction of the gravity vector to reduce the spatial spacing. If the semantic similarity value is less than or equal to the semantic clustering determination threshold, then the two feature points are determined to belong to heterogeneous semantic clusters with mutually exclusive semantic logic. A discrete repulsive force vector is generated that deviates from the coordinate orientation of the target feature point, and the feature point is driven to perform reverse displacement along the direction of the discrete repulsive force vector to increase the spatial distance. At the end of a single iteration cycle, the spatial coordinate data of the feature points are refreshed in real time, and the coordinate offset of the feature points in the two iterations is monitored. When the coordinate offset is less than the preset stable convergence micro value or the number of operation rounds reaches the maximum number of iterations, the distribution adjustment operation is stopped and high-dimensional spatial mapping coordinate data is output.

[0008] As a further aspect of the present invention, the step of obtaining the attribute projection modulus parameter specifically includes: S201: Obtain the physical attribute reference vector that characterizes color tendency, surface weathering degree and geometric roughness, call the material semantic feature vector and the physical attribute reference vector to perform a dot product operation in the vector space, calculate the projection length of the material semantic feature vector in each reference vector definition direction, analyze the angular deviation state between the semantic vector and the physical reference vector in multi-dimensional space, and generate vector dot product correlation data. S202: Based on the vector dot product correlation data, construct an analytical subspace for each physical attribute dimension, map the correlation data to the corresponding attribute subspace, analyze the directional consistency of the projection vector on the local coordinate axis of the subspace, and generate the subspace component distribution characteristics. S203: Based on the distribution characteristics of the subspace components, the physical performance intensity of each attribute is quantified by calculating the vector magnitude, the Euclidean norm of the effective feature components in each physical attribute dimension is calculated, the weight ratio of texture feature generation is matched, and the attribute projection magnitude parameter is generated.

[0009] As a further aspect of the present invention, the steps for obtaining the noise frequency coefficient and the normal intensity coefficient are specifically as follows: S301: Call the attribute projection modulus parameter, determine the numerical distribution boundary in the current projection data set, perform normalization operation, map the modulus value to the standard unit interval, perform linear scaling according to the input domain of the programmable generator, adjust the parameter distribution curve, match the linear response characteristics, and generate texture synthesis control parameters; S302: Based on the texture synthesis control parameters, analyze the independent components of the corresponding surface grain size and geometric displacement, verify the component values ​​according to the preset signal frequency limit and amplitude threshold, establish boundary constraints for signal generation, and generate signal generation constraint variables. S303: Generate constraint variables for the signal, construct independent driving links through channel separation, map frequency constraints to the coordinate scaling factor of the noise generator, define the density of texture distribution, map amplitude constraints to the height perturbation multiplier, define surface undulation slope, and generate noise frequency coefficients and normal intensity coefficients.

[0010] As a further aspect of the present invention, the steps for obtaining the height field data matrix are specifically as follows: S401: Obtain the noise frequency coefficient, input it into the grid gradient calculation logic, determine the sampling step size, perform pseudo-random gradient vector generation on the grid vertices, calculate the interpolation distribution values ​​of pixels inside the grid, and generate a texture noise layer. S402: Based on the texture noise layer, analyze the energy attenuation trend of the noise signal in the frequency domain, calculate the contribution ratio of each fractal level in texture synthesis, determine the mixing intensity of details in each frequency band, and generate fractal superposition weight parameters. S403: For the fractal superposition weight parameters, the normal intensity coefficient is called as the global amplitude scaling factor to perform multifractal superposition operation, accumulate low-frequency contours and high-frequency details, map the accumulated signal to two-dimensional matrix coordinates, and construct a height field data matrix including continuous fractal features.

[0011] As a further aspect of the present invention, the step of obtaining the design scenario generation result specifically includes: S501: Based on the height field data matrix, calculate the height difference between adjacent pixel nodes in the matrix, obtain the gradient values ​​of the pixel in the horizontal and vertical directions by performing differential operations, convert the gradient values ​​into unit normal vectors in the tangent space, map the normal vector components to the RGB color space according to the physical rendering coding standard, construct a texture map representing the geometric orientation of the surface, and generate a normal channel image. S502: Based on the normal channel image, extract the normal vector array contained in the pixel, set a local sampling window, analyze the discrete density characteristics of the normal vector in spatial distribution, quantify the intensity of surface micro-geometry disturbance, perform numerical inverse mapping processing, and generate a roughness channel image. S503: Call the normal channel image and roughness channel image to construct a composite material shader containing geometric details and optical properties, align the UV texture coordinates of the 3D model of the building component, perform multi-channel texture mapping and cover the surface of the 3D geometry, perform material simulation and lighting calculation of the components in the scene, and output the design scene generation result.

[0012] An automated intelligent generation system for architectural design scenes includes: The semantic feature extraction module obtains the material style description text, analyzes the contextual association and semantic proximity of the character sequences in the text, maps the character sequences to a high-dimensional vector space, generates material semantic feature vectors, and calculates the sparse distribution characteristics of the material semantic feature vectors in the multi-dimensional space. The physical property mapping module obtains the physical property reference vectors that characterize color tendency, surface weathering degree and geometric unevenness, calculates the vector dot product of the material semantic feature vector and the physical property reference vector, analyzes the projection component characteristics and direction consistency characteristics of the vector in each physical property subspace, and generates the attribute projection modulus parameter. The texture control calculation module extracts the attribute projection modulus parameter, performs normalization and interval scaling operations, matches the programmed input range, generates texture synthesis control parameters and parses the component dimensions, and outputs the noise frequency coefficient that controls the noise distribution density and the normal intensity coefficient that controls the surface undulation. The height field signal synthesis module extracts the noise frequency coefficients, inputs them into the grid gradient interpolation logic and performs pseudo-random gradient calculations, generates a texture noise layer and calculates the superposition weights, performs fractal superposition calculations based on the normal intensity coefficients, and constructs a height field data matrix. The scene rendering generation module performs differential operations based on the height field data matrix to obtain the slope vectors of the pixels in the horizontal and vertical directions, transforms the coordinate space to generate a normal channel image, analyzes the local distribution density characteristics of the slope vectors and performs inverse mapping processing to generate a roughness channel image, and maps it to the surface of the three-dimensional model of the building component, outputting the design scene generation result.

[0013] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, the material style description text is mapped to a high-dimensional vector space and projected onto the physical property baseline dimension. A quantitative transformation path from semantic features to geometric generation control variables is established, which drives the generation of noise frequency and normal intensity of the driver texture. Normal and roughness maps that conform to physical rendering standards are generated, realizing the automatic evolution of material details and achieving the unity of the design scene in terms of micro-geometric texture and macro-visual style. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a schematic diagram of the steps of the present invention; Figure 2 This is a detailed schematic diagram of S1 of the present invention; Figure 3 This is a detailed schematic diagram of S2 of the present invention; Figure 4 This is a detailed schematic diagram of S3 of the present invention; Figure 5 This is a detailed schematic diagram of S4 of the present invention; Figure 6 This is a detailed schematic diagram of S5 of the present invention; Figure 7 This is a system module diagram of the present invention. Detailed Implementation

[0016] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0017] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0018] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, their intended meanings are consistent. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, their intended meanings are consistent.

[0019] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0020] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0021] Please see Figure 1 This invention provides an automated intelligent generation method for architectural design scenes, comprising the following steps: S1: Obtain the material style description text, analyze the contextual association and semantic proximity of the character sequence in the text, map the character sequence to a high-dimensional vector space, generate the material semantic feature vector, and calculate the sparse distribution characteristics of the material semantic feature vector in the multi-dimensional space. S2: Obtain the physical attribute reference vectors that characterize color tendency, surface weathering degree and geometric roughness, calculate the vector dot product of material semantic feature vector and physical attribute reference vector, analyze the projection component characteristics and direction consistency characteristics of vectors in each physical attribute subspace, and generate attribute projection modulus parameters. S3: Extract the attribute projection modulus parameter, perform normalization and interval scaling operations, match the programmed input range, generate texture synthesis control parameters and parse component dimensions, and output the noise frequency coefficient that controls the noise distribution density and the normal intensity coefficient that controls the surface undulation. S4: Extract noise frequency coefficients, input them into the mesh gradient interpolation logic and perform pseudo-random gradient calculations to generate a texture noise layer and calculate the superposition weights. Perform fractal superposition operations based on the normal intensity coefficients to construct a height field data matrix. S5: Based on the height field data matrix, perform differential operations to obtain the slope vectors of the pixels in the horizontal and vertical directions, transform the coordinate space to generate the normal channel image, analyze the local distribution density characteristics of the slope vector and perform reverse mapping processing to generate the roughness channel image, and map it to the surface of the 3D model of the building component, outputting the design scene generation result.

[0022] The material semantic feature vector includes semantic dimension values, context association weights, and abstract attribute feature values. The attribute projection modulus parameters include color tendency projection values, surface weathering projection values, and geometric bump projection values. The texture synthesis control parameters include noise distribution density coefficients, surface undulation intensity coefficients, and signal amplitude constraint values. The height field data matrix includes continuous fractal geometric data, multi-layer noise superposition values, and mesh vertex height information. The design scene generation results include physically rendered standard normal maps, adaptive roughness maps, and materialized 3D component models.

[0023] Please see Figure 2 The specific steps for obtaining the material semantic feature vector are as follows: S101: Obtain the material style description text describing the material style of the target building component, analyze the arrangement and combination of character sequences, calculate the semantic proximity of adjacent character nodes, identify the core keywords in the material description, weight the character sequence according to the contextual association strength, aggregate the semantic logic information in the text, and generate semantic sequence association features. First, the descriptive text for the ancient building restoration scenario, "severely weathered red brick wall with deep cracks and covered with moss," is obtained through a text acquisition interface. A character index list is built, and the character nodes in the text are traversed. The sliding window size is set to 3 character units, and the position index difference between adjacent character nodes in the sequence is calculated. The reciprocal of the position index difference is used as the basic metric for semantic proximity. For example, for the word nodes "severe" and "weathered," whose index positions are 2 and 3 respectively, the basic semantic proximity is... ; A pre-built keyword library for the construction field is invoked, and the TF-IDF algorithm is used to calculate the weight value of each word in the text. "Crack", "moss", "weathering", and "red brick" are identified as core keywords. The weight coefficient of "weathering" is calculated to be 0.85, and the weight coefficient of "red brick" is 0.72. Based on the contextual association strength, that is, the semantic dependency relationship between the core keywords and their modifiers, a weighted encoding matrix is ​​constructed. The weight coefficients are multiplied by the above semantic proximity to obtain the comprehensive feature value of each character node. The entire text sequence is traversed, and the comprehensive feature values ​​of each node are aggregated according to temporal logic to construct a feature sequence containing part-of-speech tagging, dependency relations and semantic weights. Non-core words are weighted down, and the semantic orientation of key morphological features is preserved. Finally, semantic sequence association features that represent the internal logical structure of the text are generated.

[0024] S102: Based on semantic sequence association features, determine the mapping position of character nodes in high-dimensional feature space, calculate the spatial distance between semantic feature points and quantify similarity, adjust the distribution and clustering state of feature points in coordinate system according to similarity, convert unstructured text into a set of spatial coordinate points, and generate high-dimensional spatial mapping coordinate data. By calling semantic sequence association features, a semantic similarity metric matrix is ​​constructed to characterize the semantic association strength between character nodes. The semantic clustering judgment threshold for distinguishing semantic closeness is set to 0.75, and the maximum number of iterations in the iteration operation cycle is limited to 50. For the two feature points "crack" and "weathering" in the matrix, their semantic similarity value in the current high-dimensional feature space (set as 128-dimensional space) is extracted and is 0.82. This value of 0.82 is compared with the threshold of 0.75. Since 0.82 is greater than 0.75, it is determined that the two feature points belong to the same semantic cluster. The Euclidean distance between them in the current coordinate system is calculated to be 1.2 units. Based on this distance, a convergent gravitational vector is generated pointing towards the geometric center of the two points (the coordinates of which are the average of the coordinates of the two points). The gravitational coefficient is set to 0.5, and the displacement is calculated as follows: This drives the two feature points to move 0.6 units towards the geometric center respectively; Meanwhile, for the two feature points "red brick" and "moss", if their semantic similarity is 0.4 (less than 0.75), a discrete repulsive force vector is generated to drive their mutually exclusive movement. At the end of a single iteration, the coordinates of all feature points are refreshed, and the sum of the coordinate offsets of the two iterations is calculated. For example, if the offset of the 12th iteration is 0.05, which is less than the preset stable convergence micro value of 0.01, the iteration continues until the convergence condition is met or 50 iterations are reached. The final stable set of point coordinates is then used as the high-dimensional space mapping coordinate data.

[0025] S103: For high-dimensional spatial mapping coordinate data, perform vectorization synthesis, construct a multi-dimensional feature array, calculate the sparse distribution density in each dimension, identify redundant dimensions and separate attribute components to generate material semantic feature vectors. For high-dimensional spatial mapping coordinate data, a vectorization synthesis operation is performed to integrate the discrete coordinate point set into a multi-dimensional feature array with a dimension of 128. The array is then iterated through each dimension and the sliding window statistical method is used to calculate the variance of the numerical distribution on the dimension axis to characterize the sparse distribution density. A sparsity threshold of 0.05 is set. If the variance of a dimension is less than 0.05, the dimension is considered redundant, indicating that it contributes little to the distinguishing power of the material features. For example, if the variance of the 12th dimension is 0.02 and the variance of the 45th dimension is 0.89, the 12th dimension is removed and the 45th dimension is retained. The retained effective dimensions are then subjected to attribute component separation. Principal component analysis (PCA) is used to extract the principal feature components, reducing the data to a 32-dimensional key feature space. This ensures that the feature vector retains more than 95% of the information while removing noise interference, ultimately generating a material semantic feature vector consisting of 32 floating-point numbers that represents the abstract attributes of the material.

[0026] Please see Figure 3 The specific steps for obtaining the attribute projection modulus parameter are as follows: S201: Obtain the physical property reference vectors that characterize color tendency, surface weathering degree, and geometric roughness; call the material semantic feature vector and the physical property reference vector to perform a dot product operation in the vector space; calculate the projection length of the material semantic feature vector in the direction defined by each reference vector; analyze the angular deviation between the semantic vector and the physical reference vector in the multidimensional space; and generate vector dot product correlation data. Retrieve standardized physical attribute baseline vectors from the physical attribute database, including baseline vectors representing color tendency. A baseline vector characterizing surface weathering and the reference vector representing the geometric convexity / concavity. Each reference vector is a 32-dimensional unit vector; the previously generated material semantic feature vector is called. Perform a dot product operation with each of the three reference vectors mentioned above, and the calculation formula is as follows: ; For example, the component value for the "red" dimension in the material semantic feature vector is 0.8. The corresponding component in the middle is 1.0. After the contributions of other dimension components are accumulated, the dot product result of the color attribute is calculated to be 0.76. Similarly, the calculated weathering degree dot product is 0.92 (corresponding to the description of "severe weathering"), and the geometric convexity dot product is 0.65 (corresponding to the description of "cracks"). Analyzing the cosine value of the angle between the semantic vector and the physical reference vector, if the cosine value is close to 1, it indicates that the projection purity of the semantic features in the corresponding physical attribute direction is high. The dot product scalar values ​​obtained from the above calculations and their corresponding direction cosine values ​​are combined to generate vector dot product correlation data that describes in detail the mapping relationship between semantics and physical attributes.

[0027] S202: Based on vector dot product correlation data, construct an analytical subspace for each physical attribute dimension, map the correlation data to the corresponding attribute subspace, analyze the directional consistency of the projection vector on the local coordinate axis of the subspace, and generate subspace component distribution characteristics; Based on vector dot product correlation data, independent analytical subspaces for three dimensions—color, weathering, and convexity—are constructed. Each subspace is defined as a local coordinate system with the reference vector as the Z-axis. The correlation data is mapped to the corresponding attribute subspace. For example, the weathering dot product result 0.92 is mapped to the weathering subspace. The projection components of the projection vector on the local XY plane of the subspace are analyzed. If the projection modulus on the XY plane is less than the preset consistency threshold of 0.1, the feature is determined to have high directional consistency in weathering attributes and belongs to a pure physical attribute mapping; if the projection modulus on the XY plane is 0.4, it indicates that the semantic feature is mixed with other attribute features (such as color interference) in addition to weathering attributes. Based on this analysis, the projection vector is filtered to remove components that deviate from the reference axis by more than 30 degrees, while retaining the effective principal axis components. This generates the subspace component distribution characteristics that describe the purity of each physical property, as shown in Table 1.

[0028]

[0029] As shown in Table 1, the system calculates the correction coefficient based on the axial deviation angle, which is the cosine value of the deviation angle. The corrected effective component value is the original dot product value multiplied by the correction coefficient.

[0030] S203: Based on the distribution characteristics of subspace components, the physical performance intensity of each attribute is quantified by calculating the vector magnitude, the Euclidean norm of the effective feature components in each physical attribute dimension is calculated, the weight ratio of texture feature generation is matched, and the attribute projection magnitude parameter is generated. Based on the characteristics of the subspace component distribution, the corrected effective component values ​​are extracted, such as the effective component value of weathering degree of 0.91 and the effective component value of unevenness of texture of 0.60; the Euclidean norm calculation logic is used to merge the multidimensional attribute strengths into a unified physical performance modulus. However, in this embodiment, in order to retain the independent control capability of each attribute, the scalar modulus of each attribute dimension is calculated separately. The weighting ratios for texture feature generation are set. Based on the semantic description of "severe weathering," the weathering attribute is assigned a weight of 1.5 times, the bumpiness attribute a weight of 1.2 times, and the color a weight of 1.0 times. A weighted calculation is then performed; for example, the final modulus parameter for weathering is... The final mold length parameter for the textured surface is: ; These values ​​quantify the intensity of the material's intuitive performance in the physical world. The larger the value, the more significant the corresponding physical characteristics. Finally, these weighted scalar values ​​are combined to generate the attribute projection modulus parameter used to control the generation of subsequent steps.

[0031] Please see Figure 4 The specific steps for obtaining the noise frequency coefficient and normal intensity coefficient are as follows: S301: Call the attribute projection modulus parameter, determine the numerical distribution boundary in the current projection data set, perform normalization operation, map the modulus value to the standard unit interval, perform linear scaling according to the input domain of the procedural generator, adjust the parameter distribution curve, match the linear response characteristics, and generate texture synthesis control parameters; Calling the attribute projection modulus parameter, such as a weathering modulus of 1.365, first determines the numerical distribution boundary of the current projection dataset, assuming the system's preset theoretical maximum modulus is 2.0 and the minimum is 0.0; then performs Min-Max normalization, calculated using the following formula: Map the modulus to the standard unit interval [0,1]. Subsequently, linear scaling is performed based on the input domain of the programmed generator. Assuming the frequency input range of the noise generator is [0, 100], the normalized value 0.6825 is mapped to this range, and the result is calculated. ; At the same time, the parameter distribution is adjusted according to the preset exponential distribution curve to match the non-linear perception of texture changes by the human eye, and gamma correction (Gamma=2.2) is performed on the linear values. Then, it is mapped to the amplitude range [0,10] to obtain 4.3 as the basic control value; finally, texture synthesis control parameters containing linear scaling values ​​and nonlinear correction values ​​are generated.

[0032] S302: Based on the texture synthesis control parameters, analyze the independent components of the corresponding surface grain size and geometric displacement, verify the component values ​​according to the preset signal frequency limit and amplitude threshold, establish the boundary constraints for signal generation, and generate signal generation constraint variables. Based on texture synthesis control parameters, independent components corresponding to surface grain size (corresponding to weathering degree) and geometric displacement (corresponding to unevenness) are analyzed; the signal frequency limit is set to 80Hz, and the amplitude threshold is 5.0 units; the initial frequency control value of 68.25 corresponding to the aforementioned weathering degree is extracted and verified against the frequency limit of 80Hz, because... This value is valid; extract the initial amplitude control value corresponding to the convexity / concave feel (assuming it is 6.5 calculated by S301), and compare it with the amplitude threshold of 5.0, because... This triggers the boundary clamping logic, forcibly correcting it to version 5.0; Simultaneously, the lower limit of the component values ​​is detected. If a component is lower than 0.1, it is raised to 0.1 to avoid generating an invalid plane. Through the above numerical comparison and clamping operations, a safe boundary for signal generation is established to prevent model interleaving due to excessively large parameters or loss of details due to excessively small parameters, and to generate signal generation constraint variables that conform to the input specifications of the procedural generator.

[0033] S303: For signal generation constraint variables, through channel separation, construct independent driving links, map frequency constraints to the coordinate scaling factor of the noise generator, define the density of texture distribution, map amplitude constraints to the height perturbation multiplier, define surface undulation slope, and generate noise frequency coefficients and normal intensity coefficients. For the signal generation constraint variables, a channel separation operation is performed to construct two independent signal driving links; Link 1 maps the weathering degree-related constraint variable (68.25) to the coordinate scaling factor of the noise generator. This factor determines the density of noise sampling points per unit area, i.e. the density of texture distribution. The higher the value, the finer the texture (such as moss pores). Link 2 maps the convexity-related constraint variable (5.0) to a height perturbation multiplier, which defines the noise value's normal offset from the geometric surface, i.e., the surface undulation slope; specifically, during the mapping process, a coordinate scaling factor is applied. The fine-tuning coefficients yield the final noise frequency coefficient as follows: ; Application of high disturbance multiplier The strength coefficient is used to obtain the final normal strength coefficient. These two coefficients will be directly input as mathematical parameters into the subsequent signal synthesis function.

[0034] Please see Figure 5 The specific steps for obtaining the height field data matrix are as follows: S401: Obtain the noise frequency coefficient, input it into the grid gradient calculation logic, determine the sampling step size, perform pseudo-random gradient vector generation on the grid vertices, calculate the interpolation distribution values ​​of pixels inside the grid, and generate a texture noise layer. The noise frequency coefficient 6.825 was obtained and input into the Simplex Noise mesh gradient calculation logic; the mesh sampling step size was determined based on the frequency coefficient. In a two-dimensional plane coordinate system The grid is divided into sections; for each grid vertex, a hash function is called to generate a pseudo-random gradient vector. For example, the gradient vector of vertex A is... ; For any pixel P within the grid, calculate its distance vector to each vertex, and then perform a dot product operation between the distance vector and the gradient vector of the corresponding vertex; using the easing curve function... Smoothly interpolate the dot product result to calculate the basic gray value of the pixel (range [-1, 1]). Iterate through all pixels of the entire texture plane to generate a texture with a resolution of The base texture noise layer exhibits continuous and irregular grayscale variations, representing the basic random distribution characteristics of the material.

[0035] S402: Based on the texture noise layer, analyze the energy attenuation trend of the noise signal in the frequency domain, calculate the contribution ratio of each fractal level in texture synthesis, determine the mixing intensity of details in each frequency band, and generate fractal superposition weight parameters. Based on the texture noise layer, the fractal Brownian motion (FBM) algorithm logic is used to analyze the energy attenuation trend of the noise signal in the frequency domain; the number of fractal iteration layers (Octaves) is set to 5 to simulate the rich details required for "severe weathering"; the frequency multiplication factor (Lacunarity) of each fractal layer is calculated to be 2.0, and the amplitude attenuation factor (Persistence) is calculated to be 0.5. The weights are calculated and stacked layer by layer, with the first layer having a weight of 1.0, the second layer having a weight of 0.5, the third layer having a weight of 0.25, and so on. Based on the characteristics of the keyword "crack," the mixing intensity of the high-frequency bands (layers 4 and 5) is dynamically adjusted, increasing its weight by 20% based on the basic attenuation rule. That is, the weight of layer 4 is adjusted to... To enhance the sharpness of micro-cracks, a fractal superposition weight parameter containing the frequency and corresponding amplitude weight of each level is finally generated, as shown in Table 2.

[0036]

[0037] As shown in Table 2, the system establishes complete construction rules from macroscopic outlines to microscopic details through hierarchical parameter configuration.

[0038] S403: For fractal superposition weight parameters, the normal intensity coefficient is called as the global amplitude scaling factor to perform multifractal superposition operation, accumulate low-frequency contours and high-frequency details, map the accumulated signal to two-dimensional matrix coordinates, and construct a height field data matrix including continuous fractal features. For the fractal stacking weight parameters, the previously generated normal intensity coefficient of 1.0 is used as the global amplitude scaling factor; traversing each point in the texture coordinate space. Perform multifractal superposition operation, the formula logic is as follows:

[0039] Substituting the data from Table 2, the noise value of the first layer is multiplied by 1.0, the noise value of the second layer is multiplied by 0.5, and the calculation results of all 5 layers are accumulated. For example, the accumulated value calculated for a certain point is 0.65, which represents the height displacement of that point relative to the reference plane. The accumulated values ​​of all pixels are mapped to a floating-point two-dimensional matrix to construct a height field data matrix containing continuous fractal features. The values ​​in this matrix accurately record the unevenness and crack depth of the red brick wall surface caused by weathering and peeling, which constitute the geometric basis for the generation of physical properties.

[0040] Please see Figure 6 The specific steps for obtaining the design scene generation results are as follows: S501: Based on the height field data matrix, calculate the height difference between adjacent pixel nodes in the matrix, obtain the gradient values ​​of the pixel in the horizontal and vertical directions by performing differential operations, convert the gradient values ​​into unit normal vectors in the tangent space, map the normal vector components to the RGB color space according to the physical rendering coding standard, construct a texture map representing the geometric orientation of the surface, and generate a normal channel image. Based on the height field data matrix, the central difference method is used to calculate the height difference between adjacent pixel nodes within the matrix; for the coordinates in the matrix... Get the height value of the pixel. and adjacent points and Calculate the horizontal slope and vertical slope ; Assuming a certain point is calculated to obtain Construct unnormalized normal vectors , where Scale is the intensity coefficient; normalize the normal vector to a unit vector, and then... The components are mapped from the interval [-1,1] to the RGB color space [0,255], using the following formula: ; Finally, a normal map representing the surface geometry is generated. In the image, the blue-purple areas represent flat surfaces, and the pink-green areas represent slopes and crack edges.

[0041] S502: Based on the normal channel image, extract the normal vector array contained in the pixel, set a local sampling window, analyze the discrete density characteristics of the normal vector in spatial distribution, quantify the intensity of surface micro-geometric disturbance, perform numerical inverse mapping processing, and generate a roughness channel image. Based on the normal channel image, extract the normal vector array contained in each pixel; set the local sampling window size to... For each pixel, the variance of the normal vectors of nine pixels within a window is calculated, traversing the entire image. This variance quantifies the severity of perturbations in the surface's micro-geometry. For example, in the "moss" region, the normal vectors are randomly oriented, resulting in a variance of 0.8 (high roughness); while in the smooth "red brick" region, the normal vectors converge, resulting in a variance of 0.1 (low roughness). Based on the microsurface scattering theory, a reverse mapping process is performed to map the variance value to a grayscale value of 0-255, using the following formula: A roughness map is generated; in this image, the bright areas (white) represent the rough surface, which produces diffuse reflection under light, and the dark areas (black) represent the smooth surface, which produces specular reflection, thus accurately restoring the dry and rough physical texture of the weathered brick wall.

[0042] S503: Calls the normal channel image and roughness channel image to construct a composite material shader containing geometric details and optical properties, aligns the UV texture coordinates of the 3D model of the building component, performs multi-channel texture mapping and covers the surface of the 3D geometry, performs material simulation and lighting calculation of the components in the scene, and outputs the design scene generation result. By calling the normal channel image and roughness channel image, and combining them with the preset base color map (generated based on the "red" semantic in S1), a standard PBR (physically based rendering) composite material shader containing geometric details and optical properties is constructed. The UV texture coordinate data of the 3D model of the building component is read, and the generated channel images are aligned and mapped to the surface of the 3D geometry. In the virtual lighting environment, the rendering engine changes the light reflection angle according to the normal map to simulate bump shadows, and changes the specular range according to the roughness map to simulate material texture. Finally, without modifying the model mesh topology, a red brick wall with realistic weathering marks, moss coverage and crack details is presented, and the design scene generation result that meets the design requirements is output.

[0043] Please see Figure 7 An automated intelligent generation system for architectural design scenarios, comprising: The semantic feature extraction module obtains the material style description text, analyzes the contextual association and semantic proximity of the character sequences in the text, maps the character sequences to a high-dimensional vector space, generates material semantic feature vectors, and calculates the sparse distribution characteristics of the material semantic feature vectors in the multi-dimensional space. The physical property mapping module obtains the physical property reference vectors that characterize color tendency, surface weathering degree and geometric unevenness, calculates the vector dot product of the material semantic feature vector and the physical property reference vector, analyzes the projection component characteristics and directional consistency characteristics of the vectors in each physical property subspace, and generates the attribute projection modulus parameter. The texture control calculation module extracts the attribute projection modulus parameter, performs normalization and interval scaling operations, matches the programmed input range, generates texture synthesis control parameters and parses the component dimensions, and outputs the noise frequency coefficient that controls the noise distribution density and the normal intensity coefficient that controls the surface undulation. The height field signal synthesis module extracts noise frequency coefficients, inputs them into the mesh gradient interpolation logic and performs pseudo-random gradient calculations, generates a texture noise layer and calculates the superposition weights, performs fractal superposition calculations based on the normal intensity coefficients, and constructs a height field data matrix. The scene rendering generation module performs differential operations based on the height field data matrix to obtain the slope vectors of pixels in the horizontal and vertical directions. It then transforms the coordinate space to generate a normal channel image, analyzes the local distribution density characteristics of the slope vectors, performs inverse mapping processing, generates a roughness channel image, maps it to the surface of the 3D model of the building components, and outputs the design scene generation result.

[0044] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for automatically and intelligently generating architectural design scenes, characterized in that, Includes the following steps: S1: Obtain the material style description text, analyze the contextual association and semantic proximity of the character sequence in the text, map the character sequence to a high-dimensional vector space, generate the material semantic feature vector, and calculate the sparse distribution characteristics of the material semantic feature vector in the multi-dimensional space. S2: Obtain the physical attribute reference vector that characterizes color tendency, surface weathering degree and geometric roughness, calculate the vector dot product of the material semantic feature vector and the physical attribute reference vector, analyze the projection component characteristics and direction consistency characteristics of the vector in each physical attribute subspace, and generate attribute projection modulus parameters. S3: Extract the attribute projection modulus parameter, perform normalization and interval scaling operations, match the programmed input range, generate texture synthesis control parameters and parse component dimensions, and output the noise frequency coefficient that controls the noise distribution density and the normal intensity coefficient that controls the surface undulation. S4: Extract the noise frequency coefficients, input them into the mesh gradient interpolation logic and perform pseudo-random gradient calculations to generate a texture noise layer and calculate the superposition weights. Perform fractal superposition operations based on the normal intensity coefficients to construct a height field data matrix.

2. The automated intelligent generation method for architectural design scenes according to claim 1, characterized in that, The material semantic feature vector includes semantic dimension values, context association weights, and abstract attribute feature values. The attribute projection modulus parameters include color tendency projection values, surface weathering projection values, and geometric concavity and convexity projection values. The texture synthesis control parameters include noise distribution density coefficients, surface undulation intensity coefficients, and signal amplitude constraint values. The height field data matrix includes continuous fractal geometric data, multi-layer noise superposition values, and mesh vertex height information.

3. The automated intelligent generation method for architectural design scenes according to claim 1, characterized in that, The specific steps for obtaining the material semantic feature vector are as follows: S101: Obtain the material style description text describing the material style of the target building component, analyze the arrangement and combination of character sequences, calculate the semantic proximity of adjacent character nodes, identify the core keywords in the material description, weight the character sequence according to the contextual association strength, aggregate the semantic logic information in the text, and generate semantic sequence association features. S102: Based on the semantic sequence association features, determine the mapping position of character nodes in the high-dimensional feature space, calculate the spatial distance between semantic feature points and quantify the similarity, adjust the distribution and clustering state of feature points in the coordinate system according to the similarity, convert unstructured text into a set of spatial coordinate points, and generate high-dimensional spatial mapping coordinate data. S103: For the high-dimensional spatial mapping coordinate data, perform vectorization synthesis, construct a multi-dimensional feature array, calculate the sparse distribution density in each dimension, identify redundant dimensions and separate attribute components to generate a material semantic feature vector.

4. The automated intelligent generation method for architectural design scenes according to claim 3, characterized in that, The process of adjusting the clustering state of feature points in the coordinate system based on similarity is as follows: Based on the semantic sequence association features, a semantic similarity measurement matrix is ​​constructed to quantify the semantic association strength between character nodes, and a semantic clustering judgment threshold for classifying semantic closeness and distance and a maximum number of iterations are set to limit the iteration operation cycle. For any two feature points in the semantic similarity measurement matrix, extract the corresponding semantic similarity value, and compare the semantic similarity value with the semantic clustering judgment threshold. If the semantic similarity value is greater than the semantic clustering determination threshold, then the two feature points are determined to belong to the same semantic cluster with strong contextual association. The Euclidean distance between the two feature points in the current high-dimensional coordinate system is calculated, a gravity vector pointing to the geometric center of the feature points is generated, and the vector magnitude is adjusted according to the Euclidean distance to drive the feature points to perform coordinate displacement along the direction of the gravity vector to reduce the spatial spacing. If the semantic similarity value is less than or equal to the semantic clustering determination threshold, then the two feature points are determined to belong to heterogeneous semantic clusters with mutually exclusive semantic logic. A discrete repulsive force vector is generated that deviates from the coordinate orientation of the target feature point, and the feature point is driven to perform reverse displacement along the direction of the discrete repulsive force vector to increase the spatial distance. At the end of a single iteration cycle, the spatial coordinate data of the feature points are refreshed in real time, and the coordinate offset of the feature points in the two iterations is monitored. When the coordinate offset is less than the preset stable convergence micro value or the number of operation rounds reaches the maximum number of iterations, the distribution adjustment operation is stopped and high-dimensional spatial mapping coordinate data is output.

5. The automated intelligent generation method for architectural design scenes according to claim 3, characterized in that, The specific steps for obtaining the attribute projection modulus parameter are as follows: S201: Obtain the physical attribute reference vector that characterizes color tendency, surface weathering degree and geometric roughness, call the material semantic feature vector and the physical attribute reference vector to perform a dot product operation in the vector space, calculate the projection length of the material semantic feature vector in each reference vector definition direction, analyze the angular deviation state between the semantic vector and the physical reference vector in multi-dimensional space, and generate vector dot product correlation data. S202: Based on the vector dot product correlation data, construct an analytical subspace for each physical attribute dimension, map the correlation data to the corresponding attribute subspace, analyze the directional consistency of the projection vector on the local coordinate axis of the subspace, and generate the subspace component distribution characteristics; S203: Based on the distribution characteristics of the subspace components, the physical performance intensity of each attribute is quantified by calculating the vector magnitude, the Euclidean norm of the effective feature components in each physical attribute dimension is calculated, the weight ratio of texture feature generation is matched, and the attribute projection magnitude parameter is generated.

6. The automated intelligent generation method for architectural design scenes according to claim 5, characterized in that, The steps for obtaining the noise frequency coefficient and normal intensity coefficient are as follows: S301: Call the attribute projection modulus parameter, determine the numerical distribution boundary in the current projection data set, perform normalization operation, map the modulus value to the standard unit interval, perform linear scaling according to the input domain of the programmable generator, adjust the parameter distribution curve, match the linear response characteristics, and generate texture synthesis control parameters; S302: Based on the texture synthesis control parameters, analyze the independent components of the corresponding surface grain size and geometric displacement, verify the component values ​​according to the preset signal frequency limit and amplitude threshold, establish boundary constraints for signal generation, and generate signal generation constraint variables. S303: Generate constraint variables for the signal, construct independent driving links through channel separation, map frequency constraints to the coordinate scaling factor of the noise generator, define the density of texture distribution, map amplitude constraints to the height perturbation multiplier, define surface undulation slope, and generate noise frequency coefficients and normal intensity coefficients.

7. The automated intelligent generation method for architectural design scenes according to claim 6, characterized in that, The specific steps for obtaining the height field data matrix are as follows: S401: Obtain the noise frequency coefficient, input it into the grid gradient calculation logic, determine the sampling step size, perform pseudo-random gradient vector generation on the grid vertices, calculate the interpolation distribution values ​​of pixels inside the grid, and generate a texture noise layer. S402: Based on the texture noise layer, analyze the energy attenuation trend of the noise signal in the frequency domain, calculate the contribution ratio of each fractal level in texture synthesis, determine the mixing intensity of details in each frequency band, and generate fractal superposition weight parameters. S403: For the fractal superposition weight parameters, the normal intensity coefficient is called as the global amplitude scaling factor to perform multifractal superposition operation, accumulate low-frequency contours and high-frequency details, map the accumulated signal to two-dimensional matrix coordinates, and construct a height field data matrix including continuous fractal features.

8. The automated intelligent generation method for architectural design scenes according to claim 1, characterized in that, The method further includes: S5: Based on the height field data matrix, perform differential operations to obtain the slope vectors of the pixels in the horizontal and vertical directions, transform the coordinate space to generate a normal channel image, analyze the local distribution density characteristics of the slope vectors and perform reverse mapping processing to generate a roughness channel image, and map it onto the surface of the three-dimensional model of the building component to output the design scene generation result. The generated design scene includes physically rendered standard normal maps, adaptive roughness maps, and materialized 3D component models.

9. The automated intelligent generation method for architectural design scenes according to claim 8, characterized in that, The specific steps for obtaining the design scenario generation result are as follows: S501: Based on the height field data matrix, calculate the height difference between adjacent pixel nodes in the matrix, obtain the gradient values ​​of the pixel in the horizontal and vertical directions by performing differential operations, convert the gradient values ​​into unit normal vectors in the tangent space, map the normal vector components to the RGB color space according to the physical rendering coding standard, construct a texture map representing the geometric orientation of the surface, and generate a normal channel image. S502: Based on the normal channel image, extract the normal vector array contained in the pixel, set a local sampling window, analyze the discrete density characteristics of the normal vector in spatial distribution, quantify the intensity of surface micro-geometry disturbance, perform numerical inverse mapping processing, and generate a roughness channel image. S503: Call the normal channel image and roughness channel image to construct a composite material shader containing geometric details and optical properties, align the UV texture coordinates of the 3D model of the building component, perform multi-channel texture mapping and cover the surface of the 3D geometry, perform material simulation and lighting calculation of the components in the scene, and output the design scene generation result.

10. An automated intelligent generation system for architectural design scenes, characterized in that, The system is used to implement the automated intelligent generation method for architectural design scenes according to any one of claims 1-9, the system comprising: The semantic feature extraction module obtains the material style description text, analyzes the contextual association and semantic proximity of the character sequences in the text, maps the character sequences to a high-dimensional vector space, generates material semantic feature vectors, and calculates the sparse distribution characteristics of the material semantic feature vectors in the multi-dimensional space. The physical property mapping module obtains the physical property reference vectors that characterize color tendency, surface weathering degree and geometric unevenness, calculates the vector dot product of the material semantic feature vector and the physical property reference vector, analyzes the projection component characteristics and directional consistency characteristics of the vector in each physical property subspace, and generates the attribute projection modulus parameter. The texture control calculation module extracts the attribute projection modulus parameter, performs normalization and interval scaling operations, matches the programmed input range, generates texture synthesis control parameters and parses the component dimensions, and outputs the noise frequency coefficient that controls the noise distribution density and the normal intensity coefficient that controls the surface undulation. The height field signal synthesis module extracts the noise frequency coefficients, inputs them into the grid gradient interpolation logic and performs pseudo-random gradient calculations, generates a texture noise layer and calculates the superposition weights, performs fractal superposition calculations based on the normal intensity coefficients, and constructs a height field data matrix. The scene rendering generation module performs differential operations based on the height field data matrix to obtain the slope vectors of the pixels in the horizontal and vertical directions, transforms the coordinate space to generate a normal channel image, analyzes the local distribution density characteristics of the slope vectors and performs inverse mapping processing to generate a roughness channel image, and maps it to the surface of the three-dimensional model of the building component, outputting the design scene generation result.

Images