Wooden door style intelligent generation method and system based on deep learning

Abstract

The invention provides an intelligent wooden door style generation method and system based on deep learning, and relates to the technical field of intelligent design and intelligent manufacturing, and the method comprises the steps: constructing a wood database, collecting sample images, and analyzing feature parameters through a classification network to recognize the types and differences of wood, thereby achieving the intelligent generation of a wooden door style. Inputting the style instruction and the difference parameters into a generative network, generating a style effect picture under the constraint of wood characteristics, and associating processing precision data with physical parameters; and after the verification module performs structural rationality and implementability evaluation, design parameters are adjusted according to results, and finally a wooden door style design scheme meeting actual manufacturing requirements is output, so that full-process intelligence from wood characteristic analysis to style generation and process verification can be realized, and wooden door design is ensured to have both aesthetic value and manufacturing feasibility.

Description

Technical Field

[0001] This application relates to the fields of intelligent design and intelligent manufacturing technology, and in particular to a method and system for intelligent generation of wooden door styles based on deep learning. Background Technology

[0002] With the increasing demand for consumption upgrades and personalized customization, the wooden door industry is facing a transformation from standardized production to flexible and intelligent manufacturing. Consumers not only pursue harmony and unity between wooden doors and the overall home style, but also hope that the material texture and processing details can reflect their unique aesthetic taste. This requires design systems to respond quickly and generate feasible solutions that are both aesthetically pleasing and conform to the specific physical properties and processing techniques of wood, thereby facilitating the data flow from design conception to actual manufacturing.

[0003] Currently, some technical solutions attempt to use deep learning-based generative adversarial networks (GANs) to address the stylization generation problem. These solutions train on massive amounts of wooden door design image data to learn the distribution of their aesthetic features, thereby generating high-resolution, visually realistic wooden door images based on user-input style tags (such as "minimalist" or "European"). This method automates the design process to some extent and shortens the creative conception cycle.

[0004] However, such existing solutions have obvious drawbacks: First, their generation process relies heavily on the aesthetic features of the image and fails to deeply integrate with the specific physical properties of the wood (such as hardness and grain direction) and processing constraints (such as the carving precision of different tools on specific materials), which may result in the generated design drawings being unable to be actually processed or having effects that do not meet expectations; Second, the solutions lack a closed-loop verification and adjustment mechanism, making it impossible to effectively evaluate the structural rationality, mechanical feasibility, and matching degree with the target wood of the generated results, and their output is still mostly "conceptual drawings", which are far from actual production applications. Summary of the Invention

[0005] The purpose of this application is to provide a method, system, electronic device and storage medium for intelligent generation of wooden door styles based on deep learning, so as to solve the problem that the design renderings generated in the prior art are disconnected from the physical properties and processing technology of wood.

[0006] To address the aforementioned technical problems, firstly, this application provides a deep learning-based intelligent generation method for wooden door styles, comprising: A wood database containing physical properties and aesthetic parameters of different wood species is constructed, while surface morphology images of wood samples provided by users are acquired through image acquisition devices; The surface morphology image is analyzed to include various feature parameters such as wood door texture direction, contrast degree, and texture grain size. These feature parameters are then input into a classification network, which outputs the type identifier of the wood sample and the difference parameters between the wood sample and the standard material. The style instructions and the difference parameters are input into the material-aware generation network, and combined with the constraints of wood characteristics in the difference parameters, a wooden door style effect picture adapted to the wood characteristics is generated. At the same time, the processing accuracy data of different processing tools in the preset processing technology constraint library are associated and mapped with the physical parameters in the wood database. The structural rationality of the wooden door style rendering is evaluated by the verification module. Combined with the physical parameters and processing accuracy data after correlation mapping, the feasibility of the design pattern on the corresponding wood is verified in multiple dimensions to obtain the verification results. Based on the verification results, the style rendering of the wooden door is adaptively adjusted, and during the adjustment process, the generation parameters are adjusted based on the difference parameters and the processing accuracy data to output a wooden door style design scheme that meets the actual production requirements.

[0007] Optionally, the style instructions and the difference parameters are input into a material-aware generative network, and combined with the constraints of wood characteristics in the difference parameters, a wooden door style rendering adapted to the wood characteristics is generated, including: Style instructions and difference parameters are input into the encoder part of the generator network, which combines the style instructions and difference parameters into a joint vector; The joint vector is progressively compressed through multiple coding layers in the encoder section to obtain a latent space representation; The latent space representation is input into the decoder part of the generator network, and the decoder part progressively expands the latent space representation through multiple decoding layers to generate an initial image. During the processing of the decoder section, the difference parameter is introduced as a constraint. By adjusting the output value of the decoding layer, the texture direction, color distribution, and grain pattern in the initial effect image are made consistent with the wood characteristics described in the difference parameter, and a wooden door style effect image is output.

[0008] Optionally, the processing accuracy data of different processing tools on wood in the preset processing technology constraint library are associated and mapped with the physical parameters in the wood database, including: The machining accuracy data of multiple different machining tools are obtained from the preset machining process constraint library. The machining accuracy data of each machining tool includes the minimum engraving width, the maximum cutting depth, and the minimum drilling diameter. Physical parameters of multiple wood species are obtained from a wood database. The physical parameters of each wood species include hardness, density and brittleness values. For each processing tool, a mapping function is defined based on the relationship between the processing accuracy data and the physical parameters. The mapping function takes the physical parameter values ​​of the wood as input and outputs the actual achievable accuracy value of the processing tool on the wood, thus obtaining the corresponding mapping function.

[0009] Optionally, the structural rationality of the wooden door style rendering is evaluated through a verification module. Combined with the associated mapped physical parameters and processing precision data, the feasibility of the design pattern on the corresponding wood is verified from multiple dimensions to obtain verification results, including: Extract design patterns, including line width, depth, and complexity, from the style renderings of wooden doors; The structural evaluation unit of the verification module calculates the support strength value of the pattern and compares the support strength value with a preset threshold to check whether the features of the design pattern meet the structural stability requirements. The processing feasibility unit of the verification module uses the physical parameters and processing accuracy data after the association mapping to check whether the design pattern is processable by comparing the line width with the minimum engraving width of the corresponding processing tool for each line in the pattern. The output includes verification results for structural rationality indicators and processing feasibility indicators.

[0010] Optionally, the style rendering of the wooden door is adaptively adjusted based on the verification results, and during the adjustment process, parameters are adjusted and generated based on the difference parameters and the processing accuracy data to output a wooden door style design scheme that meets the actual manufacturing requirements, including: If the verification results indicate that the structure is unreasonable or the processing is not feasible, then adjust the generation parameters in the generative network, including texture generation intensity, color adjustment factor and line scaling factor; Based on the difference parameters, the adjustment process is constrained to ensure that the adjusted renderings remain compatible with the characteristics of the wood. Based on the processing accuracy data, adjust the width and depth of the pattern lines so that the line width is greater than or equal to the minimum engraving width of the corresponding processing tool, and the depth is less than or equal to the maximum cutting depth. The rendering can be regenerated using the generated network, or the existing rendering can be directly modified to output the final wooden door style design scheme.

[0011] Optionally, the feature parameters are input into a classification network, which outputs the species identifier of the wood sample and the difference parameters between the wood sample and the standard material, including: The feature parameters are input into the input layer of the classification network, whereby the input layer receives the feature parameters and converts them into an internal representation; The data represented internally is transformed layer by layer through multiple hidden layers of the classification network. Each hidden layer contains multiple computational units, which perform weighted summation on the input data and apply a nonlinear function. The first output unit of the output layer calculates the species identifier of the wood sample based on the transformed data and by comparing the calculated value with the pre-stored species threshold. The second output unit of the output layer calculates the difference parameters between the transformed data and the standard material. The difference parameters include color deviation value, texture deviation value and grain size deviation value. The color deviation value is obtained by calculating the difference between the input feature parameter and the standard color parameter. The texture deviation value is obtained by calculating the difference between the input feature parameter and the standard texture parameter. The grain size deviation value is obtained by calculating the difference between the input feature parameter and the standard grain size parameter.

[0012] Optionally, the decoder portion progressively expands the latent space representation through multiple decoding layers to generate an initial rendering, including: The latent space representation is input into the first decoding layer in the multi-layer decoding layer. The first decoding layer performs a first expansion operation on the input data, increasing the data size by increasing the number of data points to obtain the first expanded data. The first extended data is input into the second decoding layer in the multi-layer decoding layer. The second decoding layer performs a second expansion operation on the first extended data, further increasing the data size by adjusting the arrangement of data points to obtain the second extended data. The second extended data is input into the third decoding layer in the multi-layer decoding layer. The third decoding layer performs a third extension operation on the second extended data to generate initial data in the image format by changing the distribution of data points. During the expansion operation of each decoding layer, a difference parameter is introduced as a guiding condition. By combining the difference parameter with the data of the current layer, the direction and extent of the expansion operation are controlled, so that the generated initial data remains consistent with the characteristics of the wood. The initial data is then adjusted to output the corresponding initial effect diagram.

[0013] Secondly, this application provides a deep learning-based intelligent generation system for wooden door styles, comprising: The acquisition module is used to build a wood database containing physical properties and aesthetic parameters of different wood species, and at the same time to acquire surface morphology images of wood samples provided by users through image acquisition devices; The analysis module is used to analyze various feature parameters, including the direction of wood grain, contrast degree, and grain size, from the surface morphology image, and input the feature parameters into the classification network, and output the type identifier of the wood sample and the difference parameters between the wood sample and the standard material through the classification network. The generation module is used to input the style instructions and the difference parameters into the material-aware generation network, and combine the constraints of wood characteristics in the difference parameters to generate a wooden door style rendering that matches the wood characteristics. At the same time, it associates and maps the processing accuracy data of different processing tools on wood in the preset processing technology constraint library with the physical parameters in the wood database. The verification module is used to evaluate the structural rationality of the wooden door style rendering, and to verify the feasibility of the design pattern on the corresponding wood in multiple dimensions by combining the associated mapping physical parameters and processing accuracy data, and to obtain the verification results. The adjustment module is used to adaptively adjust the wooden door style rendering based on the verification results, and adjust the generation parameters based on the difference parameters and the processing accuracy data during the adjustment process, and output a wooden door style design scheme that meets the actual production requirements.

[0014] Thirdly, this application provides an electronic device, comprising: Memory, used to store computer programs; A processor is configured to execute the computer program to implement the steps of a deep learning-based intelligent generation method for wooden door styles as described in the first aspect above.

[0015] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps of the deep learning-based intelligent generation method for wooden door styles described in the first aspect above.

[0016] This application presents a deep learning-based intelligent generation method for wooden door styles. By constructing a wood database containing physical properties and aesthetic parameters and collecting sample images, it provides a comprehensive data foundation for subsequent analysis. By analyzing feature parameters from images and inputting them into a classification network, it can accurately identify wood species and quantify their differences from standard materials. By inputting style instructions and difference parameters into a material-aware generation network and associating them with processing constraints, it can generate feasible effect diagrams that meet aesthetic requirements while respecting material properties. Through a verification module, it conducts multi-dimensional evaluations of structural rationality and feasibility, enabling the early detection and avoidance of manufacturing risks. Finally, based on the verification results, it adjusts the generation parameters in a coordinated manner, outputting design schemes that truly meet actual production requirements.

[0017] Furthermore, the style instructions and difference parameters are encoded into a joint vector, compressed to obtain a latent space representation, and then expanded by a decoder to generate an initial image. During the decoding process, the output is dynamically adjusted with the difference parameters as constraints to ensure that the texture, color, and other characteristics are consistent with the wood sample. The technical effect lies in achieving a deep coupling between style creativity and material physical properties, making the generated image not only visually realistic but also highly realistic and achievable in terms of material representation, avoiding material mismatch problems caused by subjective design. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 A flowchart illustrating a deep learning-based intelligent generation method for wooden door styles provided in this application embodiment; Figure 2 This is a schematic diagram of the structure of a deep learning-based intelligent generation system for wooden door styles, provided in an embodiment of this application. Detailed Implementation

[0020] Currently, while deep learning-based wood door style generation technology can quickly produce visual solutions, its core flaw lies in the "disconnect between design and manufacturing": the generation model only learns the aesthetic features of the image, completely ignoring the physical properties of wood as a natural material (such as hardness and grain direction) and the technological constraints in actual processing (such as carving precision). This often results in the generated renderings becoming "digital artworks"—beautiful in appearance but unable to accurately guide production, or forced to undergo significant modifications due to improper material selection or impractical processes, thus prolonging the design cycle and increasing costs.

[0021] To address this fundamental problem, this application proposes an intelligent generation method that integrates material perception and process integration. This method first constructs a wood database and analyzes sample images to accurately identify wood species and quantify their differences from standard materials. Then, these difference parameters, along with user style instructions, are input into the generation network to ensure that the generated textures, colors, and other visual effects are strictly constrained by the characteristics of real wood. Most importantly, it innovatively introduces a processing technology constraint library and performs associated mapping. A verification module performs multi-dimensional simulation verification of the structural rationality and feasibility of the generated scheme, and dynamically adjusts design parameters based on the results. This solution fundamentally bridges the gap between "design" and "manufacturing," making the generated renderings no longer isolated images, but digital production guidance schemes with both aesthetic value and manufacturing feasibility. It effectively solves the problem of unusable schemes in existing technologies due to their detachment from the constraints of the physical world.

[0022] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0023] The core of this application is to provide a method for intelligent generation of wooden door styles based on deep learning, and a flowchart of one specific implementation is shown below. Figure 1 As shown, the method includes: S101. Construct a wood database containing physical properties and aesthetic parameters of different wood species, and simultaneously acquire surface morphology images of wood samples provided by users through image acquisition devices. In the above scheme, the wood database refers to a structured data set that centrally stores physical properties of different wood species, such as density, hardness, moisture content, elastic modulus, and aesthetic parameters such as texture pattern, color distribution, and gloss. Surface morphology images refer to digital images of the micro and macro features of the surface texture, particles, and pores of wood samples obtained through image acquisition devices such as high-resolution scanners or macroscopic photography systems. Physical properties are quantitative indicators that describe the mechanical and physical behavior of wood, while aesthetic parameters are subjective and objective descriptive quantities used to characterize the visual properties of wood.

[0024] In this embodiment, firstly, a data acquisition module integrates the physical properties and aesthetic parameters of wood from wood standard manuals, industry databases, and laboratory measurement records to construct a wood database containing multi-dimensional data. For example, the density of wood species A is recorded as 650 kg / m³, the hardness as Brinell 3.2, and the grain type as straight grain using a structured storage method. Secondly, surface morphology images of user-provided wood samples are acquired using an image acquisition device such as a high-resolution linear scanner. For example, a user-provided oak sample is scanned under multi-angle illumination to generate an RGB image with a resolution of 1200 pixels per inch. An image preprocessing process is then used for noise reduction, brightness equalization, and geometric correction to ensure clear and standardized image features. Finally, the preprocessed image is initially aligned with reference images in the database to provide input for subsequent feature analysis.

[0025] In practical applications, taking Company A, a custom wooden door manufacturer, as an example, it first extracts physical parameters such as density range of 450 to 750 kg per cubic meter and aesthetic parameters such as texture contrast level from the industry standard library for 10 types of wood, including pine and walnut, to build a local database. Then, when user B provides a sample of an unknown type of wood, Company A uses a desktop scanner C to acquire an image of the sample surface under constant lighting conditions, generating a digital image of 2000×2000 pixels. The image is then processed using image processing software to remove the background and perform color calibration. Finally, the calibrated image is initially matched with the standard texture image in the database to complete the data preparation stage.

[0026] This solution provides a high-quality and consistent data foundation for subsequent analysis by systematically integrating physical and aesthetic data of wood and combining it with a standardized image acquisition process. This ensures the reliability of input for species identification and style generation, while reducing errors caused by missing data or differences in image quality.

[0027] S102. Analyze the feature parameters of various types, including wood door texture direction, contrast degree and texture grain, from the surface morphology image, and input the feature parameters into the classification network, and output the type identification of the wood sample and the difference parameters between the wood sample and the standard material through the classification network; Optionally, step S102 may specifically include the following steps: S1021. The feature parameters are input into the input layer of the classification network, and the input layer receives the feature parameters and converts them into an internal representation; S1022. The data represented internally is transformed layer by layer through multiple hidden layers of the classification network. Each hidden layer contains multiple computing units. The computing units perform weighted summation on the input data and apply a nonlinear function. S1023. The first output unit of the output layer calculates the species identifier of the wood sample based on the transformed data and by comparing the calculated value with the pre-stored species threshold. S1024. The second output unit of the output layer calculates the difference parameters between the transformed data and the standard material, the difference parameters including color deviation value, texture deviation value and grain size deviation value; S1025. The color deviation value is obtained by calculating the difference between the input feature parameter and the standard color parameter; the texture deviation value is obtained by calculating the difference between the input feature parameter and the standard texture parameter; and the grain size deviation value is obtained by calculating the difference between the input feature parameter and the standard grain size parameter.

[0028] In the above scheme, feature parameters refer to quantitative indicators extracted from the wood surface morphology image, including texture direction (the principal angle of wood fiber orientation), contrast (the intensity of difference between light and dark areas of texture), and texture graininess (the coarseness or density of texture features). The classification network is an artificial neural network model used for pattern recognition and classification based on input features. The species identifier is the classification result output by the network to indicate the specific type of wood, such as walnut or pine. The difference parameters are numerical values ​​used to quantify the differences between user samples and standard samples in terms of color, texture, and graininess, including color deviation value (the Euclidean distance in the color space), texture deviation value (the difference measure of texture feature vectors), and graininess deviation value (the average difference in particle size distribution).

[0029] In this embodiment, firstly, the feature parameters are input to the input layer of the classification network via S1021 and undergo transformation processing. The extracted texture direction θ, contrast C, and texture granularity G are input to the input layer of the classification network. This layer converts each parameter into a standardized value using a linear normalization method. For example, when processing a wood sample provided by user B, the 82-degree texture direction is converted to 0.82, the 0.78 contrast remains unchanged, and the high granularity is converted to 0.9, ultimately forming a feature vector [0.82, 0.78, 0.9] as the network input. Secondly, through the hidden layer feature transformation processing in S1022, the input features are nonlinearly transformed through multiple hidden layers of the classification network, where the computational unit of each hidden layer performs a weighted summation operation. ; in This represents the connection weight between the j-th neuron in layer l and the i-th neuron in layer (l-1). For bias terms, This is the output of the previous layer. For example, the first hidden layer expands the 3D input vector into a 128-dimensional feature representation through the above calculation. Then, through the category identification calculation in S1023, the softmax function is used in the first output unit of the output layer to compare the calculated value with the pre-stored category threshold to calculate the category probability. ; in This represents the linear output of the k-th node in the output layer, where K is the total number of categories. For example, if the calculated probability of walnut wood is 0.92 and the probability of pine wood is 0.05, since 0.92 exceeds the threshold of 0.9, the walnut wood category identifier is output. Simultaneously, through S1024, the second output unit in the output layer calculates the difference parameter between the transformed data and the standard material, using the Euclidean distance formula in the LAB color space for the color deviation value. ; Where L represents lightness, a represents red-green value, and b represents yellow-blue value. For example, the standard walnut wood LAB value is [60, 8, 25], and the sample value is [58, 10, 22], then the calculation is: ; Texture deviation values ​​are calculated using cosine similarity: ; Where A and B are the texture feature vectors of the sample and the standard sample, respectively. For example, if the similarity is calculated to be 0.88, then the texture deviation value is 1 - 0.88 = 0.12. Finally, the distribution difference is calculated using the Earth Mover's Distance algorithm based on the S1025 granularity deviation value. The mathematical expression is as follows: ; Where P and Q are two particle size distributions, respectively. This represents the optimal flow from cluster i of P to cluster j of Q. This represents the inter-cluster distance. For example, the calculated EMD value between the sample and standard walnut wood is 0.05, indicating a small difference in grain size distribution. The final output includes a complete analysis result containing species identification and three difference parameters, providing a quantitative basis for subsequent style generation.

[0030] In practical applications, within Company A's custom wooden door system, after feature extraction is completed for user B's wood sample image, the system inputs parameters such as texture direction (82 degrees), contrast (0.78), and grain size level into a pre-defined classification network. The network input layer converts these parameters into vectors of 0.82, 0.78, and 0.9. After processing through three hidden layers, the final output layer identifies the sample as a walnut wood species identifier and calculates a color deviation of 4.1 (due to the sample's reddish hue), a texture deviation of 0.12 (due to denser texture), and a grain size deviation of 0.05 (due to slightly finer grains) compared to standard walnut wood. These parameters provide a quantitative basis for subsequent style generation.

[0031] This solution achieves accurate identification of wood species and quantification of differences in standard materials through automated feature extraction and neural network classification, providing reliable data input for style adaptation generation, while improving the objectivity and consistency of the identification process.

[0032] S103. Input the style instructions and the difference parameters into the material-aware generation network, and combine the constraints of wood characteristics in the difference parameters to generate a wooden door style effect picture that matches the wood characteristics. At the same time, associate and map the processing accuracy data of different processing tools on wood in the preset processing technology constraint library with the physical parameters in the wood database.

[0033] Optionally, step S103 may specifically include the following steps: S1031. The style instruction and the difference parameter are input into the encoder part of the generator network, and the encoder part merges the style instruction and the difference parameter into a joint vector; S1032. The joint vector is progressively compressed through multiple coding layers of the encoder part to obtain the latent space representation; S1033. The latent space representation is input into the decoder part of the generator network, and the decoder part expands the latent space representation step by step through multiple decoding layers to generate an initial effect diagram; S1034. During the processing of the decoder section, the difference parameter is introduced as a constraint condition. By adjusting the output value of the decoding layer, the texture direction, color distribution and grain pattern in the initial effect image are made consistent with the wood characteristics described in the difference parameter, and a wooden door style effect image is output.

[0034] S1035. Obtain machining accuracy data of multiple different machining tools from the preset machining process constraint library. The machining accuracy data of each machining tool includes the minimum engraving width, the maximum cutting depth, and the minimum drilling diameter. S1036. Obtain physical parameters of multiple wood species from the wood database. The physical parameters of each wood species include hardness value, density value and brittleness value. S1037. For each processing tool, a mapping function is defined based on the relationship between the processing accuracy data and the physical parameters. The mapping function takes the physical parameter value of the wood as input and outputs the actual achievable accuracy value of the processing tool on the wood, thus obtaining the corresponding mapping function.

[0035] Specifically, step S1033 may include the following process: inputting the latent space representation into the first decoding layer of the multi-layer decoding layer, the first decoding layer performs a first expansion operation on the input data, increasing the data size by increasing the number of data points to obtain first expanded data; inputting the first expanded data into the second decoding layer of the multi-layer decoding layer, the second decoding layer performs a second expansion operation on the first expanded data, further increasing the data size by adjusting the arrangement of data points to obtain second expanded data; inputting the second expanded data into the third decoding layer of the multi-layer decoding layer, the third decoding layer performs a third expansion operation on the second expanded data, generating initial data in image format by changing the distribution of data points; during the expansion operation of each decoding layer, a difference parameter is introduced as a guiding condition, and by combining the difference parameter with the data of the current layer, the direction and degree of the expansion operation are controlled to keep the generated initial data consistent with the characteristics of wood, the initial data is finally adjusted, and the corresponding initial effect image is output.

[0036] In the above scheme, style instructions refer to user-specified design style requirements, such as modern minimalist or classical carving; difference parameters refer to quantitative difference indicators such as color deviation value, texture deviation value, and grain size deviation value output from the classification network; the material-aware generative network is a deep learning model based on conditional generative adversarial networks, which can generate effect images that conform to physical laws according to material characteristics; the processing technology constraint library refers to a structured database that stores the precision parameters of various processing tools; and the association mapping refers to establishing a mathematical relationship model between the precision of processing tools and the physical parameters of wood.

[0037] In this embodiment, firstly, through S1031 joint vector generation processing, style instructions and difference parameters are input into the encoder part of the generator network, and the two types of inputs are merged into a joint vector using vector concatenation. For example, user B specifies a modern minimalist style instruction and difference parameters: color deviation value 4.1, texture deviation value 0.12, and granularity deviation value 0.05, which are merged into a six-dimensional joint vector. Secondly, through S1032 latent space compression processing, the three encoding layers of the encoder part progressively reduce the dimensionality of the joint vector. Each encoding layer uses a fully connected layer and batch normalization operation. For example, the first encoding layer compresses the six-dimensional input into four dimensions, and the second encoding layer further compresses it into two dimensions, finally obtaining the latent space representation of the design elements. Then, through S1033 layered expansion generation processing, the latent space representation is input into the decoder part of the generator network: the first decoding layer expands the two-dimensional data into an 8×8×64 feature map through deconvolution operation, the second decoding layer continues to expand it into a 16×16×32 feature map, and the third decoding layer finally generates 256×256×3 RGB image data. In each extended operation, a difference parameter is introduced as a constraint to generate an initial effect image. For example, in the second decoding layer, the texture deviation value of 0.12 is fused with the feature map using channel attention to ensure that the generated texture features are consistent with the characteristics of wood. Simultaneously, the difference parameter is introduced as a constraint in S1034. During the decoding process, an adaptive instance normalization technique is used to integrate the difference parameter into the generation process to output the wooden door style effect image. Specifically, style control is achieved using the following formula: ; Where x is the feature map, y is the difference parameter, and μ and σ represent the mean and standard deviation, respectively. For example, by adjusting the color deviation value of 4.1, the color distribution of the generated image is controlled to ensure that the effect image is consistent with the actual color difference of the wood. Finally, through the processing mapping process of S1035-S1037, the processing accuracy data of multiple different processing tools are obtained from the preset processing technology constraint library. The processing accuracy data of each processing tool includes the minimum sculptable width, the maximum cutting depth, and the minimum drillable diameter; physical parameters such as the walnut hardness value of 5.2 are obtained from the wood database. The physical parameters of each wood species include hardness value, density value, and brittleness value; a mapping function is established: ; in, For practically achievable accuracy, Here, H represents the nominal precision of the tool, H is the hardness value of the wood, and k is the material influence coefficient. For example, the actual carving width of the engraving machine on walnut wood is calculated to be 0.5×e^{-0.1×5.2}≈0.3mm.

[0038] In practical applications, in Company A's design system, after user B selects a modern minimalist style, the system inputs the style instructions and the difference parameters of the walnut wood sample into the generation network; the encoder generates a joint vector and then compresses it through three layers to obtain the latent representation; the decoder generates an initial rendering through three layers of expansion operations, during which the difference parameters are used to control texture generation; at the same time, the system queries the engraving machine precision data in the processing constraint library and calculates the actual achievable engraving precision in combination with the walnut wood hardness value to ensure the feasibility of the design pattern.

[0039] This solution organically combines design style with the characteristics of wood through an intelligent generative network, producing aesthetically pleasing design schemes that conform to the essential characteristics of the material. At the same time, it ensures the feasibility of the design schemes through process constraint mapping, achieving a seamless connection from virtual design to actual manufacturing.

[0040] S104. The structural rationality of the wooden door style rendering is evaluated by the verification module. Combined with the physical parameters and processing accuracy data after the correlation mapping, the feasibility of the design pattern on the corresponding wood is verified in multiple dimensions to obtain the verification results. Optionally, step S104 may specifically include the following steps: S1041. Extract design patterns from the wooden door style renderings, including pattern line width, pattern depth, and pattern complexity. S1042. The structural evaluation unit of the verification module calculates the support strength value of the pattern and compares the support strength value with a preset threshold to check whether the features of the design pattern meet the structural stability requirements. S1043. The processing feasibility unit of the verification module uses the physical parameters and processing accuracy data after the association mapping to check whether the design pattern is processable by comparing the line width with the minimum engraving width of the corresponding processing tool for each line in the pattern. S1044. Output verification results including structural rationality indicators and processing feasibility indicators.

[0041] In the above scheme, structural rationality assessment refers to the process of systematically examining the mechanical stability and structural integrity of the design pattern; the design pattern refers to the set of decorative elements extracted from the rendering, including the pattern line width (i.e., the physical width of the carved lines), the pattern depth (i.e., the depth of the carved recessed parts), and the pattern complexity (i.e., the complexity and interrelationship of the geometric elements in the pattern); the support strength value refers to the quantitative index of the pattern structure's ability to remain stable under stress; the structural stability requirement refers to the basic mechanical standard that ensures the wooden door does not deform or become damaged under normal use conditions; and the processing feasibility refers to the evaluation index of whether the design pattern can be accurately realized using existing processing tools.

[0042] In this embodiment, firstly, through pattern feature extraction processing (S1041), an image segmentation algorithm is used to identify and quantify the key features of the design pattern from the wooden door style rendering. Edge detection technology is used to extract line contours, and a depth estimation algorithm is used to calculate the pattern depth distribution. Topological analysis is then used to evaluate the pattern complexity. For example, analyzing the modern minimalist style rendering of User B, the main decorative line width is extracted to be 0.3 mm, the pattern depth is 2 mm, and the complexity level is medium. Secondly, through structural stability verification processing (S1042), the structural evaluation unit of the verification module calculates the support strength value of the pattern using the finite element analysis method. The calculation formula is: ; Where E is the elastic modulus of wood, w is the line width, d is the pattern depth, and l is the span length. For example, to calculate the support strength value of walnut wood, where E is taken as 10 GPa, w as 0.3 mm, d as 2 mm, and l as 10 mm, then: ; The support strength value is compared with a preset threshold of 5 × 10^4 Pa. Since 6 × 10^4 Pa is greater than the threshold, it is determined that the structural stability requirements are met. Then, through the processing feasibility verification process S1043, the width of each line in the pattern is compared with the minimum engraving width of the processing tool. For example, the minimum line width in User B's design is 0.3 mm, while the actual engraving width of the engraving machine C on walnut wood is 0.25 mm. Since 0.3 mm is greater than 0.25 mm, the design pattern can be accurately processed. Finally, through the verification result output process S1044, a comprehensive verification result including structural rationality and processing feasibility indicators is generated. For example, if the output structural rationality indicator is "passed" and the processing feasibility indicator is "passed," it indicates that the design pattern meets both the structural stability requirements and processing feasibility.

[0043] In practical applications, in Company A's design verification system, the design renderings of user B's walnut wood door are verified from multiple dimensions: First, the pattern features are extracted to obtain parameters such as the line width of 0.3 mm; then, the support strength value of 6 × 10^4 Pa is calculated, which is higher than the threshold of 5 × 10^4 Pa, thus determining that the structure is reasonable; next, the line width is compared with the processing capability of the engraving machine, confirming that 0.3 mm is greater than the minimum engraving width of 0.25 mm, thus determining that the processing is feasible; finally, the verification result with double marks is output.

[0044] This solution, through a systematic verification process, ensures that the design pattern meets both structural mechanics requirements and is feasible for actual processing, effectively avoiding design defects and production problems, and improving the success rate of the design solution.

[0045] S105. Based on the verification results, make adaptive adjustments to the wooden door style rendering, and adjust the generation parameters based on the difference parameters and the processing accuracy data during the adjustment process, and output a wooden door style design scheme that meets the actual production requirements.

[0046] Optionally, step S105 may specifically include the following steps: S1051. If the verification results indicate that the structure is unreasonable or the processing is not feasible, then adjust the generation parameters in the generator network, including texture generation intensity, color adjustment factor and line scaling factor. S1052. Based on the difference parameters, constrain the adjustment process to ensure that the adjusted rendering remains compatible with the characteristics of the wood. S1053. Based on the processing accuracy data, adjust the width and depth of the pattern lines so that the line width is greater than or equal to the minimum engraving width of the corresponding processing tool, and the depth is less than or equal to the maximum cutting depth. S1054. Regenerate the rendering through the generating network, or directly modify the existing rendering, and output the final wooden door style design scheme.

[0047] In the above scheme, adaptive adjustment refers to the optimization process of making targeted modifications to the design pattern based on the verification results; generation parameters refer to adjustable variables that control the output characteristics of the generation network, including texture generation intensity, which is a coefficient that controls the degree of texture clarity, color adjustment factor, which is a parameter that adjusts color saturation and brightness, and line scaling factor, which is an adjustment coefficient that changes the proportion of line size; wood property adaptability refers to the degree of coordination between the design pattern and the natural properties of wood; actual production requirements refer to the design standards that meet the limitations of processing tools and the requirements of structural stability.

[0048] In this embodiment, the system first generates parameter adjustment processing via S1051. When the verification result shows that the structure is unreasonable or processing is infeasible, the system automatically adjusts the generation parameters in the generation network, including texture generation intensity, color adjustment factor, and line scaling factor. For example, if user B's design pattern is deemed infeasible to process because the line width of 0.25mm is less than the minimum sculptable width of 0.3mm, the system adjusts the line scaling factor from 1.0 to 1.2 and calculates the new line width according to the formula: ; in, The original width is 0.25mm. A scaling factor of 1.2 results in a new width of 0.3mm. Next, the S1052 material property preservation process introduces a difference parameter as a constraint during adjustment to ensure the modified design maintains compatibility with wood properties. For example, if User B's walnut wood sample has a color deviation value of 4.1, the system controls the color variation range using a color adjustment factor to ensure the adjusted image's color difference remains within a reasonable range of 3.5-4.5. Then, the S1053 processing adaptability optimization process adjusts the pattern's line width and depth based on processing accuracy data. For example, if the maximum cutting depth of engraving machine C is 3mm, while a certain pattern depth in User B's design is 3.2mm, the system adjusts it according to the formula: ; in, The original depth is 3.2mm. To achieve a maximum cutting depth of 3mm, a new depth of 3.0mm is obtained. Finally, the system is regenerated using the S1054 scheme, employing parametric modifications to directly adjust the existing rendering or to regenerate the complete rendering via a generative network. For example, the system maintains User B's modern minimalist style, only adjusting the line width from 0.25mm to 0.3mm and the pattern depth from 3.2mm to 3.0mm, outputting the final wooden door design scheme.

[0049] In practical applications, in Company A's design optimization system, User B's walnut door design was deemed unfeasible to process due to some lines being too thin. The system automatically adjusted the line scaling factor, increasing the minimum line width from 0.25mm to 0.3mm. At the same time, it constrained the color adjustment range according to the color deviation value of 4.1 to maintain the natural color characteristics of the wood. The final output was an optimized solution that met both the processing capabilities of the engraving machine and maintained the aesthetic effect.

[0050] This solution, through intelligent parameter adjustment and constraint optimization, ensures that the design fully meets actual production requirements while maintaining artistic effect, effectively solving the connection problem between design and manufacturing, and improving the success rate of the design solution.

[0051] This application presents a deep learning-based intelligent generation method for wooden door styles. Step 101 involves constructing a wood database containing multi-dimensional physical properties and aesthetic parameters, and collecting standardized surface images to establish an accurate and reliable material data foundation. Step 102 employs neural networks to intelligently analyze texture features and accurately identify wood species and quantify material differences, achieving digital representation of material characteristics. Step 103 integrates design styles with material characteristics and correlates them with processing constraints, creating a design scheme that combines aesthetic value and manufacturing feasibility. Step 104 verifies the design scheme from multiple dimensions, including structural mechanics and processing feasibility, ensuring the practicality and safety of the design results. Step 105 optimizes and adjusts the design based on the verification results and outputs the final scheme, achieving a seamless transition from virtual design to actual manufacturing. This entire method, through an intelligent technology chain, automates, refines, and enhances the practicality of wooden door style design, significantly improving design efficiency and output quality.

[0052] Figure 2 This application provides a schematic diagram of a specific implementation of a deep learning-based intelligent generation system for wooden door styles, referring to... Figure 2 The system may include: The acquisition module 21 is used to construct a wood database containing physical properties and aesthetic parameters of different wood species, and at the same time to acquire surface morphology images of wood samples provided by the user through an image acquisition device. Analysis module 22 is used to analyze various feature parameters, including wood door texture direction, contrast degree and texture grain, from the surface morphology image, and input the feature parameters into the classification network, and output the type identification of the wood sample and the difference parameters between the wood sample and the standard material through the classification network; The generation module 23 is used to input the style instructions and the difference parameters into the material-aware generation network, and combine the constraints of wood characteristics in the difference parameters to generate a wooden door style effect picture that matches the wood characteristics. At the same time, it associates and maps the processing accuracy data of different processing tools on wood in the preset processing technology constraint library with the physical parameters in the wood database. Verification module 24 is used to evaluate the structural rationality of the wooden door style rendering through the verification module, and to verify the feasibility of the design pattern on the corresponding wood in multiple dimensions by combining the physical parameters and processing accuracy data after association mapping, so as to obtain the verification result. The adjustment module 25 is used to adaptively adjust the wooden door style rendering based on the verification results, and adjust the generation parameters based on the difference parameters and the processing accuracy data during the adjustment process, and output a wooden door style design scheme that meets the actual production requirements.

[0053] This application provides an embodiment of a deep learning-based intelligent generation system for wooden door styles, which is used to implement the aforementioned deep learning-based intelligent generation method for wooden door styles. Therefore, the specific implementation of the deep learning-based intelligent generation system for wooden door styles can be found in the embodiment section of the deep learning-based intelligent generation method for wooden door styles described above. The specific implementation can be referred to the description of the corresponding embodiments, and will not be repeated here.

[0054] This application also provides an electronic device, including: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of the deep learning-based intelligent generation method for wooden door styles described above.

[0055] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of any of the above-described deep learning-based intelligent generation methods for wooden door styles.

[0056] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory, random access memory, portable hard drives, magnetic disks, or optical disks.

[0057] Embodiments of the present invention also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the embodiments of the deep learning-based intelligent generation method for wooden door styles.

[0058] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0059] The above provides a detailed description of a deep learning-based intelligent generation method, system, electronic device, and storage medium for wooden door styles provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A deep learning-based intelligent generation method for wooden door styles, characterized in that, The method comprises the following steps: constructing a wood database containing physical properties and aesthetic parameters of different wood species, while obtaining surface topography images of user-provided wood samples through an image acquisition device; analyzing feature parameters of various types including wood door texture direction, contrast degree, and texture granularity from the surface topography images, and inputting the feature parameters into a classification network to output the species identification of the wood sample and difference parameters from standard materials; inputting style instructions and the difference parameters into a material-aware generation network, and combining the wood characteristics in the difference parameters to generate wood door style renderings that adapt to the wood characteristics, while associating and mapping the processing accuracy data of different processing tools in a preset processing technology constraint library with the physical parameters in the wood database; performing structural rationality evaluation on the wood door style renderings through a verification module, and performing multidimensional verification on the implementability of the design patterns on the corresponding wood based on the associated and mapped physical parameters and processing accuracy data to obtain verification results; performing adaptive adjustment on the wood door style renderings according to the verification results, and generating adjustment parameters based on the difference parameters and the processing accuracy data during the adjustment process to output a wood door style design scheme that meets actual production requirements.

2. The method of claim 1, wherein, inputting style instructions and the difference parameters into a material-aware generation network, and combining the wood characteristics in the difference parameters to generate wood door style renderings that adapt to the wood characteristics, comprising: inputting style instructions and difference parameters into an encoder part of the generation network, which combines the style instructions and the difference parameters into a joint vector; gradually compressing the joint vector through multiple encoding layers of the encoder part to obtain a latent space representation; inputting the latent space representation into a decoder part of the generation network, which gradually expands the latent space representation through multiple decoding layers to generate an initial rendering; during the processing of the decoder part, introducing the difference parameters as a constraint condition, adjusting the output value of the decoding layer to make the texture direction, color distribution, and granularity mode in the initial rendering consistent with the wood characteristics described in the difference parameters, and outputting a wood door style rendering.

3. The method of claim 1, wherein, associating and mapping the processing accuracy data of different processing tools in a preset processing technology constraint library with the physical parameters in the wood database, comprising: obtaining the processing accuracy data of multiple different processing tools from the preset processing technology constraint library, and the processing accuracy data of each processing tool includes the minimum engraveable width, the maximum cuttable depth, and the minimum drillable diameter; obtaining the physical parameters of multiple wood species from the wood database, and the physical parameters of each wood species include the hardness value, the density value, and the brittleness value; for each processing tool, defining a mapping function according to the relationship between the processing accuracy data and the physical parameters, which inputs the wood physical parameter value and outputs the actual achievable accuracy value of the processing tool on the wood, to obtain the corresponding mapping function.

4. The method of claim 1, wherein, The structure rationality of the wooden door style effect picture is evaluated by the verification module, and the implementability of the design pattern on the corresponding wood is verified in multiple dimensions by combining the associated mapped physical parameters and processing precision data, to obtain a verification result, including: Extracting the design pattern including pattern line width, pattern depth and pattern complexity from the wooden door style effect picture; Using the structure evaluation unit of the verification module to calculate the support strength value of the pattern, and comparing the support strength value with a preset threshold to check whether the characteristics of the design pattern meet the structural stability requirements; Using the processing feasibility unit of the verification module to compare the line width with the minimum engraveable width of the corresponding processing tool for each line in the pattern according to the associated mapped physical parameters and processing precision data to check whether the design pattern is processable; Outputting the verification result including the structure rationality mark and the processing feasibility mark.

5. The method of claim 1, wherein, According to the verification result, the wooden door style effect picture is adaptively adjusted, and adjustment generation parameters are generated based on the difference parameters and the processing precision data during the adjustment process, and a wooden door style design scheme meeting the actual production requirements is output, including: If the verification result indicates that the structure is not reasonable or the processing is not feasible, generate the generation parameters including texture generation strength, color adjustment factor and line scaling factor in the generation network; According to the difference parameters, the adjustment process is constrained to ensure that the adjusted effect picture remains adaptive to the characteristics of the wood; According to the processing precision data, the pattern line width and depth are adjusted so that the line width is greater than or equal to the minimum engraveable width of the corresponding processing tool, and the depth is less than or equal to the maximum cuttable depth; The effect picture is regenerated through the generation network, or the existing effect picture is directly modified, and the final wooden door style design scheme is output.

6. The method of claim 1, wherein, The feature parameters are input into the classification network, and the species identification of the wood sample and the difference parameters from the standard material are output by the classification network, including: The feature parameters are input into the input layer of the classification network, and the input layer receives the feature parameters and converts them into internal representations; The data of the internal representations are transformed layer by layer by multiple hidden layers of the classification network, each hidden layer containing multiple calculation units that perform weighted summation on the input data and apply a nonlinear function; The species identification of the wood sample is calculated based on the transformed data at the first output unit of the output layer by comparing the calculation value with the pre-stored species threshold value; The difference parameters from the standard material are calculated based on the transformed data at the second output unit of the output layer, including color deviation value, texture deviation value and granularity deviation value; The color deviation value is obtained by calculating the difference between the input feature parameters and the standard color parameters, the texture deviation value is obtained by calculating the difference between the input feature parameters and the standard texture parameters, and the granularity deviation value is obtained by calculating the difference between the input feature parameters and the standard granularity parameters.

7. The method of claim 2, wherein, The decoder part expands the latent space representation step by step through multiple decoding layers to generate an initial effect picture, including: inputting the latent space representation into a first decoding layer in a multi-layer decoding layer, the first decoding layer performing a first expansion operation on the input data to increase the data size by increasing the number of data points, to obtain first expanded data; inputting the first expanded data into a second decoding layer in the multi-layer decoding layer, the second decoding layer performing a second expansion operation on the first expanded data to further increase the data size by adjusting the arrangement of data points, to obtain second expanded data; inputting the second expanded data into a third decoding layer in the multi-layer decoding layer, the third decoding layer performing a third expansion operation on the second expanded data to generate initial data in an image format by changing the distribution of data points; in the expansion operation process of each decoding layer, a difference parameter is introduced as a guide condition, the direction and degree of the expansion operation are controlled by combining the difference parameter with the data of the current layer, so that the generated initial data maintains consistency with the wood characteristics, and the initial effect picture is output after final adjustment of the initial data.

8. A deep learning-based wooden door style intelligent generation system, characterized in that, comprises: a collection module configured to construct a wood database containing physical attributes and aesthetic parameters of different wood species, and to obtain a surface topography image of a wood sample provided by a user through an image collection device; an analysis module configured to analyze feature parameters including wood door texture direction, contrast degree, and texture granularity from the surface topography image, and to input the feature parameters into a classification network to output a species identification of the wood sample and a difference parameter with a standard material through the classification network; a generation module configured to input a style instruction and the difference parameter into a material perception generation network, and to generate a wood door style effect picture adapted to the wood characteristics by combining the difference parameter with a constraint of the wood characteristics in the material perception generation network, and to associate and map processing precision data of different processing tools in a preset processing process constraint library with physical parameters in the wood database; a verification module configured to perform structural rationality evaluation on the wood door style effect picture through the verification module, to perform multidimensional verification on the implementability of the design pattern on the corresponding wood by combining the associated and mapped physical parameters and processing precision data, and to obtain a verification result; an adjustment module configured to adaptively adjust the wood door style effect picture according to the verification result, to generate adjustment parameters based on the difference parameter and the processing precision data during the adjustment, and to output a wood door style design scheme meeting actual manufacturing requirements.

9. An electronic device, comprising: comprises: a memory configured to store a computer program; a processor configured to execute the computer program to implement the steps of the wood door style intelligent generation method based on deep learning according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer program is stored in the computer readable storage medium, and the computer program is executed by the processor to implement the wood door style intelligent generation method based on deep learning according to any one of claims 1 to 7.

Images