Dynamic visual identification adaptive design system based on AI parameterization generation

Abstract

The application relates to the field of computer vision design, in particular to a dynamic visual identification adaptive design system based on AI parameterization generation, which integrates data collection, processing, parameterization generation, emotional feedback and dynamic output, realizes an automatic process from brand semantics to dynamic visual identification, and accurately captures brand official website basic data such as standard colors and logo patterns through a data collection module; a data processing module deeply analyzes semantics, extracts theme words and parameterization control data; a parameterization generation module constructs an SVG layer model according to the data, is flexible and controllable, an emotional feedback module integrates user emotional data, realizes adaptive adjustment of the visual identification, deepens brand and user emotional connection, and a dynamic output module combines a physical engine to create a lively visual identification.

Description

Technical Field

[0001] This invention relates to the field of computer vision design, and in particular to a dynamic visual identifier adaptive design system based on artificial intelligence technology, parametric design methods, and emotional feedback mechanisms. Background Technology

[0002] With the rapid development of digital media and interactive technologies, traditional static visual identity design can no longer meet the needs of brand image display in a multimedia environment. Existing visual identity design mainly suffers from the following problems:

[0003] First, traditional visual identity design mostly adopts a static form, which makes it difficult to adapt to different media environments, resulting in consistency issues in the brand's digital presentation. The application of static logos on different sizes, backgrounds, and media often requires manual adjustments, which is time-consuming, labor-intensive, and yields inconsistent results.

[0004] Secondly, existing dynamic visual identity designs primarily rely on the designer's subjective creation, lacking a systematic extraction and transformation of brand semantic characteristics, resulting in a disconnect between visual expression and brand connotation. This disconnect makes it difficult for the logo to accurately convey the brand's core values ​​and characteristics.

[0005] Furthermore, current visual identity design lacks a direct connection to user emotional feedback and cannot adaptively adjust to user emotional responses. This one-way communication model limits the emotional connection between the brand and users, reducing the personalization and interactivity of the brand experience.

[0006] Finally, existing technologies lack comprehensive solutions that organically integrate multiple technologies such as artificial intelligence, parametric design, sentiment analysis, and physics engines, making it difficult to achieve automated conversion and personalized presentation from brand semantics to dynamic visuals.

[0007] To address the aforementioned issues, this invention proposes an adaptive design system for dynamic visual identity based on AI parameterized generation. This system achieves a complete process from brand semantic analysis to dynamic visual generation through the collaborative work of multiple modules, and can adaptively adjust based on user emotional feedback, providing a brand-new technical path for brand visual identity design. Summary of the Invention

[0008] The purpose of this invention is to provide an adaptive design system for dynamic visual identity based on AI parameterized generation. This system can solve the problems existing in the prior art, realize the automatic conversion of brand semantics to dynamic visual identity, and make adaptive adjustments based on user emotional feedback, providing a personalized and consistent brand visual experience.

[0009] Another objective of this invention is to provide a system solution that can organically integrate artificial intelligence, parametric design, sentiment analysis, and physics engine technology, establishing a complete transformation chain from abstract brand concepts to concrete visual expressions.

[0010] Another objective of this invention is to construct a mapping relationship between user emotions and visual design parameters, realize adaptive adjustment of design based on emotional feedback, and enhance the emotional connection between brands and users.

[0011] This invention proposes a dynamic visual signage adaptive design system based on AI parameterized generation, comprising:

[0012] The data collection module is used to collect basic information from the brand's official website, including standard colors and logo patterns;

[0013] The data processing module is used to perform semantic analysis on the basic data and generate subject terms and parameterized control data;

[0014] A parameterization generation module, connected to the data processing module, is used to construct a parameterized model of the SVG layer based on the topic words;

[0015] The emotion feedback module is used to collect user emotion data and adjust parameters;

[0016] The dynamic output module is used to generate dynamic visual identifiers in conjunction with the physics engine.

[0017] Preferably, the data processing module includes:

[0018] The semantic analysis model, based on the Word2Vec and BERT natural language processing algorithms, uses brand phrases and brand standard colors as input to extract and express the semantic features of the brand.

[0019] The topic word generator model, based on the GAN architecture, consists of a discriminator and a generator. After training on the topic word corpus dataset, it extracts random numbers z as input to generate topic words related to brand semantics.

[0020] Preferably, the parameterization generation module includes:

[0021] An ellipse envelope generator that automatically generates ellipse envelopes using multiple random parameters;

[0022] An SVG generator matches the brand feature words obtained from the data processing module with the envelope generated by the ellipse envelope generator, and controls the size of the SVG dynamic layer.

[0023] The semantic mapping module includes a sentiment analysis model and a GAN adversarial network model. The sentiment analysis model analyzes user sentiment based on LSTM and attention mechanisms, and the GAN adversarial network model generates a random number z as input to generate ellipse parameters.

[0024] Preferably, the method for constructing the elliptical envelope parameterization model is as follows:

[0025] The envelope of a 2D ellipse is parameterized, where x0 and y0 are the coordinates of the center point of the ellipse, x and y are the lengths of the major and minor axes of the ellipse, and θ is the angle of the ellipse.

[0026] Fix x0 and y0 to random numbers, and set x = y and θ = y to generate multiple ellipses;

[0027] The keyword groups mentioned in user interaction comments are matched with feature words and marked as design style tendency, main words and style tendency;

[0028] The random number is matched with keywords related to the theme words generated by the theme word generator, and multiple color schemes are generated according to style preferences.

[0029] Preferably, the emotion feedback module includes:

[0030] EEG sensing devices are used to collect emotional feedback data from users when they view dynamic visual symbols in real time.

[0031] The emotion analysis unit uses LSTM and attention mechanisms to process the EEG data and generate user emotion feature vectors.

[0032] The parameter mapping unit maps the user emotion feature vector to design style tendency, main words, and style tendency, and generates parameter adjustment instructions.

[0033] The feedback training unit stores the emotional feedback data and adjustment results in a database for continuous optimization of the GAN adversarial network model.

[0034] Preferably, the dynamic output module is based on a physics engine, which includes:

[0035] 3D engine, used to generate dynamic visual identifiers representing three-dimensional models;

[0036] A 2D engine for generating dynamic visual identifiers in SVG format;

[0037] A physical material property simulator is used to simulate the mechanical and kinematic properties of signs and adjust their shape and boundaries.

[0038] The three-mode interactive controller is used to coordinate the synchronous changes of three output formats: video stream, 3D model, and dynamic SVG.

[0039] Preferably, a database system is also included, the database system comprising:

[0040] The first database is used to store basic information collected from the brand's official website;

[0041] The second database is used to store data after it has been evaluated by the data evaluation and storage module.

[0042] The second database includes a data information storage module, a data sentiment feedback module, and a data interaction module. The data interaction module is used to match the elliptical envelope with the topic words based on the random number z.

[0043] Preferably, the data evaluation and storage module performs the following data evaluations on the basic data:

[0044] Identify the range of R, G, B, C, M, Y, and K values ​​for brand identifiers;

[0045] Identify the features of the collected logo patterns, parse them for semantic analysis, and obtain design style tendencies, main words, and style tendency tags;

[0046] The logo pattern feature analysis includes extracting the main color of the logo, analyzing the color matching, and extracting the graphic. The extracted results are then extracted using a semantic analysis model, and keywords are added.

[0047] Preferably, the generator in the topic term generator model adopts the following structure:

[0048] GRU layer, used to capture word order features;

[0049] A multi-head attention mechanism layer is used to focus on different aspects of the semantics of the topic word;

[0050] LSTM layers are used for long-term dependency modeling.

[0051] The loss function of the GAN architecture is: LG = α * LA + β * Ldis, where LG represents the total loss, LA represents the loss of aesthetic features, Ldis represents the discrimination effect of the discriminator network, and α and β are the weight coefficients of different loss terms.

[0052] Preferably, the system operates by comprising the following steps:

[0053] Collect basic information from the brand's official website, evaluate it, and store it in the database;

[0054] Keywords are generated through semantic analysis, and a parameterized model is constructed simultaneously.

[0055] Interactive testing using a computer virtual engine;

[0056] Adjust parameters based on user emotional feedback to generate an elliptic parameterized envelope;

[0057] A three-modal interactive design for dynamic visual identity based on a physics engine, including video stream, 3D model and dynamic SVG;

[0058] The generated results and user emotional feedback data are stored in a database for continuous system learning and optimization.

[0059] The beneficial effects of this invention are as follows:

[0060] 1. It has achieved automated generation from brand semantics to dynamic visual identity, reducing design costs and improving design efficiency;

[0061] 2. A mapping relationship between user emotions and design parameters was established, enabling adaptive adjustment of visual identity and enhancing the emotional connection between the brand and users;

[0062] 3. Dynamic effects driven by a physics engine give visual logos natural and realistic changing characteristics, enhancing the visual experience;

[0063] 4. By adopting multimodal output technology, unified output of video stream, 3D model and dynamic SVG is achieved, ensuring the consistent presentation of brand logo in different media environments;

[0064] 5. A complete closed-loop system for data acquisition, processing, generation, feedback, and adjustment has been constructed, providing a technical framework for sustainable optimization of visual identity design. Attached Figure Description

[0065] Figure 1 This is the overall architecture diagram of the AI-parameterized dynamic visual signage adaptive design system of the present invention;

[0066] Figure 2 This is a structural diagram of the data processing module of the present invention;

[0067] Figure 3 This is a flowchart of the parameterization generation module of the present invention;

[0068] Figure 4 This is a diagram illustrating the structure and workflow of the emotional feedback module of this invention.

[0069] Figure 5 This is a parameter definition diagram of the elliptical envelope parameterization model of the present invention;

[0070] Figure 6 This is a diagram of the three-mode interactive structure of the dynamic output module of this invention;

[0071] Figure 7 This is a schematic diagram of the keyword generation process of this invention;

[0072] Figure 8 This is a schematic diagram of parameter generation based on GAN in this invention;

[0073] Figure 9 This is a flowchart of the system workflow of the present invention. Detailed Implementation

[0074] Please refer to the attached document. Figure 1-9 The present invention will now be described in further detail with reference to the accompanying drawings and embodiments.

[0075] like Figure 1 As shown, the AI-parameterized dynamic visual identifier adaptive design system provided by the present invention includes: a data collection module 1, a data processing module 2, a parametric generation module 3, an emotional feedback module 4, a dynamic output module 5, and a database system 6.

[0076] Data collection module 1 is used to collect basic information from the brand's official website, including standard colors and logo patterns. Specifically, data collection module 1 uses web crawling technology to crawl data from the target brand's official website, obtaining the brand's standard color information (including RGB and CMYK color values) and vector or bitmap files of the logo pattern. Preferably, data collection module 1 can also collect textual information such as brand description text, corporate culture, and brand history, providing richer input data for subsequent semantic analysis. In one embodiment of the present invention, data collection module 1 uses a Selenium-based web crawling framework combined with OpenCV for image processing, which can accurately extract the color value range of the brand's standard colors. For example, for the blue logo of a technology company, the standard color with RGB values ​​of (0, 122, 255) can be extracted.

[0077] Data processing module 2 is used to perform semantic analysis on the basic data, generating keyword terms and parameterized control data. For example... Figure 2 As shown, the data processing module 2 includes a semantic analysis model 21 and a topic word generator model 22.

[0078] Semantic analysis model 21, based on the natural language processing algorithms Word2Vec and BERT, uses brand phrases and brand standard colors as input to extract and express brand semantic features. Specifically, Word2Vec is mainly used to process shallow semantic associations and obtain semantic similarity between words, while BERT is used to capture contextual semantics and understand deeper brand connotations. The working process of semantic analysis model 21 can be represented as follows:

[0079] V semantic =f BERT (fWord2Vec (T brand C brand )),

[0080] Among them, V semantic f represents the brand semantic feature vector. BERT f represents the BERT model function. Word2Vec T represents the Word2Vec model function. brand Represents brand text information, C brand It indicates brand color information.

[0081] In a preferred embodiment of the present invention, the semantic analysis model 21 employs a pre-trained BERT-based model (containing a 12-layer Transformer encoder, each layer containing 12 attention heads), with a hidden layer dimension of 768 and a vocabulary of 30,000. The model can transform brand-related text into a 768-dimensional semantic vector, and integrate color information into this vector through a fully connected layer to form a unified brand semantic representation.

[0082] The topic word generator model 22 is based on a GAN architecture and consists of a discriminator and a generator. After training on a topic word corpus dataset, it extracts random numbers z as input to generate topic words related to brand semantics. For example... Figure 7 As shown, the keyword generation process first initializes with random noise z, and then combines the brand semantic feature vector V in the generator. semantic The generator generates candidate keywords. The discriminator then evaluates the differences between the generated keywords and the actual keyword data, providing feedback to optimize the generator. This process can be represented as:

[0083] T theme =G(z,V) semantic ),

[0084]

[0085] Among them, T theme Let L represent the generated keywords, G represent the generator function, z represent random noise, and D represent the discriminator function. GAN Let p represent the loss function of GAN. data p represents the true data distribution. z This indicates the noise distribution.

[0086] In one embodiment of the present invention, the generator of the topic word generator model 22 includes a GRU layer, a multi-head attention mechanism layer, and an LSTM layer. The GRU layer has 256 hidden units and is used to capture word order features; the multi-head attention mechanism layer has 8 attention heads with a dimension of 512 and is used to focus on different aspects of the semantics of the topic words; the LSTM layer has 512 hidden units and is used for long-term dependency modeling. This combination of three layers can effectively capture the sequence features of the text, focus on key semantic information, and maintain the semantic coherence of long sequences, thereby generating topic words that are more consistent with brand characteristics.

[0087] The parameterization generation module 3 is connected to the data processing module 2 and is used to construct a parameterized model of the SVG layer based on topic words. For example... Figure 3 As shown, the parameterization generation module 3 includes an ellipse envelope generator 31, an SVG generator 32, and a semantic mapping module 33.

[0088] Elliptical envelope generator 31 automatically generates elliptical envelopes using multiple random parameters. For example... Figure 5 As shown, the method for constructing the parametric model of the elliptical envelope is as follows: The 2D elliptical envelope is parametrically defined, where x0 and y0 are the coordinates of the ellipse's center point, x and y are the lengths of the ellipse's major and minor axes, and θ is the angle of the ellipse. The parametric equation of the ellipse can be expressed as:

[0089] X(t)=x0+xcosθcost-ysinθsint,

[0090] Y(t)=y0+xsinθcost+ycosθsint,

[0091] Where t is a parameter variable, and its value range is [0, 2π].

[0092] To generate diverse and regular envelope shapes, this invention fixes x0 and y0 as random numbers and sets x = y and θ = y, thereby simplifying the parameter space and making shape changes more controllable. In one embodiment of this invention, the values ​​of x0 and y0 are in the range of [-100, 100], while the values ​​of x (equal to y) are in the range of [20, 80]. This setting is based on experimental findings that the elliptical shapes within this range are the most aesthetically pleasing and practical.

[0093] The SVG generator 32 matches the brand feature words obtained from the data processing module 2 with the envelope generated by the ellipse envelope generator 31, and controls the size of the SVG dynamic layer. Specifically, the SVG generator 32 converts the ellipse envelope into an SVG path and sets the attributes of the SVG elements, such as fill color, stroke width, and transparency, according to the brand feature words. For example, when the brand feature words contain keywords such as "simple" and "technology," the SVG generator 32 tends to use a visual style with thin lines and high contrast; while when the feature words contain keywords such as "warm" and "traditional," it uses softer lines and transition effects.

[0094] The semantic mapping module 33 includes a sentiment analysis model 331 and a GAN adversarial network model 332. The sentiment analysis model 331 analyzes user sentiment based on LSTM and attention mechanisms; its specific implementation will be detailed in the sentiment feedback module 4 later. The GAN adversarial network model 332 generates a random number z as input to generate ellipse parameters. For example... Figure 8 As shown, this model receives user sentiment feature vectors and brand semantic feature vectors as conditional inputs, and generates ellipse parameters that conform to specific sentiments and brand characteristics through adversarial training. The loss function of the GAN adversarial network model 332 is:

[0095] L G =α·L A +β·L dis ,

[0096] Among them, L G L represents the total loss. A L represents the loss of aesthetic features. d is represents the discriminator network's discrimination performance, and α and β are the weight coefficients of different loss terms. In a preferred embodiment of the present invention, α is set to 0.7 and β is set to 0.3. This weight ratio was obtained through extensive experiments and can ensure the consistency of the generated results with the brand's semantics while maintaining the aesthetic quality of the generated results.

[0097] The emotion feedback module 4 is used to collect user emotion data and adjust parameters. For example... Figure 4 As shown, the emotion feedback module 4 includes an EEG sensing device 41, an emotion analysis unit 42, a parameter mapping unit 43, and a feedback training unit 44.

[0098] The brainwave sensing device 41 is used to collect emotional feedback data from users when viewing dynamic visual cues in real time. This invention employs a portable EEG (electroencephalography) device, typically containing 14-64 electrodes with a sampling rate of 128-1024 Hz. These devices can capture the user's brainwave activity, particularly frequency bands related to emotion, such as alpha waves (8-13 Hz, associated with relaxation), beta waves (13-30 Hz, associated with focus), and gamma waves (30-100 Hz, associated with higher cognitive processes).

[0099] The sentiment analysis unit 42 uses LSTM and an attention mechanism to process EEG data and generate user emotion feature vectors. LSTM networks are particularly well-suited for processing temporal signals, capturing the characteristics of EEG changes over time, while the attention mechanism focuses on the most critical time segments for emotion judgment. The mathematical expression of the sentiment analysis model is as follows:

[0100] h t =LSTM(x t ,h t-1 ),

[0101] α t =softmax(W a ·h t +b a ),

[0102]

[0103] Among them, h t This represents the hidden state of the LSTM at time step t, x. t α represents the input feature at time step t. t The attention weight W a and b a These are the weights and biases of the attention mechanism, V emotion This represents the final user sentiment feature vector.

[0104] In an embodiment of the present invention, the LSTM network adopts a bidirectional architecture with a hidden layer dimension of 256, and the attention mechanism adopts multi-head self-attention with 8 heads. The emotion analysis unit 42 can map EEG data into probability distributions of 8 basic emotions (joy, excitement, satisfaction, calmness, frustration, irritability, boredom, and confusion).

[0105] The parameter mapping unit 43 maps user emotion feature vectors to design style preferences, subject words, and style tendencies, generating parameter adjustment instructions. Specifically, this unit establishes a mapping table between emotion and design parameters, for example:

[0106] Joy / excitement → Increase color saturation, improve contrast, and speed up animation;

[0107] Satisfy / calm emotions → Maintain current parameters, slightly optimize smoothness;

[0108] Frustration / irritability → Reduce complexity, slow down animation speed, adjust color tone;

[0109] Boredom / confusion → Increase the frequency of change, introduce new elements, and adjust morphological parameters;

[0110] When the sentiment analysis unit 42 detects that the user's mood is biased towards boredom (probability exceeding 0.65), the parameter mapping unit 43 will generate adjustment instructions, such as increasing the ellipse deformation frequency by 30% or increasing the morphological complexity by 20%.

[0111] The feedback training unit 44 stores emotional feedback data and adjustment results in a database for continuous optimization of the GAN adversarial network model. This unit enables the system to learn independently, continuously optimizing parameter mapping rules and the generation model by accumulating the correspondence between user emotional feedback and design parameters. Preferably, this unit employs reinforcement learning, using positive user emotions as reward signals to guide the system in adjusting its parameter generation strategy.

[0112] The dynamic output module 5 is used to generate dynamic visual identifiers in conjunction with the physics engine. For example... Figure 6 As shown, this module is based on a physics engine and includes a 3D engine 51, a 2D engine 52, a physical material property simulator 53, and a three-mode interactive controller 54.

[0113] The 3D engine 51 is used to generate dynamic visual symbols representing three-dimensional models. This invention uses WebGL or Three.js as the 3D rendering engine, supporting material mapping, lighting effects, and camera animation, enabling the generation of symbol models with spatial depth and realism. Preferably, the physical parameter settings of the 3D engine 51 include: a gravity coefficient of 9.8 m / s². 2 The elastic coefficient ranges from 0.3 to 0.8, and the friction coefficient ranges from 0.1 to 0.5. These parameter values ​​are derived from appropriate adjustments to parameters in the real physical world, which can provide a better visual experience while ensuring physical realism.

[0114] The 2D engine 52 is used to generate dynamic visual logos in SVG format. This engine is primarily responsible for handling the generation and animation of vector graphics, ensuring that the logos maintain clarity and scalability at different sizes. In one embodiment of the invention, the 2D engine 52 uses D3.js in conjunction with the GSAP animation library to achieve smooth path deformation and attribute transition effects.

[0115] The physical material property simulator 53 is used to simulate the mechanical and kinematic properties of signage, adjusting the shape and boundaries of the signage. This simulator translates brand characteristics into physical property parameters; for example, robustness corresponds to a high elastic modulus (15000-20000MPa), and flexibility corresponds to a low elastic modulus (500-2000MPa). The physical simulation equation can be expressed as:

[0116] F = k·Δx,

[0117]

[0118] v t+1 =v t +a·Δt,

[0119] x t+1 =x t +v t+1 ·Δt,

[0120] Where F represents the applied force, k represents the elastic coefficient, Δx represents the deformation, a represents the acceleration, m represents the mass, v represents the velocity, x represents the position, and Δt represents the time step.

[0121] The tri-mode interactive controller 54 coordinates the synchronous changes of three output formats: video stream, 3D model, and dynamic SVG. This controller ensures consistency across different output formats, enabling the brand identity to maintain a unified visual experience across various media environments. Specifically, the controller employs an event-driven mechanism; when parameters in the physics engine change, the corresponding parameters for all output formats are updated synchronously, guaranteeing the synchronicity of animation effects.

[0122] Database system 6 includes a first database 61 and a second database 62. The first database 61 is used to store basic information collected from the brand's official website. The second database 62 is used to store data after data evaluation by the data evaluation and storage module 7. The second database 62 includes a data information storage module 621, a data sentiment feedback module 622, and a data interaction module 623.

[0123] The data information storage module 621 is used to store the evaluated brand basic information, including standard color R, G, B, C, M, Y, K values, logo pattern features, and extracted design style tendencies, main words, style tendencies, and other tagged information. In a preferred embodiment of the present invention, this module adopts a hybrid storage architecture combining a relational database (such as PostgreSQL) and a document-oriented database (such as MongoDB), which can efficiently process both structured color data and unstructured visual feature data simultaneously.

[0124] The data sentiment feedback module 622 is used to store user sentiment feedback data and corresponding design parameters, establishing a sentiment-parameter mapping knowledge base. This module uses a time-series database (such as InfluxDB) to store EEG data and sentiment analysis results, supporting efficient time-series query and analysis.

[0125] The data interaction module 623 is used to match the elliptical envelope with the keyword based on the random number z. This module implements a deterministic mapping between the random seed and the design elements, ensuring that the system can generate consistent visual output under the same input. Preferably, this module uses a hash mapping algorithm to map the random number z to a specific keyword index, achieving a stable and repeatable matching process.

[0126] The data evaluation and storage module 7 evaluates the basic data, including identifying the R, G, B, C, M, Y, and K value ranges of the brand identity and recognizing the features of the collected logo pattern. This module extracts the main color, analyzes color matching, and extracts graphics from the logo pattern. It also extracts and labels keyword groups using a semantic analysis model. Regarding color analysis, this invention uses the K-means clustering algorithm to determine the dominant color tone, through the following steps:

[0127] 1. Convert the image to the LAB color space (which is more in line with human visual perception than RGB);

[0128] 2. Apply the K-means algorithm, where K is usually set to 3-5 (based on experimental findings that most brand logos use 3-5 primary colors);

[0129] 3. Determine the primary color tone based on the number of pixels in each cluster and its distance from the center;

[0130] 4. Convert the cluster centers back to the RGB and CMYK color spaces;

[0131] More details about the topic generator model 22: This model's generator uses a combination of GRU, multi-head attention mechanism, and LSTM structure. The update formula for the GRU (Gated Recurrent Unit) layer is:

[0132] z t =σ(W z ·[h t-1 ,x t ]+b z ),

[0133] r t =σ(W r ·[h t-1 ,x t ]+b r ),

[0134]

[0135] Among them, z t Indicates the update gate, r t This indicates that the door is being reset. H represents the candidate hidden state. t σ represents the current hidden state, σ represents the sigmoid activation function, (W) represents the element-wise product, and W and b represent the weight matrix and bias vector, respectively.

[0136] The formula for calculating the multi-head attention mechanism is:

[0137]

[0138] MultiHead(Q,K,V)=Concat(head1,…,head h W O ,

[0139]

[0140] Where Q, K, and V represent the query, key, and value matrices, respectively, and d k W represents the dimension of the key. O , and This represents the learnable parameter matrix.

[0141] The formula for calculating LSTM (Long Short-Term Memory) layers is:

[0142] f t =σ(W f ·[h t-1 ,x t ]+b f ),

[0143] i t =σ(W i ·[h t-1 ,x t ]+b i ),

[0144]

[0145] o t =σ(W o ·[h t-1 ,x t ]+b o ),

[0146] h t =o t ⊙tanh(C t ),

[0147] Among them, f t Represents the forget gate, i t Indicates the input gate. Indicates the candidate cell state, C t Indicates the current cell state, o t Indicates the output gate, h t Indicates a hidden state.

[0148] The loss function of the GAN architecture is L G =α·L A +β·L dis Where α and β take values ​​of 0.7 and 0.3 respectively in this invention. Aesthetic feature loss L A A combination of content loss and style loss is used, while the discriminator loss L... dis The standard binary cross-entropy loss is used.

[0149] The working method of this system includes the following steps:

[0150] 1. Collect basic information from the brand's official website and evaluate and store it in the database: The data collection module 1 performs web crawling tasks to collect the brand's standard colors and logo patterns. The data evaluation and storage module 7 evaluates the collected data, including color value recognition and logo feature analysis, and stores the evaluation results in the database system 6.

[0151] 2. Generate keywords through semantic analysis and construct a parameterized model: Semantic analysis model 21 in data processing module 2 analyzes basic brand information and extracts semantic features; keyword generator model 22 generates relevant keywords based on these features; parameterized generation module 3 constructs an elliptical envelope parameterized model based on these keywords to prepare for the generation of dynamic visual identity.

[0152] 3. Interactive testing via computer virtual engine: The system generates initial dynamic visual logos, which are then tested using a virtual engine to evaluate their visual effects and dynamic performance.

[0153] 4. Adjust parameters based on user emotional feedback to generate elliptical parameterized envelope: The emotional feedback module 4 collects EEG data when the user views the logo, analyzes the emotional characteristics, and maps them into design parameter adjustment instructions to guide the parameter generation process of the elliptical envelope.

[0154] 5. Three-mode interactive design of dynamic visual identity based on physics engine: Dynamic output module 5 combines the simulation capabilities of physics engine to generate three output forms: video stream, 3D model and dynamic SVG, to achieve a unified visual experience in multimedia environment.

[0155] 6. Store the generated results and user emotional feedback data in the database for continuous system learning and optimization: The system stores the results and feedback of each interaction in the database system 6, continuously accumulating experience data, optimizing model parameters and mapping rules, and realizing the continuous evolution of the system.

[0156] Through the above steps, this invention realizes a complete process from brand semantic analysis to dynamic visual generation, and can adaptively adjust according to user emotional feedback, providing an innovative technical solution for brand visual identity design.

[0157] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

[0158] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An adaptive design system for dynamic visual signage based on AI parameterized generation, characterized in that, include: The data collection module is used to collect basic information from the brand's official website, including standard colors and logo patterns; The data processing module is used to perform semantic analysis on the basic data and generate subject terms and parameterized control data; The data processing module includes: a semantic analysis model, based on Word2Vec and BERT natural language processing algorithms, which takes brand phrases and brand standard colors as input to extract and express brand semantic features and generate brand semantic feature vectors; and a topic word generator model, based on a GAN architecture, consisting of a discriminator and a generator. After training on a topic word corpus dataset, the generator takes random noise z and the brand semantic feature vectors as input to generate topic words related to brand semantics. The discriminator evaluates the difference between the generated words and the real topic words and provides feedback to optimize the generator. A parameterized generation module, connected to the data processing module, is used to construct a parameterized model controlling the SVG layer based on the topic words. The parameterized generation module includes: an ellipse envelope generator, which automatically generates an ellipse envelope using multiple random parameters; an SVG generator, which matches the brand feature words obtained from the data processing module with the envelope generated by the ellipse envelope generator, converts the ellipse envelope into an SVG path, and sets the attributes of the SVG elements according to the brand feature words to control the size of the SVG dynamic layer; and a semantic mapping module, including a sentiment analysis model and a GAN adversarial network model. The sentiment analysis model analyzes user sentiment based on LSTM and attention mechanisms, and the GAN adversarial network model uses a random number z, a user sentiment feature vector, and a brand semantic feature vector as conditional inputs, and generates ellipse parameters that conform to specific sentiment and brand characteristics through adversarial training. An emotion feedback module is used to collect user emotion data and adjust the ellipse parameters and SVG element attribute parameters in the parameterization generation module. The emotion feedback module includes: an EEG sensing device for real-time collection of user emotion feedback data when viewing dynamic visual icons; an emotion analysis unit that processes the EEG data using LSTM and attention mechanisms to generate user emotion feature vectors; a parameter mapping unit that maps the user emotion feature vectors to design style tendencies, main words, and style tendencies to generate parameter adjustment instructions; and a feedback training unit that stores the emotion feedback data and adjustment results in a database for continuous optimization of the GAN adversarial network model. The dynamic output module is used to generate dynamic visual identifiers in conjunction with the physics engine.

2. The system according to claim 1, characterized in that, The generator in the keyword generator model adopts the following structure: GRU layer, used to capture word order features; A multi-head attention mechanism layer is used to focus on different aspects of the semantics of the topic word; LSTM layers are used for long-term dependency modeling. The loss function of the GAN architecture is as follows: ,in Indicates the total loss. This indicates a loss of aesthetic features. This indicates the discrimination effect of the discriminator network. , These are the weighting coefficients for different loss terms.

3. The system according to claim 1, characterized in that, The method for constructing the elliptical envelope parameterization model in the elliptical envelope generator is as follows: Parameterize the 2D elliptical envelope. , Here are the coordinates of the center point of the ellipse. , The length of the major and minor axes of the ellipse. Let be the rotation angle of the ellipse; Will , Fixed as random numbers, and set , To simplify the parameter space and make morphological changes more controllable, multiple envelope morphologies are generated. The keyword groups mentioned in user interaction comments are matched with feature words and marked as design style tendency, main words and style tendency; The random number is matched with keywords related to the theme words generated by the theme word generator, and multiple color schemes are generated according to style preferences.

4. The system according to claim 1, characterized in that, The dynamic output module is based on a physics engine, which includes: 3D engine, used to generate dynamic visual identifiers representing three-dimensional models; A 2D engine for generating dynamic visual identifiers in SVG format; A physical material property simulator is used to simulate the mechanical and kinematic properties of signs and adjust their shape and boundaries. The three-mode interactive controller is used to coordinate the synchronous changes of three output formats: video stream, 3D model, and dynamic SVG.

5. The system according to claim 1, characterized in that, It also includes a database system, which comprises: The first database is used to store basic information collected from the brand's official website; The second database is used to store data after it has been evaluated by the data evaluation and storage module. The second database includes a data information storage module, a data sentiment feedback module, and a data interaction module. The data interaction module is used to match the elliptical envelope with the topic words based on the random number z.

6. The system according to claim 5, characterized in that, The data evaluation and storage module performs the following data evaluations on the basic data: Identify the range of R, G, B, C, M, Y, and K values ​​for brand identifiers; Identify the features of the collected logo patterns, parse them for semantic analysis, and obtain design style tendencies, main words, and style tendency tags; The logo pattern feature analysis includes extracting the main color of the logo, analyzing the color matching, and extracting the graphic. The extracted results are then extracted using a semantic analysis model, and keywords are added.

7. The system according to claim 2, characterized in that, In the loss function of the GAN architecture, The value is 0.

7. The value is 0.

3.

8. The system according to claim 1, characterized in that, The system operates by including the following steps: Collect basic information from the brand's official website, evaluate it, and store it in the database; Keywords are generated through semantic analysis, and a parameterized model is constructed simultaneously. Interactive testing using a computer virtual engine; The emotional feedback module collects emotional feedback data when users view dynamic visual icons through the EEG sensing device. The emotional analysis unit generates a user emotional feature vector. The parameter mapping unit maps the user emotional feature vector to design style tendency, main words, and style tendency, and generates parameter adjustment instructions to adjust the ellipse parameters and generate an elliptic parameterized envelope. A three-modal interactive design for dynamic visual identity based on a physics engine, including video stream, 3D model and dynamic SVG; The generated results and user emotional feedback data are stored in a database for continuous system learning and optimization.

Images