Convolutional neural network driven online resource automatic design generation method

Abstract

The invention discloses an online resource automatic design generation method driven by a convolutional neural network, and belongs to the technical field of man-machine interaction and cooperative operation, and the method comprises the following steps: S1, data collection: collecting multiple types of design data as a training basis; s2, data preprocessing: carrying out cleaning and structured processing on the original data; s3, constructing a convolutional neural network model; s4, designing a generation process: realizing automatic design based on model output; s5, feedback improvement; s6, performing model evaluation and multi-dimensional verification; s7, performing multi-modal input fusion and semantic alignment; S8, performing dynamic resource optimization and hardware adaptation; s9, performing compliance examination and ethical constraint embedding; and S10, performing collaborative creation and version evolution management. Through technical innovation and engineering closed-loop design, the core problems of data splitting, equipment dependence, compliance risk and the like in the field of automatic design are solved, and an efficient, safe and sustainable intelligent solution is provided for digital content creation.

Description

Technical Field

[0001] The present invention relates to the field of human-computer interaction and collaborative computing technology, and in particular to a method for automatically designing and generating online resources driven by a convolutional neural network. Background Art

[0002] With the rapid development of internet technology, the demand for online resources has exploded. From online courseware and teaching materials in education to advertising pages and product display materials in the commercial sector, and even digital content creation in the entertainment industry, there is an urgent need for efficient, accurate, and personalized design generation methods. However, traditional design processes rely on manual operations by professional designers, which are inefficient, costly, and difficult to meet large-scale personalized needs. Therefore, we propose a convolutional neural network-driven method for automatic online resource design generation to address this problem. Summary of the Invention

[0003] The purpose of the present invention is to provide a convolutional neural network driven online resource automatic design and generation method to solve the problems raised in the above background technology.

[0004] In order to achieve the above object, the present invention adopts the following technical solutions:

[0005] The convolutional neural network-driven automatic design and generation method for online resources includes the following steps:

[0006] S1. Data collection: Collect various types of design data as the basis for training;

[0007] S2. Data preprocessing: cleaning and structuring the raw data;

[0008] S3. Convolutional neural network model construction: Design a deep model architecture for generation tasks;

[0009] S4, Design Generation Process: Automated design implementation based on model output;

[0010] S5. Feedback improvement: Collect user ratings and modification records of generated designs, inject user preference data into the model, and fine-tune the generation strategy;

[0011] S6. Model Evaluation and Multi-Dimensional Verification: Establish an automated evaluation system, including aesthetic scoring, user experience testing, cross-platform compatibility testing, and generation efficiency indicators, to verify whether the design output meets functional, artistic, and performance requirements;

[0012] S7, Multimodal Input Fusion and Semantic Alignment: Introducing a multimodal input parsing module for text, speech, sketches, etc., aligning user intent with design elements through a cross-modal attention mechanism to address the issue of insufficient information from a single data source;

[0013] S8. Dynamic Resource Optimization and Hardware Adaptation: Design lightweight model branches and dynamic computational graph compression strategies, automatically switch model scales based on the computing power of terminal devices, and achieve a balance between low-latency response and resource consumption;

[0014] S9. Compliance review and ethical constraint embedding: Integrate copyright detection, culturally sensitive content filtering, and an accessible design rule library to ensure that generated content complies with legal regulations and social ethics.

[0015] S10. Collaborative creation and version evolution management: Supports multi-user real-time editing conflict resolution, historical version difference visualization, and style inheritance functions to meet the iterative needs of team collaboration.

[0016] Preferably, in said S1, the specific steps are as follows:

[0017] S101. Design material library collection: Collect design templates from multiple industries, covering formats such as PSD, Sketch, and Figma, and mark the design element types and layout parameters;

[0018] S102. User behavior data collection: Record user operation logs in mainstream design tools, including element dragging trajectory, adjustment frequency, and undo / redo behavior, to build a behavior pattern database;

[0019] S103, multimodal input integration: collect text descriptions, voice commands, hand-drawn sketches and reference images to establish a multimodal input sample library;

[0020] S104. Construction of domain knowledge base: Organize industry design specifications and form a structured rule base.

[0021] Preferably, in said S2, the specific steps are as follows:

[0022] S201, data standardization processing: unify the image resolution to 1920×1080, convert the color space to sRGB, and standardize the text encoding to UTF-8 format;

[0023] S202, feature labeling and mapping: Use semi-automatic tools to label design elements and establish a mapping relationship table between element categories and layout parameters;

[0024] S203, data enhancement and noise filtering: rotate, crop, and color-perturb the image, while filtering low-quality samples such as blurred images and low-contrast layouts.

[0025] Preferably, in S3, the specific steps are as follows:

[0026] S301, encoder-decoder architecture design: The encoder uses multi-scale CNN to extract local image features, ViT to extract global semantics, and the decoder combines Transformer to generate layout parameters and generates visual styles through GAN;

[0027] S302, multi-task joint training: simultaneous optimization of layout generation, aesthetic scoring, and responsive adaptation;

[0028] S303, cross-modal attention fusion: construct a text-image attention matrix to align keywords and image regions.

[0029] Preferably, in said S4, the specific steps are as follows:

[0030] S401, element positioning and style assignment: Generate a grid layout based on the model output coordinates, and match fonts, colors, and shadow effects from a preset style library;

[0031] S402, Dynamic Responsive Adaptation: Dynamically adjust the stacking order and proportion of elements based on the device screen size;

[0032] S403, multiple solution candidate output: Generate 3-5 style differentiation solutions for user selection or further optimization;

[0033] Preferably, in S5, the specific steps are as follows:

[0034] S501, user behavior log analysis: extracting user modification behavior as preference signals;

[0035] S502, incremental learning and model fine-tuning: inject user preference data into the training set and use online learning strategies to update model weights;

[0036] S503, anomaly detection and self-repair: Identify layout errors such as element overlap and overflow boundaries, trigger regeneration or recommend correction suggestions;

[0037] Preferably, in said S6, the specific steps are as follows:

[0038] S601, Automated Aesthetic Scoring: Uses a pre-trained aesthetic model to evaluate the color harmony and visual balance of the layout, outputting a 0-1 score.

[0039] S602, User Experience Simulation Test: Analyze the user's visual focus distribution through eye tracking simulation tools to optimize the reading flow design;

[0040] S603, cross-platform compatibility test: Render the layout on multiple terminals to verify the element's adaptive effect;

[0041] Preferably, in said S7, the specific steps are as follows:

[0042] S701, Multimodal Feature Extraction: BERT extracts keywords and sentiment, ASR converts text into voiceprint analysis to identify user intent, and edge detection algorithms extract contour features.

[0043] S702, cross-modal attention alignment: Calculate the joint embedding vector of text, sketch, and speech to generate a unified semantic space;

[0044] S703, Intent-Design Element Mapping: Establish association rules between user intent and design elements.

[0045] Preferably, in said S8, the specific steps are as follows:

[0046] S801, Lightweight Model Branch Design: Build a pruned CNN lightweight branch for mobile devices, retaining the core feature extraction capabilities;

[0047] S802, dynamic computation graph compression: automatically switches model complexity based on the device GPU computing power;

[0048] S803, Resource Consumption Monitoring: Monitor memory usage and inference latency in real time, and dynamically adjust batch processing size to balance efficiency and performance.

[0049] The beneficial effects of the present invention are:

[0050] 1. The convolutional neural network-driven automatic online resource design generation method described in this invention achieves high-precision design intent understanding and element matching through multimodal data integration and deep feature alignment technology. The system can simultaneously parse text descriptions, hand-drawn sketches, and user behavior logs, and utilize a cross-modal attention mechanism to accurately associate semantic keywords with visual elements to generate layout solutions that meet user needs. This method solves the problem of intention bias caused by traditional tools relying on a single input, significantly improving the accuracy of design solutions and user satisfaction. It is particularly suitable for scenarios such as advertising and e-commerce that require rapid response to needs;

[0051] 2. The convolutional neural network-driven online resource automatic design and generation method described in the present invention achieves low-latency, high-concurrency design and generation services through dynamic resource optimization and hardware adaptation strategies. The system automatically switches between lightweight models and complete models based on the computing power of the terminal device, prioritizing response speed on mobile devices and focusing on generation quality in the cloud. At the same time, it reduces memory usage through computational graph compression technology. This capability enables it to be widely used in scenarios such as real-time collaboration and cross-platform content creation, breaking the reliance of traditional generation tools on high-performance devices.

[0052] 3. In the present invention, the convolutional neural network-driven automatic online resource design and generation method ensures the legality and social responsibility of the design output through compliance review and ethical constraint embedding mechanisms. The system has a built-in copyright database and sensitive content recognition model, automatically blocking unauthorized fonts, images, and political violence elements. At the same time, it enforces accessibility design standards to ensure readability for color-blind users. This mechanism not only reduces legal risks but also enhances the company's brand image in the dimension of social responsibility.

[0053] 4. The convolutional neural network-driven online resource automatic design generation method described in the present invention achieves efficient team collaboration and controllable design iteration through collaborative creation and version evolution management functions. It supports real-time editing by multiple people based on a conflict-free synchronization algorithm, automatically merges operation instructions and records historical version differences, and supports extracting style features from old versions for application to new designs. This function significantly shortens the collaboration cycle of enterprise-level projects and avoids rework caused by version confusion.

[0054] 5. In the present invention, the convolutional neural network-driven automatic design and generation method for online resources achieves continuous self-optimization of design strategies through a feedback loop and incremental learning mechanism. The system analyzes user modification behavior and A / B test data in real time, identifies the correlation between high-frequency adjustment patterns and business indicators, and dynamically injects them into the model training process. This process enables the generated solutions to keep up with the evolution of user preferences and market trends, and maintain the advanced nature and practicality of the design output in the long term. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 This is a flowchart of the convolutional neural network-driven online resource automatic design and generation method proposed in the present invention. DETAILED DESCRIPTION

[0056] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0057] Reference Figure 1 ,The convolutional neural network driven online resource automatic design and generation method includes the following steps:

[0058] S1. Data collection: Collect various types of design data as the basis for training;

[0059] S2. Data preprocessing: cleaning and structuring the raw data;

[0060] S3. Convolutional neural network model construction: Design a deep model architecture for generation tasks;

[0061] S4, Design Generation Process: Automated design implementation based on model output;

[0062] S5. Feedback improvement: Collect user ratings and modification records of generated designs, inject user preference data into the model, and fine-tune the generation strategy;

[0063] S6. Model Evaluation and Multi-Dimensional Verification: Establish an automated evaluation system, including aesthetic scoring, user experience testing, cross-platform compatibility testing, and generation efficiency indicators, to verify whether the design output meets functional, artistic, and performance requirements;

[0064] S7, Multimodal Input Fusion and Semantic Alignment: Introducing a multimodal input parsing module for text, speech, sketches, etc., aligning user intent with design elements through a cross-modal attention mechanism to address the issue of insufficient information from a single data source;

[0065] S8. Dynamic Resource Optimization and Hardware Adaptation: Design lightweight model branches and dynamic computational graph compression strategies, automatically switch model scales based on the computing power of terminal devices, and achieve a balance between low-latency response and resource consumption;

[0066] S9. Compliance review and ethical constraint embedding: Integrate copyright detection, culturally sensitive content filtering, and an accessible design rule library to ensure that generated content complies with legal regulations and social ethics.

[0067] S10. Collaborative Creation and Version Evolution Management: Supports multi-user real-time editing conflict resolution, historical version difference visualization, and style inheritance to meet the iterative needs of team collaboration;

[0068] In this embodiment, in S1, the specific steps are as follows:

[0069] S101. Design material library collection: Collect design templates from multiple industries, covering formats such as PSD, Sketch, and Figma, and mark the design element types and layout parameters;

[0070] S102. User behavior data collection: Record user operation logs in mainstream design tools, including element dragging trajectory, adjustment frequency, and undo / redo behavior, to build a behavior pattern database;

[0071] S103, multimodal input integration: collect text descriptions, voice commands, hand-drawn sketches and reference images to establish a multimodal input sample library;

[0072] S104. Construction of domain knowledge base: Organize industry design specifications and form a structured rule base.

[0073] In this embodiment, in S2, the specific steps are as follows:

[0074] S201, data standardization processing: unify the image resolution to 1920×1080, convert the color space to sRGB, and standardize the text encoding to UTF-8 format;

[0075] S202, feature labeling and mapping: Use semi-automatic tools to label design elements and establish a mapping relationship table between element categories and layout parameters;

[0076] S203, data enhancement and noise filtering: rotate, crop, and color-perturb the image, while filtering low-quality samples such as blurred images and low-contrast layouts.

[0077] In this embodiment, in S3, the specific steps are as follows:

[0078] S301, encoder-decoder architecture design: The encoder uses multi-scale CNN to extract local image features, ViT to extract global semantics, and the decoder combines Transformer to generate layout parameters and generates visual styles through GAN;

[0079] S302, multi-task joint training: simultaneous optimization of layout generation, aesthetic scoring, and responsive adaptation;

[0080] S303, cross-modal attention fusion: construct a text-image attention matrix to align keywords and image regions.

[0081] In this embodiment, in S4, the specific steps are as follows:

[0082] S401, element positioning and style assignment: Generate a grid layout based on the model output coordinates, and match fonts, colors, and shadow effects from a preset style library;

[0083] S402, Dynamic Responsive Adaptation: Dynamically adjust the stacking order and proportion of elements based on the device screen size;

[0084] S403, multiple solution candidate output: Generate 3-5 style differentiation solutions for user selection or further optimization;

[0085] In this embodiment, in S5, the specific steps are as follows:

[0086] S501, user behavior log analysis: extracting user modification behavior as preference signals;

[0087] S502, incremental learning and model fine-tuning: inject user preference data into the training set and use online learning strategies to update model weights;

[0088] S503, anomaly detection and self-repair: Identify layout errors such as element overlap and overflow boundaries, trigger regeneration or recommend correction suggestions;

[0089] In this embodiment, in S6, the specific steps are as follows:

[0090] S601, Automated Aesthetic Scoring: Uses a pre-trained aesthetic model to evaluate the color harmony and visual balance of the layout, outputting a 0-1 score.

[0091] S602, User Experience Simulation Test: Analyze the user's visual focus distribution through eye tracking simulation tools to optimize the reading flow design;

[0092] S603, cross-platform compatibility test: Render the layout on multiple terminals to verify the element's adaptive effect;

[0093] In this embodiment, in S7, the specific steps are as follows:

[0094] S701, Multimodal Feature Extraction: BERT extracts keywords and sentiment, ASR converts text into voiceprint analysis to identify user intent, and edge detection algorithms extract contour features.

[0095] S702, cross-modal attention alignment: Calculate the joint embedding vector of text, sketch, and speech to generate a unified semantic space;

[0096] S703, Intent-Design Element Mapping: Establish association rules between user intent and design elements.

[0097] In this embodiment, in S8, the specific steps are as follows:

[0098] S801, Lightweight Model Branch Design: Build a pruned CNN lightweight branch for mobile devices, retaining the core feature extraction capabilities;

[0099] S802, dynamic computation graph compression: automatically switches model complexity based on the device GPU computing power;

[0100] S803, Resource Consumption Monitoring: Monitor memory usage and inference latency in real time, and dynamically adjust batch processing size to balance efficiency and performance.

[0101] Through technological innovation and engineering closed-loop design, this invention solves core problems in the field of automated design, such as data fragmentation, equipment dependence, and compliance risks, and provides an efficient, secure, and sustainable intelligent solution for digital content creation.

[0102] The above is a detailed introduction to the convolutional neural network-driven automatic design and generation method for online resources provided by the present invention. Specific embodiments are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. It should be pointed out that for ordinary technicians in this technical field, without departing from the principles of the present invention, the present invention can also be improved and modified in several ways, and these improvements and modifications also fall within the scope of protection of the claims of the present invention.

Claims

1. A convolutional neural network driven method for automatically designing and generating online resources, characterized in that: The steps include: S1. Data collection: Collect various types of design data as the basis for training; S2. Data preprocessing: cleaning and structuring the raw data; S3. Convolutional neural network model construction: Design a deep model architecture for generation tasks; S4, Design Generation Process: Automated design implementation based on model output; S5, feedback improvement; Collect user ratings and modification records of generated designs, inject user preference data into the model, and fine-tune the generation strategy; S6. Model Evaluation and Multi-Dimensional Verification: Establish an automated evaluation system, including aesthetic scoring, user experience testing, cross-platform compatibility testing, and generation efficiency indicators, to verify whether the design output meets functional, artistic, and performance requirements; S7. Multimodal Input Fusion and Semantic Alignment: This introduces a multimodal input parsing module that integrates text, speech, and sketches. This uses a cross-modal attention mechanism to align user intent with design elements, addressing the issue of insufficient information from a single data source. S8. Dynamic Resource Optimization and Hardware Adaptation: Design lightweight model branches and dynamic computational graph compression strategies, automatically switch model scales based on the computing power of terminal devices, and achieve a balance between low-latency response and resource consumption; S9. Compliance review and ethical constraint embedding: Integrate copyright detection, culturally sensitive content filtering, and an accessible design rule library to ensure that generated content complies with legal regulations and social ethics. S10. Collaborative creation and version evolution management: Supports multi-user real-time editing conflict resolution, historical version difference visualization, and style inheritance functions to meet the iterative needs of team collaboration.

2. The method for automatically designing and generating online resources driven by a convolutional neural network according to claim 1, characterized in that: In S1, the specific steps are as follows: S101. Design material library collection: Collect design templates from multiple industries, covering formats such as PSD, Sketch, and Figma, and mark the design element types and layout parameters; S102. User behavior data collection: Record user operation logs in mainstream design tools, including element dragging trajectory, adjustment frequency, and undo / redo behavior, to build a behavior pattern database; S103, multimodal input integration: collect text descriptions, voice commands, hand-drawn sketches and reference images to establish a multimodal input sample library; S104. Construction of domain knowledge base: Organize industry design specifications and form a structured rule base.

3. The method for automatically designing and generating online resources driven by a convolutional neural network according to claim 1, characterized in that: In said S2, the specific steps are as follows: S201, data standardization processing: unify the image resolution to 1920×1080, convert the color space to sRGB, and standardize the text encoding to UTF-8 format; S202, feature labeling and mapping: Use semi-automatic tools to label design elements and establish a mapping relationship table between element categories and layout parameters; S203, data enhancement and noise filtering: rotate, crop, and color-perturb the image, while filtering low-quality samples such as blurred images and low-contrast layouts.

4. The method for automatically designing and generating online resources driven by a convolutional neural network according to claim 1, characterized in that: In S3, the specific steps are as follows: S301, encoder-decoder architecture design: The encoder uses multi-scale CNN to extract local image features, ViT to extract global semantics, and the decoder combines Transformer to generate layout parameters and generates visual styles through GAN; S302, multi-task joint training: simultaneous optimization of layout generation, aesthetic scoring, and responsive adaptation; S303, cross-modal attention fusion: construct a text-image attention matrix to align keywords and image regions.

5. The method for automatically designing and generating online resources driven by a convolutional neural network according to claim 1, characterized in that: In said S4, the specific steps are as follows: S401, element positioning and style assignment: Generate a grid layout based on the model output coordinates, and match fonts, colors, and shadow effects from a preset style library; S402, Dynamic Responsive Adaptation: Dynamically adjust the stacking order and proportion of elements based on the device screen size; S403. Multiple-solution candidate output: Generate 3-5 style-differentiated solutions for user selection or further optimization.

6. The method for automatically designing and generating online resources driven by a convolutional neural network according to claim 1, characterized in that: In said S5, the specific steps are as follows: S501, user behavior log analysis: extracting user modification behavior as preference signals; S502, incremental learning and model fine-tuning: inject user preference data into the training set and use online learning strategies to update model weights; S503, anomaly detection and self-repair: Identify incorrect layouts with overlapping elements or overflowing boundaries, and trigger regeneration or recommend correction suggestions.

7. The method for automatically designing and generating online resources driven by a convolutional neural network according to claim 1, characterized in that: In said S6, the specific steps are as follows: S601, Automated Aesthetic Scoring: Uses a pre-trained aesthetic model to evaluate the color harmony and visual balance of the layout, outputting a 0-1 score. S602, User Experience Simulation Test: Analyze the user's visual focus distribution through eye tracking simulation tools to optimize the reading flow design; S603. Cross-platform compatibility test: Render the layout on multiple terminals to verify the element's adaptive effect.

8. The method for automatically designing and generating online resources driven by a convolutional neural network according to claim 1, characterized in that: In said S7, the specific steps are as follows: S701, Multimodal Feature Extraction: BERT extracts keywords and sentiment, ASR converts text into voiceprint analysis to identify user intent, and edge detection algorithms extract contour features. S702, cross-modal attention alignment: Calculate the joint embedding vector of text, sketch, and speech to generate a unified semantic space; S703, Intent-Design Element Mapping: Establish association rules between user intent and design elements.

9. The method for automatically designing and generating online resources driven by a convolutional neural network according to claim 1, characterized in that: In said S8, the specific steps are as follows: S801, Lightweight Model Branch Design: Build a pruned CNN lightweight branch for mobile devices, retaining the core feature extraction capabilities; S802, dynamic computation graph compression: automatically switches model complexity based on the device GPU computing power; S803, Resource Consumption Monitoring: Monitor memory usage and inference latency in real time, and dynamically adjust batch processing size to balance efficiency and performance.

Images