A digital media design element optimization method and system based on AIGC technology

Abstract

The present invention relates to the field of artificial intelligence generation technology, and in particular to a digital media design element optimization method and system using AIGC technology. The method comprises the following steps: collecting digital media multi-source data; performing digital media design element analysis based on the digital media multi-source data to generate digital media design element data; designing a digital media design element association matrix based on the digital media design element data; analyzing personalized digital media design preference feature data; establishing a user-preferred digital media design element optimization generation model based on a preset adversarial generation network model and personalized digital media design preference feature data, and using the digital media design element association matrix for training processing to generate a personalized digital media design element optimization generation model; performing a user digital media design element intelligent optimization operation on the digital media design element to-be-optimized data using the personalized digital media design element optimization generation model. The present invention realizes the intelligent optimization generation of high-quality digital media design elements.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence generation technology, and in particular to a digital media design element optimization method and system using AIGC technology. Background Art

[0002] AIGC technology (AI Generated Content) is the use of artificial intelligence (AI) technology to automatically generate various types of digital information content. With the development of deep learning and large-scale models (such as GPT-3, DALL-E, Stable Diffusion, etc.), AIGC has been widely used in various industries and application scenarios. With the rapid development of digital media technology, the demand for digital content creation is growing. As a key component of digital content creation, digital media design elements play an important role in film and television, advertising, games, animation and other fields. Through the automation of digital media design with AIGC technology, the design efficiency is significantly improved, the design cost is reduced, and more creative information is provided for designers and ordinary users. However, the existing AIGC technology has an insufficient understanding of different design elements in the optimization and generation of digital media design elements, which makes it difficult to meet complex and diverse design needs. In addition, it lacks in-depth correlation analysis of digital media design elements and random digital media design element generation, resulting in poor optimization and generation of digital media design elements, which cannot meet the quality requirements of digital media. Summary of the invention

[0003] Based on this, the present invention provides a digital media design element optimization method and system based on AIGC technology to solve at least one of the above technical problems.

[0004] To achieve the above purpose, a digital media design element optimization method using AIGC technology includes the following steps:

[0005] Step S1: using a preset digital media API calling interface to collect and process digital media multi-source data to generate digital media multi-source data;

[0006] Step S2: performing regional fitting digital media visual analysis based on digital media multi-source data to generate regional fitting digital media visual data; performing digital media design element analysis based on the regional fitting digital media visual data to generate digital media design element data;

[0007] Step S3: Design an association matrix of each type of digital media design element based on the digital media design element data to generate a digital media design element association matrix;

[0008] Step S4: obtaining user digital media design data; performing personalized digital media design preference feature analysis based on the user digital media design data to generate personalized digital media design preference feature data;

[0009] Step S5: Establishing a digital media design element optimization generation model of user preference based on a preset generative adversarial network model and personalized digital media design preference feature data to generate an initial digital media design element optimization generation model; transmitting the digital media design element association matrix to the initial digital media design element optimization generation model to perform personalized optimization generation training processing on the digital media design elements to generate a personalized digital media design element optimization generation model;

[0010] Step S6: Obtain real-time user digital media design element data to be optimized; transmit the real-time user digital media design element data to be optimized to the personalized digital media design element optimization generation model for real-time user digital media design element optimization generation processing to generate real-time user digital media design element optimization data; transmit the real-time user digital media design element optimization data to the terminal to execute the user digital media design element intelligent optimization operation.

[0011] The present invention collects relevant data from multiple data sources through a preset digital media API interface and performs standardized processing on the data. The collection of such multi-source data can cover design elements of different dimensions, so that the subsequent optimization and generation process is based on more comprehensive and rich data, and the diversity and generation effect of digital media design elements are improved. Regional fitting visual analysis is performed on multi-source data, and the visual features of different regions in digital media are extracted and understood to generate regional fitting digital media visual data, which can find the distribution and correlation of design elements in different visual regions, and provide basic data for optimizing design elements. Design element analysis based on regional fitting data can extract the visual features of different design elements and reveal the overall visual effect of design elements. By establishing an association matrix of digital media design elements, the association features between different design elements are deeply excavated and characterized. The association matrix reflects the intrinsic connection of design elements in terms of vision, style, color, etc., and provides an organized structured information for subsequent optimization generation. The mining of such association features helps to maintain the coordination and consistency between design elements during the generation process, thereby improving the overall quality and visual effect of the optimization generation results, so that the generated design is more in line with actual application requirements. Obtain the user's digital media design data, analyze the user's personalized design preferences, better understand the user's style and habits in design, use algorithms such as convolutional neural networks to extract the style in the user's design data, and analyze the corresponding interaction features, so as to achieve quantitative analysis of the user's personalized preferences, and provide user-specific reference standards for the subsequent optimization and generation of design elements, so that the optimized and generated design elements can better meet the user's personalized needs, improve user satisfaction and the pertinence of digital media design. Using the adversarial generative network (GAN) and user preference feature data, establish a user-preferred digital media design element optimization generation model, so that the model can obtain personalized design element optimization generation data according to the user's preferences, and make full use of the user's preference characteristics to enable the generation model to have the ability to learn and generate the user's favorite design style. Transmit the digital media design element association matrix to the generation model for personalized optimization generation training, further strengthen the model's understanding of the association characteristics of different design elements and analyze the needs of optimization generation of various types of design elements, so as to better coordinate the relationship between different design elements during optimization generation. By acquiring real-time user digital media design element data to be optimized, and performing real-time user digital media design element optimization generation processing on the real-time user digital media design element data to be optimized based on the personalized digital media design element optimization generation model, the real-time optimization capability of digital media design elements is realized, and when the user provides the design element data to be optimized, the optimized design elements can be quickly generated and transmitted to the terminal. This meets the user's needs for instant feedback and adjustment during the design process, and greatly improves the design efficiency.

[0012] This specification provides a digital media design element optimization system of AIGC technology, which is used to execute the digital media design element optimization method of AIGC technology as described above. The digital media design element optimization system of AIGC technology includes:

[0013] The digital media multi-source data acquisition module is used to collect and process digital media multi-source data using a preset digital media API call interface to generate digital media multi-source data;

[0014] The digital media design element analysis module is used to perform regional fitting digital media visual analysis based on digital media multi-source data to generate regional fitting digital media visual data; perform digital media design element analysis based on regional fitting digital media visual data to generate digital media design element data;

[0015] A digital media design element association analysis module is used to design an association matrix of each digital media design element type based on the digital media design element data, and generate a digital media design element association matrix;

[0016] The personalized digital media design preference feature analysis module is used to obtain user digital media design data; perform personalized digital media design preference feature analysis based on the user digital media design data to generate personalized digital media design preference feature data;

[0017] A design element optimization generation model construction module is used to establish a digital media design element optimization generation model of user preference based on a preset adversarial generative network model and personalized digital media design preference feature data, and generate an initial digital media design element optimization generation model; transmit the digital media design element association matrix to the initial digital media design element optimization generation model to perform personalized optimization generation training processing on the digital media design elements, and generate a personalized digital media design element optimization generation model;

[0018] The user digital media design element optimization generation module is used to obtain real-time user digital media design element data to be optimized; transmit the real-time user digital media design element data to be optimized to the personalized digital media design element optimization generation model for real-time user digital media design element optimization generation processing to generate real-time user digital media design element optimization data; and transmit the real-time user digital media design element optimization data to the terminal to execute the user digital media design element intelligent optimization operation.

[0019] The beneficial effect of the present application is that, by utilizing the preset digital media API call interface, multi-source data is collected and standardized conversion processing of heterogeneous data is performed, so that the generated digital media design elements can integrate the characteristics of multi-source data, avoid the problem of single data source in the prior art, and more comprehensively understand different digital media design elements, provide more diversified basic data for subsequent optimization generation, and improve the diversity and applicability of the optimization generation of digital media design elements. By analyzing the visual features of digital media to obtain digital media design element data, and analyzing the associated features of digital media design elements, in-depth mining of the associated features between each design element is achieved, and the understanding of different types of design elements is improved, ensuring that the optimized and generated design elements are more in line with user needs in terms of visual effects and overall relevance. By analyzing the user's digital media design content and interactive behavior, it is used to guide the optimized generation of design elements, ensuring that the generated content is more in line with the user's specific preferences, and based on the optimization generation model of the adversarial generation network, the personalized preference characteristics of the user and the design element associated data are fully utilized, and the intelligent and personalized optimization generation of digital media design elements is realized, the diversity of generated content and the matching degree of user preferences are enhanced, and the efficiency and quality of optimized generation are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A schematic diagram of the steps of a digital media design element optimization method using AIGC technology according to the present invention;

[0021] Figure 2 for Figure 1 Detailed implementation steps of step S3 in FIG.

[0022] Figure 3 for Figure 1 Detailed implementation steps of step S4 in FIG.

[0023] Figure 4 for Figure 1 Detailed implementation steps of step S5;

[0024] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0025] The technical method of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0026] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0027] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0028] To achieve this, please refer to Figures 1 to 4 The present invention provides a digital media design element optimization method using AIGC technology, comprising the following steps:

[0029] Step S1: using a preset digital media API calling interface to collect and process digital media multi-source data to generate digital media multi-source data;

[0030] Step S2: performing regional fitting digital media visual analysis based on digital media multi-source data to generate regional fitting digital media visual data; performing digital media design element analysis based on the regional fitting digital media visual data to generate digital media design element data;

[0031] Step S3: Design an association matrix of each type of digital media design element based on the digital media design element data to generate a digital media design element association matrix;

[0032] Step S4: obtaining user digital media design data; performing personalized digital media design preference feature analysis based on the user digital media design data to generate personalized digital media design preference feature data;

[0033] Step S5: Establishing a digital media design element optimization generation model of user preference based on a preset generative adversarial network model and personalized digital media design preference feature data to generate an initial digital media design element optimization generation model; transmitting the digital media design element association matrix to the initial digital media design element optimization generation model to perform personalized optimization generation training processing on the digital media design elements to generate a personalized digital media design element optimization generation model;

[0034] Step S6: Obtain real-time user digital media design element data to be optimized; transmit the real-time user digital media design element data to be optimized to the personalized digital media design element optimization generation model for real-time user digital media design element optimization generation processing to generate real-time user digital media design element optimization data; transmit the real-time user digital media design element optimization data to the terminal to execute the user digital media design element intelligent optimization operation.

[0035] The present invention collects relevant data from multiple data sources through a preset digital media API interface and performs standardized processing on the data. The collection of such multi-source data can cover design elements of different dimensions, so that the subsequent optimization and generation process is based on more comprehensive and rich data, and the diversity and generation effect of digital media design elements are improved. Regional fitting visual analysis is performed on multi-source data, and the visual features of different regions in digital media are extracted and understood to generate regional fitting digital media visual data, which can find the distribution and correlation of design elements in different visual regions, and provide basic data for optimizing design elements. Design element analysis based on regional fitting data can extract the visual features of different design elements and reveal the overall visual effect of design elements. By establishing an association matrix of digital media design elements, the association features between different design elements are deeply excavated and characterized. The association matrix reflects the intrinsic connection of design elements in terms of vision, style, color, etc., and provides an organized structured information for subsequent optimization generation. The mining of such association features helps to maintain the coordination and consistency between design elements during the generation process, thereby improving the overall quality and visual effect of the optimization generation results, so that the generated design is more in line with actual application requirements. Obtain the user's digital media design data, analyze the user's personalized design preferences, better understand the user's style and habits in design, use algorithms such as convolutional neural networks to extract the style in the user's design data, and analyze the corresponding interaction features, so as to achieve quantitative analysis of the user's personalized preferences, and provide user-specific reference standards for the subsequent optimization and generation of design elements, so that the optimized and generated design elements can better meet the user's personalized needs, improve user satisfaction and the pertinence of digital media design. Using the adversarial generative network (GAN) and user preference feature data, establish a user-preferred digital media design element optimization generation model, so that the model can obtain personalized design element optimization generation data according to the user's preferences, and make full use of the user's preference characteristics to enable the generation model to have the ability to learn and generate the user's favorite design style. Transmit the digital media design element association matrix to the generation model for personalized optimization generation training, further strengthen the model's understanding of the association characteristics of different design elements and analyze the needs of optimization generation of various types of design elements, so as to better coordinate the relationship between different design elements during optimization generation. By acquiring real-time user digital media design element data to be optimized, and performing real-time user digital media design element optimization generation processing on the real-time user digital media design element data to be optimized based on the personalized digital media design element optimization generation model, the real-time optimization capability of digital media design elements is realized, and when the user provides the design element data to be optimized, the optimized design elements can be quickly generated and transmitted to the terminal. This meets the user's needs for instant feedback and adjustment during the design process, and greatly improves the design efficiency.

[0036] As an embodiment of the present invention, refer to Figure 1The above is a schematic flow chart of the steps of a digital media design element optimization method using AIGC technology of the present invention. In this embodiment, the digital media design element optimization method using AIGC technology includes the following steps:

[0037] Step S1: using a preset digital media API calling interface to collect and process digital media multi-source data to generate digital media multi-source data;

[0038] In an embodiment of the present invention, a preset digital media API interface is used to call existing data sources (such as image libraries, video libraries, design material libraries, etc.) to collect multi-source data. The API interface obtains original design element data according to the formats of different data sources (JSON, XML, CSV, etc.), including but not limited to images, videos, etc. The data collection frequency is set, and 100-500 data items can be collected for each request. The specific number is determined by the limitations of the API interface. Multi-source data in different formats are uniformly converted into a standard format to generate digital media multi-source data, such as unifying image data into RGB mode and standardizing the size to 256×256 pixels.

[0039] Step S2: performing regional fitting digital media visual analysis based on digital media multi-source data to generate regional fitting digital media visual data; performing digital media design element analysis based on the regional fitting digital media visual data to generate digital media design element data;

[0040] In an embodiment of the present invention, regional fitting digital media visual analysis is performed based on digital media multi-source data, and each image is divided into a grid of fixed size using an image processing library such as OpenCV or PIL. Assuming that the image size is 256×256 pixels, it is divided into 8×8 grids, and the size of each grid is 32×32 pixels. Color histogram, texture features, shape features and other information are extracted in each grid, including: color histogram analysis (using 256 color levels) and texture extraction (using Gabor filter). Visual feature similarity analysis is performed on adjacent grids, cosine similarity or Euclidean distance is calculated, and regional fitting processing is performed on each local grid based on the calculated visual feature similarity, similar regions are fitted as a whole, and regional fitting visual data is generated, including information such as color and shape of the region, and the fitting region is analyzed to extract digital media design elements (such as main color and main shape), and the extracted elements are identified and classified to generate digital media design element data, including detailed information such as the category, quantity, and spatial position of each element.

[0041] Step S3: Design an association matrix of each type of digital media design element based on the digital media design element data to generate a digital media design element association matrix;

[0042] In an embodiment of the present invention, an association rule mining algorithm (such as the Apriori algorithm) is used to analyze the design element data to extract the association relationship between the elements. The association threshold is set to 0.5 (in the range of 0-1, 0.5 and above are considered strong associations). Each design element is regarded as a node in the matrix, and the association relationship is regarded as an edge in the matrix. The values ​​in the matrix represent the strength of the association, ranging from 0 to 1. For example, if the association between a certain color and a certain shape is 0.75, then the value of the corresponding position in the matrix is ​​0.75. Element types include color, shape, texture, layout, etc. The generated digital media design element association matrix is ​​an N×N matrix, where N is the number of design element types, and the association characteristics between elements of each type are clarified.

[0043] Step S4: obtaining user digital media design data; performing personalized digital media design preference feature analysis based on the user digital media design data to generate personalized digital media design preference feature data;

[0044] In an embodiment of the present invention, a design application software (such as image editing software, graphic design tool) equipped with a user interaction interface is used to obtain the digital media design data of the user. The collected data includes the user's operation behavior on each element during the design process, the design element features used (such as color, shape, layout), etc. The embedded plug-in interface is used to collect the user's operation records in the design software in real time, such as material selection, color adjustment, shape stretching, layer movement, etc. Each operation record contains information such as timestamp, operation type, target element features, etc. It is set to collect once per second to ensure that the dynamic behavior of the user in the design process is captured. The pre-trained CNN model (such as ResNet50) is used to extract visual features of the design elements selected by the user, including color histogram, texture features, shape contours, etc. By counting the user's design behavior (such as color adjustment frequency, the number of times a specific shape is used), the user's preference is quantified as a numerical feature. A threshold is set. If the number of times a user adjusts a certain color exceeds 30 times, it is considered that the user prefers the color. The analysis results include the user's preferred colors, common shapes, layout methods, etc., and generate personalized digital media design preference feature data. The data format is a JSON object, which contains the user's main preference characteristics, weight values, and related element categories.

[0045] Step S5: Establishing a digital media design element optimization generation model of user preference based on a preset generative adversarial network model and personalized digital media design preference feature data to generate an initial digital media design element optimization generation model; transmitting the digital media design element association matrix to the initial digital media design element optimization generation model to perform personalized optimization generation training processing on the digital media design elements to generate a personalized digital media design element optimization generation model;

[0046] In an embodiment of the present invention, a preset adversarial generative network model is used to construct an optimized generation model for digital media design elements, wherein personalized digital media design preference feature data is used as a conditional feature of the generation model, including the establishment of a generator and a discriminator network, the input of conditional features, and the introduction of an association matrix for training the optimized generation model. A deep neural network is constructed to generate an optimized design element image. The input of the generator is composed of a random noise vector and personalized digital media design preference feature data, wherein the design preference feature data is used as a conditional feature to guide the generator to generate design elements that meet the personalized needs of the user. Network structure: The generator is composed of a fully connected layer, a batch normalization layer (BatchNormalization), and a deconvolution layer (Transpose Convolution). The conditional vector is extracted from the user preference feature data through the encoder network. An adversarial neural network is constructed to determine whether the input design element image is a real sample or a generated sample. The discriminator receives two parts of input: one part is the generated design element image, and the other part is the corresponding personalized design preference feature data (as a conditional feature). Network structure: The discriminator consists of a convolutional layer, a batch normalization layer, and a fully connected layer, and uses the LeakyReLU activation function. Finally, a probability value of 0-1 is output, indicating the probability that the input image is a real or generated sample. Input parameters: Receive a 256×256 pixel image and a 128-dimensional conditional vector (user preference feature data). The personalized digital media design preference feature data is encoded into a 128-dimensional vector as the conditional input of the generator and the discriminator, reflecting the user's personalized preferences for design elements such as color, shape, style, etc., and guiding the generator network to generate design elements that meet these preferences. During the training process, the generator uses the input random noise and conditional vector to generate design element images, and the discriminator simultaneously receives the generated image and the corresponding conditional vector for discrimination to ensure that the generated design elements meet the user's personalized needs. The digital media design element association matrix is ​​transferred to the generation model. The association matrix contains the association features between design elements, such as the matching relationship between color and shape, and layout. These association features serve as guiding conditions for optimizing training, guiding the design elements generated by the generator to maintain overall coordination and conform to the characteristics of a specific style. During training, the association matrix data is combined with personalized preference feature data to further refine the conditional features, so that the design elements generated by the model can satisfy user preferences and follow the association relationship between design elements. When training the model, the noise (random parameters of the digital media design element association matrix) vector and the conditional vector (user preference feature data) are randomly selected as the input of the generator to generate optimized design element images. The discriminator receives the generated image and conditional features at the same time, and evaluates the generated results in combination with the association matrix. The output of the discriminator is used to update the generator so that the design elements it generates are more in line with the user's personalized needs and associated features.Among them, the design of training parameters includes: learning rate: the learning rate of the generator and the discriminator are both set to 0.0002; batch size: 64 samples are used for each training; number of training rounds: 20,000 rounds, until the design elements generated by the generator pass the judgment of the discriminator, that is, they meet the constraints of the user's preference characteristics and the design element association matrix. The generator adjusts its internal weights according to the feedback of the discriminator in each iteration, so that the generated design elements gradually approach the user's preference characteristics. The discriminator continuously improves its ability to distinguish the authenticity and relevance of the generated image based on the input association matrix and conditional features. After multiple rounds of adversarial training, the generated personalized digital media design element optimization generation model can generate design elements that meet the user's personalized needs and the constraints of the design element association matrix based on the input digital media design element to be optimized data.

[0047] Step S6: Obtain real-time user digital media design element data to be optimized; transmit the real-time user digital media design element data to be optimized to the personalized digital media design element optimization generation model for real-time user digital media design element optimization generation processing to generate real-time user digital media design element optimization data; transmit the real-time user digital media design element optimization data to the terminal to execute the user digital media design element intelligent optimization operation.

[0048] In an embodiment of the present invention, the real-time data interface of the design application software is used to collect the element data that the user is currently designing, such as the color, layout, shape, etc. that the user is adjusting. The data collection frequency is set to collect once per second to ensure that the design element features adjusted by the user in real time are captured. The real-time collected data to be optimized is input into the trained personalized design element optimization generation model. The model optimizes and generates the design elements according to the user's real-time adjustment and preference characteristics, and generates real-time user digital media design element optimization data, including timestamps, optimized element features (such as color values, shape coordinates), etc. The generated data is transmitted to the user terminal through the application interface, replacing the corresponding elements in the current design, and presenting the optimization effect in real time. The design software automatically updates the interface to display the optimized design elements in a visual way, and the user can immediately view the optimization results and make further adjustments.

[0049] Preferably, step S1 comprises the following steps:

[0050] Step S11: using a preset digital media API calling interface to perform initial digital media multi-source data acquisition and processing to generate initial digital media multi-source data;

[0051] Step S12: performing digital media information heterogeneous source type analysis based on the initial digital media multi-source data to generate digital media information heterogeneous source type data;

[0052] Step S13: performing digital media information heterogeneous source structure analysis according to the digital media information heterogeneous source type data to generate digital media information heterogeneous source structure data;

[0053] Step S14: Design a digital media multi-source heterogeneous conversion decision based on the digital media information heterogeneous source structure data, and perform heterogeneous data standardization conversion processing on the initial digital media multi-source data based on the digital media multi-source heterogeneous conversion decision to generate digital media multi-source data.

[0054] The present invention performs initial digital media multi-source data acquisition and processing through a digital media API interface, realizes automatic collection of digital media design-related data from multiple sources, covers multiple types of media data such as images, videos, texts, and sounds, and ensures the diversity and richness of data. The API interface is used to automatically collect data, improve the efficiency and accuracy of data acquisition, reduce the workload of manual data collection, and provide a comprehensive data foundation for subsequent data analysis and design optimization. Heterogeneous source type analysis is performed on the initially collected multi-source data to identify the diversity of data sources and their respective type characteristics, such as images, texts, audio, etc. The analysis can classify and summarize data types, laying the foundation for further structured processing. Heterogeneous source type analysis is used to identify and distinguish the characteristics of different data sources, ensuring that subsequent data processing and optimization processes can make full use of the characteristics of different types of data, thereby providing generalization capabilities for data diversity. Structural analysis is performed on heterogeneous source type data, and the internal structural characteristics of different types of data are deeply mined, such as the resolution and color characteristics of images, the semantic information of texts, the frequency characteristics of audio, etc., which helps to model the internal characteristics of multi-source data and provide guidance for the standardized conversion of heterogeneous data. Comprehensively describe the structure of multi-source heterogeneous data to provide more fine-grained feature information for design element optimization, ensure that subsequent optimization models can fully capture the characteristics of the data, and improve the accuracy of design element optimization generation. Design multi-source heterogeneous conversion decisions through digital media information heterogeneous source structure data, and formulate corresponding standardized processing strategies according to the structural characteristics of different data sources to ensure that heterogeneous data maintains its essential characteristics during the conversion process while being compatible with the input requirements of subsequent models. Through standardized conversion processing, multi-source heterogeneous data is processed for consistency to generate digital media multi-source data that can be used by the model, solving the structural difference problem caused by multi-source heterogeneous data. The standardized data is easier to analyze and process in the future, which improves the applicability and effect stability of the optimization generation model.

[0055] In an embodiment of the present invention, a digital media API interface is used to collect initial data from multiple data sources (such as image libraries, audio libraries, and video libraries). Data sources include open source data sets, social media platforms, digital material libraries, and the like. Set API request parameters, including data types (such as image / jpeg, video / mp4), collection quantity (such as returning 100 data items per request), and request frequency (such as 10 requests per hour). The received data is classified and stored according to type, for example, image data is stored as JPEG files, video data is stored as MP4 files, and audio data is stored as MP3 files. Generate initial digital media multi-source data, including multiple types of media files and their related metadata information. Perform type analysis on the collected initial digital media multi-source data to identify the heterogeneous source types of the data, including file formats, content types, and the like, to generate heterogeneous source type data of digital media information. Use the OpenCV library to read image files and obtain information such as image resolution, color mode, and compression format. For example, read image001.jpg and obtain a resolution of 1024×768 pixels and a color mode of RGB. Use the FFprobe tool to obtain the resolution, frame rate, encoding format, and the like of the video file. For example, the resolution of the video file video001.mp4 is 1920×1080 and the frame rate is 30 fps. The digital media multi-source heterogeneous conversion decision is designed through the digital media information heterogeneous source structure data to unify the data format and structure. All video data are converted into image frame data, and all image data are adjusted to 256×256 resolution, converted to PNG format, and the color mode is unified to RGB. Finally, standardized digital media multi-source data, that is, various types of digital media images, are generated to ensure that all data conform to a unified format and specification.

[0056] Preferably, step S2 comprises the following steps:

[0057] Step S21: extracting digital media local grid visual data from digital media multi-source data to generate digital media local grid visual data;

[0058] Step S22: performing regional fitting processing of adjacent local grid visuals according to the digital media local grid visual data to generate regional fitting digital media visual data;

[0059] Step S23: performing digital media design element analysis on the digital media multi-source data according to the regional fitting digital media visual data to generate digital media design element data.

[0060] The present invention extracts local grid visual data from digital media multi-source data, subdivides digital media data into smaller local areas, and extracts visual features of each local area, such as color, texture, shape, structure, etc. It not only retains global information, but also performs fine analysis on the details of each local area. This makes the design element analysis more targeted at the visual features of digital media. Through the regional fitting processing of adjacent local grid vision, the visual information of the local area is integrated to generate more holistic regional visual data. The regional fitting processing method not only retains local details, but also constructs the visual structure information of the region. The information extraction of each design element of digital media is realized, providing a richer visual feature basis for the subsequent analysis and optimization of design elements. Design element analysis is performed on the digital media visual data after regional fitting, and the main design elements and their features in the digital media can be identified as a whole, including color, shape, texture, light and shadow, etc. This analysis method is based on the global visual information of the fitting area, and more accurately extracts the features of each independent design element.

[0061] In an embodiment of the present invention, each image file is locally gridded, the image is divided into multiple small areas, and visual features are extracted from each area. Each image file is divided into 8×8 grids, and the size of each local grid after division is 32×32 pixels (assuming that the original image size is 256×256 pixels). This grid division can ensure that sufficiently fine-grained visual information is extracted in each local area. Visual features are extracted for each grid, including: color features: the color histogram in the grid is statistically analyzed, and the color information is represented in the form of a distribution of a specific color channel (such as RGB). The extraction of color features can be used to analyze the main color of the grid area. Texture features: texture features such as contrast, uniformity, and entropy are calculated using the local gray level co-occurrence matrix (GLCM) to describe the texture pattern of the local area. Texture features help identify detail changes in the image, such as surface roughness and edge characteristics. Shape features: shape contours in the local grid are detected, and the number, size, and direction of the main contours are recorded. This feature helps analyze the shape elements present in the image. The visual feature data of each local grid will be represented in a table, including the index position of each grid (such as (1, 1) represents the grid in the upper left corner of the image) and the corresponding visual features (color histogram, texture parameters, shape description, etc.) to generate digital local grid visual data, providing a detailed visual feature basis for subsequent regional fitting processing. Using the local grid visual data, the visual similarity between adjacent grids is identified, and similar grids are merged into a larger visual area. The similarity of visual features of each local grid and its adjacent grids is calculated. For example, the color similarity between grids is calculated based on the color histogram (such as using the cosine similarity index), and the similarity threshold is set to 0.8. Only adjacent grids with similarity higher than the threshold are considered for regional fitting. For texture features, the differences in texture parameters (such as contrast and entropy) of adjacent grids are compared. If the difference is within the set threshold range (such as ±10%), the texture is considered similar. Adjacent grids with similarity higher than the threshold are merged to form a larger visual area. For example, if the color similarity of the four grids in the upper left corner of the image is higher than 0.8, they are merged into a large area, indicating that the part has similar visual features. Recalculate the overall visual features of each merged area, including the main color tone, texture features, and shape description of the merged area. This process is used to construct comprehensive visual information of the merged area. Record the location information of each fitting area (such as the coordinates of the upper left corner and the lower right corner), the merged visual features (such as the color value of the main color tone, the mean of the texture features), and the number of grids in the area. The generated regional fitting digital media visual data is used to describe the overall information of similar visual feature areas in the image. Based on the regional fitting digital media visual data, analyze the visual features of each fitting area, identify and extract the design element (such as color, shape, texture) information, and extract the main color of each fitting area as a design element based on the main color tone information in the regional fitting digital media visual data.Count the frequency of each color, determine the main color list in the image (such as red, blue), and use the shape description information to identify the main shape features in the area (such as circles, rectangles, lines). By analyzing the size, position and direction of the shape, determine the main shape design elements in the image, compare the texture features in the area, identify the texture type (such as smooth, rough, grid-like) and record its distribution to determine the texture style of the image, and generate a table containing design element data for each image. The table records the design element type (color, shape, texture), feature description and quantity of each fitting area, such as RGB distribution value, texture distribution value, shape feature; position - (0, 0) to (64, 64); contains 4 grids, generates digital media design element data, provides design element information for each visual area in the image, and provides an analysis basis for subsequent design element optimization and generation.

[0062] Preferably, step S22 includes the following steps:

[0063] Perform local grid visual descriptor analysis based on digital media local grid visual data to generate local grid descriptor data;

[0064] Performing digital media local grid visual similarity analysis on digital media local grid visual data based on local grid descriptor data to generate digital media local grid visual similarity data;

[0065] Based on the digital media local grid visual similarity data, the digital media local grid visual data is subjected to regional fitting processing of adjacent local grid visuals to generate regional fitting digital media visual data.

[0066] The present invention extracts the visual features of each local area, including detailed features in terms of texture, color, shape, edge, etc., by performing visual descriptor analysis on local grid visual data, and forms visual descriptor data. These descriptors, as independent representations of the area, help to fully capture the unique visual information of each local area. The generated local grid descriptor data provides accurate and rich feature basis for subsequent similarity analysis and regional fitting, ensures the detailed description of visual features, enables the visual features of different local areas to be quantitatively analyzed, and improves the accuracy and effect of overall visual data processing. Using local grid descriptor data for similarity analysis, it is possible to compare the visual features between local grids and quantify the similarity between them. It is possible to identify which local areas are visually similar, so as to determine whether these areas are merged or processed as a whole to a certain extent, effectively avoid the merging of uncoordinated areas, improve the rationality of visual area fitting, and make the generated regional visual data more coherent as a whole. Based on the digital media local grid visual similarity data, the digital media local grid visual data is subjected to regional fitting processing of adjacent local grid vision, and the local areas with similar visual features are merged into a larger overall visual area. This fitting processing method ensures the visual consistency and coherence within the merged region, thereby generating more holistic and coordinated regional visual data. The visual data after regional fitting contains more comprehensive visual feature information, providing a more global basis for subsequent design element analysis and optimization, that is, reflecting the overall information of each digital media design element. By fitting adjacent grids, not only local details are retained, but also the overall visual structure is constructed, which improves the aesthetics and practicality of the generated results.

[0067] In an embodiment of the present invention, a visual descriptor analysis is performed on each local grid to extract and characterize the visual features of each grid. Each image is divided into an 8×8 grid, and the image is divided into multiple small areas (such as each grid is 32×32 pixels). A visual descriptor is extracted for each local grid, including the following features: a color histogram of each grid is calculated to reflect the color distribution of the local area. The color histogram can be divided into 256 levels according to the RGB channels to describe the main color features of the grid. The texture features of the grid are extracted using the local binary pattern (LBP). The LBP descriptor encodes the neighborhood relationship of each pixel and generates a histogram of the texture pattern, which reflects the surface texture of the local grid. Edge detection (such as the Sobel operator) is used to identify the main shape contours in the grid. The shape features of each grid are analyzed. The visual features of each grid are combined into a descriptor dataset, including color histogram, texture histogram and shape features. Each grid descriptor is used to comprehensively characterize the visual information of the local area. The generated local grid descriptor data is recorded in a table, for example, the table contains grid index position, color histogram, texture histogram and shape feature information. Using the local grid descriptor data, the visual similarity between each grid and its adjacent grids is calculated to determine whether these grids belong to the same area visually. The color histogram of each grid is similar to the color histogram of its adjacent grid, and the cosine similarity is used to measure the similarity between the two. Histogram similarity. Set the similarity threshold to 0.8. Only when the similarity is higher than this threshold, the two grids are considered similar in color features. Compare the texture descriptors (LBP histograms) of each grid with adjacent grids, and use the chi-square statistic to measure texture similarity. Set the texture similarity threshold to 0.7 to distinguish the differences in texture patterns. Compare the shape descriptors (edge ​​features) of the grids and count the number and direction differences of similar shapes in the two grids. If the number of similar shapes exceeds 50% of the total number of shapes and the shape direction difference is less than 30 degrees, the shapes are considered similar. Record the similarity information of each grid with its adjacent grids and generate a visual similarity data table. For example, the table contains the grid index, the index of the adjacent grid, color similarity, texture similarity, shape similarity, etc. Based on the visual similarity data, the adjacent grids are regionally fitted, and the grids with similar visual features are merged into a larger region. Set the comprehensive similarity threshold for regional fitting: For adjacent grids, only when the color similarity, texture similarity, and shape similarity are higher than their respective thresholds (such as color>0.8, texture>0.7, shape>50%), these grids are fitted into a whole region. Using the regional growing algorithm, starting from the grid in the upper left corner of the image, the similarity data of its adjacent grids are checked one by one. If the fitting conditions are met, the adjacent grids are merged into the current region.According to the preset similarity conditions, starting from the grid in the upper left corner, recursively check the visual similarity of adjacent grids, and merge the grids that meet the conditions into one region. Recalculate the overall visual features for each merged region, including the main color tone (the highest value of the color histogram), overall texture pattern, and shape features of the merged region. Record the location information of each fitting region (such as the coordinates of the upper left and lower right corners), the merged visual features (color, texture, shape), and the number of grids contained in the region. The generated regional fitting digital visual data is used to describe the overall information of similar visual feature regions in the image.

[0068] Preferably, step S3 comprises the following steps:

[0069] Step S31: performing digital media visual region abstract feature analysis based on the region fitting digital media visual data to generate digital media visual region abstract feature data;

[0070] Step S32: using the digital media visual region abstract feature data as cluster labels to perform digital media visual region cluster analysis on the regional fitting digital media visual data to generate clustered digital media visual region data;

[0071] Step S33: performing digital media design element correlation feature analysis based on the clustered digital media visual area data to generate digital media design element correlation feature data;

[0072] Step S34: Design an association matrix for each type of digital media design element according to the digital media design element association feature data and the digital media design element data to generate a digital media design element association matrix.

[0073] The present invention performs abstract feature analysis on the digital media visual data after regional fitting, and extracts high-level features of the visual area, such as color distribution, texture pattern, shape outline and composition features. These abstract features reflect the overall style and characteristics of the visual area, and can express the feature information of the visual area at a higher level of abstraction. The generated abstract feature data provides a basis for subsequent clustering and association analysis, so that the features of the visual area can be better characterized and distinguished. Through the abstract feature analysis, the depth of understanding of the visual area is improved, which helps to more accurately grasp the uniqueness and coordination of each area when optimizing the design elements. The visual area is clustered using the abstract feature data as a clustering label, and the visual areas with similar abstract features are classified into one category, and different visual styles, color combinations and composition patterns in digital media are identified, so as to better understand the relationship between design elements. The generated clustered digital media visual area data provides a classification basis for the subsequent design element association feature analysis, so that the subsequent processing can be carried out for each type of visual area, thereby improving the accuracy and pertinence of the design element analysis, ensuring that the visual elements are consistent and coordinated in the subsequent optimization, and improving the overall effect of the generated design. The design element correlation feature analysis is performed on the clustered visual area data, and the correlation features within the visual area and between different areas are deeply explored, including color matching, shape combination, spatial layout, etc., revealing the mutual relationship and influence mechanism between the design elements, and providing important correlation feature data for building the optimization model. The generated design element correlation feature data provides a guiding basis for the subsequent correlation matrix design, ensuring that the optimized design elements have internal consistency in overall style and layout, and better grasping the coordination between design elements, thereby improving the adaptability and practicality of the optimization generation model. By using the design element correlation feature data and digital media design element data for correlation matrix design, the correlation relationship between the design elements can be established, and these relationships can be structured in the form of a matrix. The correlation matrix can intuitively describe the degree of mutual influence and characteristic association between design elements, and provide clear constraints and guidance for the optimization generation process. The generated correlation matrix acts as a "rule set" in the subsequent optimization generation model, ensuring that the generation of design elements meets the expected visual effects and user needs, and can better maintain the coordination and unity between design elements, thereby improving the quality and efficiency of the optimization generation of digital media design elements.

[0074] As an embodiment of the present invention, refer to Figure 2 As shown, Figure 1 Detailed implementation step flow chart of step S3 in the embodiment, step S3 in this embodiment includes:

[0075] Step S31: performing digital media visual region abstract feature analysis based on the region fitting digital media visual data to generate digital media visual region abstract feature data;

[0076] In an embodiment of the present invention, an abstract feature analysis is performed on the regional fitting digital media visual data to extract the global features of each visual area, such as color theme, texture type, shape features, etc. The color data of each fitting area is subjected to principal component analysis (PCA) to find the main color components in the area and determine the color theme of the area (such as blue, warm tones). The texture features of the area are calculated using the local binary pattern (LBP) or gray level co-occurrence matrix (GLCM) to identify the texture type (such as smooth, rough, grid-like). The main shape features of each area (such as circle, rectangle, curve) are extracted through edge detection (such as Canny operator) and shape analysis. The number, size, direction and other information of the shape are recorded. Abstract feature data of the digital media visual area is generated, for example, such as area ID (such as "Region 1"), color theme (such as "blue"), texture type (such as "smooth"), shape features (such as "circle, number = 3, average size = 20 pixels").

[0077] Step S32: using the digital media visual region abstract feature data as cluster labels to perform digital media visual region cluster analysis on the regional fitting digital media visual data to generate clustered digital media visual region data;

[0078] In the embodiment of the present invention, the abstract feature data (color theme, texture type, shape feature) of the visual area is used as the input feature, and the K-Means clustering algorithm is used to classify all the areas, and the number of clusters K is set to 5-10 to distinguish the main visual style in the image, and the color theme, texture type and shape feature of the visual area are used as cluster labels. A label is generated for each clustering result, such as "Cluster 1: Blue-Smooth Area", and the area ID contained in each cluster is recorded, and finally the clustered digital media visual area data after clustering and classifying the regional fitting digital media visual data is obtained.

[0079] Step S33: performing digital media design element correlation feature analysis based on the clustered digital media visual area data to generate digital media design element correlation feature data;

[0080] In the embodiment of the present invention, clustered digital media visual area data is used to analyze the design element association features in various visual areas to find common design element combinations. In each cluster, the deviation between each cluster subset and the midpoint is calculated, and based on the calculation results, the Apriori algorithm is used to mine the association rules between the design elements in each cluster, find strong association features, and generate digital media design element association feature data.

[0081] Step S34: Design an association matrix for each type of digital media design element according to the digital media design element association feature data and the digital media design element data to generate a digital media design element association matrix.

[0082] In an embodiment of the present invention, the digital media design element association feature data and the digital media design element data are used to construct an association matrix of the design elements to quantify the association between the types of design elements. According to the type of design element (such as color, shape, texture), an N×N matrix is ​​constructed, where N is the total number of design element types. The support and confidence in the association feature data are used to fill in the positions of different design element combinations in the matrix. For example, if the association confidence between the color "blue" and the shape "circle" is 85%, 0.85 is filled in the corresponding position in the matrix. Each element (i, j) in the matrix represents the strength of association between the i-th design element and the j-th design element, ranging from 0 to 1. The higher the value, the stronger the association, and the digital media design element association matrix is ​​generated.

[0083] Preferably, step S4 comprises the following steps:

[0084] Step S41: Obtaining user digital media design data;

[0085] Step S42: performing user digital media design content extraction processing according to the user digital media design data to generate user digital media design content data;

[0086] Step S43: using a preset convolutional neural network algorithm to perform digital media design content style analysis on the user digital media design content data to generate user digital media design content style data;

[0087] Step S44: extracting user digital media design interaction behaviors according to the user digital media design data to generate user digital media design interaction behavior data;

[0088] Step S45: performing a quantitative analysis of the user's digital media design interaction behavior preference based on the user's digital media design interaction behavior data to generate the user's digital media design interaction behavior preference quantitative data;

[0089] Step S46: Map the user digital media design interaction behavior preference quantified data to the user digital media design content style data to perform personalized digital media design preference feature analysis to generate personalized digital media design preference feature data.

[0090] The present invention obtains the digital media design data of the user, collects the information such as the elements, styles, layouts, and interaction methods used by the user in the design process, and provides basic data for analyzing the user's design habits, preferences, and design styles. It ensures that the subsequent personalized design analysis is based on the real user design behavior and work data, ensures the accuracy of personalized analysis, and provides reliable data support for the optimization and generation of design elements. By extracting the content of the user's design data, the main element features in the user's design can be obtained, including visual style, layout composition, color matching, font selection, etc., and the user's design data is analyzed as a whole to generate content data for analysis. The generated design content data can provide a rich information basis for subsequent style analysis and interactive behavior analysis, ensuring that the optimization process can fully consider the user's design content characteristics, thereby improving the personalization of design element generation. The convolutional neural network (CNN) is used to perform style analysis on the user's design content, and the design style features are extracted from multiple dimensions, such as color, texture, layout, element combination, etc. The convolutional neural network has a strong image feature extraction capability and can accurately capture the style features in the user's design. The generated user design content style data reflects the user's style preference in design, provides a style basis for the subsequent optimization of design elements, ensures that the generated design elements are more in line with the user's aesthetic and design habits in style, and improves user satisfaction. By extracting the user's interactive behavior in the design process (such as operation sequence, element adjustment, color selection, etc.), we can fully understand the user's preferences and habits in design, collect the user's adjustment and processing methods for different design elements, and provide more detailed interactive information. The generated interactive behavior data helps to further quantify the user's design behavior, providing a basis for analyzing the user's design preference characteristics, ensuring that the optimized generated design elements not only conform to the user's visual style, but also fit the user's operating habits. By quantitatively analyzing the user's interactive behavior, the user's design habits and preferences are converted into numerical form, and the user's preferences for color, layout, element usage, etc. in the design process are extracted, providing accurate quantitative indicators for the analysis of user preferences. The generated interactive behavior preference quantitative data provides a computable reference standard for subsequent personalized design, which helps to optimize the generation model to better understand the user's preferences, thereby achieving a higher degree of personalized matching when generating design elements. By mapping and comprehensively analyzing the quantitative data of users' interactive behavior preferences with the design content style data, we can comprehensively and deeply explore the users' personalized design preference characteristics, and generate more refined personalized design feature data by combining the users' style preferences and operating habits.

[0091] As an embodiment of the present invention, refer to Figure 3 As shown, Figure 1 Detailed implementation step flow chart of step S4 in the embodiment, step S4 in this embodiment includes:

[0092] Step S41: Obtaining user digital media design data;

[0093] In the embodiment of the present invention, design software (such as Photoshop, Illustrator) equipped with user behavior tracking and data recording functions is used to obtain user digital media design data. The collected data includes elements used by users in the design process, operation behaviors and adjustment parameters, etc. A plug-in interface is integrated in the user design software to record the user's design operations in real time, such as material selection, color adjustment, shape drawing, layer movement, etc. The data collection frequency is set to once per second to capture the user's design dynamic behavior and operation details. When storing data, timestamps and operation type labels (such as "color adjustment", "shape insertion"), as well as design element features (such as color value, shape size, position coordinates, etc.) are attached. Generate original user digital media design data, including operation behavior, timestamp, design element features and other information.

[0094] Step S42: performing user digital media design content extraction processing according to the user digital media design data to generate user digital media design content data;

[0095] In an embodiment of the present invention, content extraction is performed on user design data to identify the main elements and style features used by the user in the design. The colors used by the user in the design are counted, and the color space distribution of commonly used colors is determined by reading the RGB value of each design element. When extracting color data, 256 levels of color scale are used to quantify the color. The shape types (such as rectangles, circles, polygons) inserted in the user design and their attributes (such as size, position, rotation angle) are recorded. The shape size and position are measured in pixels, and the angle is expressed in degrees. The text elements in the design are analyzed, and the font type, font size, color, etc. are recorded. The extracted design content data is organized into a table form to generate user digital media design content data, including color, shape, layout, text and other features.

[0096] Step S43: using a preset convolutional neural network algorithm to perform digital media design content style analysis on the user digital media design content data to generate user digital media design content style data;

[0097] In an embodiment of the present invention, a convolutional neural network (CNN) model is used to perform style analysis on user design content data to identify the user's preferred style in design. A preset CNN model (such as ResNet-50) is used, and the model input is the user design content data (image data or extracted design element features). The input size of the network is set to 224×224 pixels. The network training uses the Adam optimizer, the learning rate is set to 0.0001, the batch size is 32, and the number of iterations is 5000. The model is trained using a pre-collected design style data set so that it can recognize a variety of design styles (such as modern, retro, minimalism, etc.). The user's design content is input into the trained CNN model, the deep features are extracted, and the design style is classified through the fully connected layer, and the style category and its corresponding confidence (such as modern style-confidence 85%) are output. At the same time, the model extracts the color, shape, layout and other features used in the user's design, and generates the user's digital media design content style data to describe the overall style of the user's design, including style categories, main design element features and their weights.

[0098] Step S44: extracting user digital media design interaction behaviors according to the user digital media design data to generate user digital media design interaction behavior data;

[0099] In an embodiment of the present invention, the user's interactive behavior during the design process is analyzed, key behavioral features are extracted, and user digital media design interactive behavior data is generated. The types and frequencies of operations performed by users during the design process are counted, such as the number of color adjustments, the number of shape insertions, layer movement operations, etc. A threshold is set, such as when the number of color adjustments exceeds 30 times, it is considered that the user has a preference for that color. The user's operation sequence and mode are recorded, for example: color adjustment-shape insertion-text addition. The user's operating habits are analyzed (such as a tendency to adjust the color before adding a shape). Each behavioral feature is quantified, such as the color adjustment frequency and the number of shape insertions, and the mean and standard deviation of the interactive behavior are calculated to reflect the user's design habits. Generate user digital media design interactive behavior data and record user operation characteristics and behavior patterns.

[0100] Step S45: performing a quantitative analysis of the user's digital media design interaction behavior preference based on the user's digital media design interaction behavior data to generate the user's digital media design interaction behavior preference quantitative data;

[0101] In an embodiment of the present invention, the user's interactive behavior data is quantitatively analyzed to extract the user's design preference characteristics. The user's interactive behavior data is statistically analyzed to calculate the average, maximum, minimum and standard deviation of each behavior. For example, the average color adjustment frequency is 25 times / hour and the standard deviation is 5 times, indicating that the user often adjusts the color and has a relatively stable preference. The Z-Score standardization method is used to quantify the behavioral characteristics. For example, the color adjustment preference score is 0.75, indicating that the user has a high preference for color adjustment. Generate quantitative data on user digital media design interactive behavior preferences, including behavior type (such as color adjustment), preference score, behavior frequency, etc.

[0102] Step S46: Map the user digital media design interaction behavior preference quantified data to the user digital media design content style data to perform personalized digital media design preference feature analysis to generate personalized digital media design preference feature data.

[0103] In an embodiment of the present invention, the user's interactive behavior preference quantitative data is mapped to the design content style data to generate comprehensive personalized design preference feature data. The interactive behavior preference quantitative data (such as color adjustment preference score) is associated with the design content style data (such as color theme) to determine the user's preference characteristics. For example, if the user has a high frequency of adjusting the blue hue and the blue color theme appears many times in the design style, the comprehensive analysis shows that the user prefers the blue design style. The preference of each design element is comprehensively scored by weighted average. The interactive behavior preference weight is set to 0.6 and the design content style weight is set to 0.4 to generate an overall preference feature. The output includes feature data such as color preference (such as blue, weight 0.8), shape preference (such as round, weight 0.6), style tendency (such as minimalist style, confidence 85%), etc. Generate personalized digital media design preference feature data to provide personalized reference for subsequent design element optimization and generation model.

[0104] Preferably, step S5 comprises the following steps:

[0105] Step S51: Establishing a generator network for optimizing digital media design elements based on the fully connected layer and the deconvolution layer of the preset adversarial generative network model, and generating a generator network for optimizing digital media design elements;

[0106] Step S52: establishing a user preference discriminator network for optimizing digital media design elements based on personalized digital media design preference feature data, and generating a digital media design element optimization discriminator network;

[0107] Step S53: constructing an adversarial generation mapping relationship between digital media design elements and user-preferred digital media design elements based on the digital media design element optimization generator network and the digital media design element optimization discriminator network, and generating an initial digital media design element optimization generation model;

[0108] Step S54: transmitting the digital media design element association matrix to the initial digital media design element optimization generation model to perform personalized optimization generation training processing of the digital media design elements, and generating a personalized digital media design element optimization generation model.

[0109] The present invention constructs a powerful digital media design element optimization generator network by using the fully connected layer and deconvolution layer of the adversarial generative network (GAN). The generator network generates design elements that meet user preferences based on the input digital media design element data. The fully connected layer can capture high-dimensional features, and the deconvolution layer can gradually restore these features to design elements with visual effects, and make full use of user preference features to generate more refined and personalized design elements, providing higher quality initial generation data for subsequent optimization models, and ensuring that the generated design elements are more in line with user needs and aesthetic standards. A user preference discriminator network is established. By using personalized digital media design preference feature data, the generated design elements can be evaluated to determine whether they meet the user's design preferences. The discriminator network provides a feedback mechanism for the optimization generation model, so that the generator network can continuously adjust the generation strategy and optimize in a direction that is more in line with user preferences. The establishment of the discriminator network ensures the accurate capture of user personalized needs during the optimization process, can timely discover and correct deviations in generation, and improve the quality of the final generated design elements and user satisfaction. By combining the generator network and the discriminator network, an adversarial generation mapping relationship is constructed, so that the optimized generation model can continuously adjust the generation strategy in the adversarial process and improve the quality of the generated design elements. The adversarial generation process can simulate the user's aesthetics and preferences to achieve more accurate optimization of design elements. It ensures that the optimized generation model is continuously optimized during the training process, and the generated design elements gradually approach the user's preferences. Through adversarial training, the initial model has a stronger generalization ability, which provides a good foundation for the further improvement of the subsequent optimization model. The digital media design element association matrix is ​​transferred to the generation model, which can integrate the association characteristics between the design elements into the optimization process, so that the generated design elements can maintain overall coordination while being personalized. This training processing method ensures that the association of design elements in terms of style, layout, color, etc. is fully utilized. After the optimization training processing of the association matrix, the generated personalized digital media design element optimization generation model can better meet the diverse needs of users. The generated model dynamically adjusts the design elements according to the user's preferences, improves the accuracy and consistency of the design effect, and significantly enhances the optimization quality of digital media design elements.

[0110] As an embodiment of the present invention, refer to Figure 4 As shown, Figure 1 FIG. 5 is a flow chart of detailed implementation steps of step S5, in this embodiment, step S5 includes:

[0111] Step S51: Establishing a generator network for optimizing digital media design elements based on the fully connected layer and the deconvolution layer of the preset adversarial generative network model, and generating a generator network for optimizing digital media design elements;

[0112] In an embodiment of the present invention, a generator network in a generative adversarial network (GAN) is used to optimize design elements. The generator network is constructed by a fully connected layer and a deconvolution layer, and is intended to generate design elements that meet user preferences based on the input personalized features. The establishment of a generator network for optimizing digital media design elements includes a generator network structure and network parameter settings. The generator network structure includes: Input layer: The input is a random noise vector (i.e., random digital media design element associated data) and personalized digital media design preference feature data (such as color, shape, style). The dimension of the random noise vector is set to 100 dimensions, and the personalized feature data is encoded as a 128-dimensional conditional vector. The final input is a 228-dimensional vector. Fully connected layer: The generator network contains 3 fully connected layers, and the number of neurons in each layer is 512, 1024, and 2048, respectively. The ReLU activation function is used, and batch normalization is used to stabilize the training process. Deconvolution layer: After the fully connected layer, 3 deconvolution layers (Transpose Convolution Layer) are used to convert the input vector into two-dimensional image data. Each deconvolution layer uses a 4×4 convolution kernel with a step size of 2 to gradually enlarge the size of the feature map, and finally outputs a 256×256 pixel design element image. Output layer: The output is a design element image that meets personalized design preferences, and the Tanh activation function is used to normalize the pixel values ​​to the range of [-1, 1]. Network parameter settings include: Optimizer: Adam optimizer, learning rate set to 0.0002, β1=0.5. Loss function: The mean square error (MSE) is used as the loss function of the generator to measure the match between the generated image and the target feature. Generate a digital media design element optimization generator network, which can generate a design element image that is biased towards personalized preferences based on the input digital media design element data.

[0113] Step S52: establishing a user preference discriminator network for optimizing digital media design elements based on personalized digital media design preference feature data, and generating a digital media design element optimization discriminator network;

[0114] In an embodiment of the present invention, a discriminator network is constructed to evaluate whether the generated design elements meet the personalized preferences of the user, and the discriminator network discriminates the input image and personalized feature data through a convolutional neural network (CNN) structure. The establishment of the user preference discriminator network for optimizing the digital media design elements includes the analysis of the discriminator network structure and the setting of network parameters. The discriminator network structure analysis includes: input layer: the input is the generated design element image (256×256 pixels) and the corresponding personalized digital media design preference feature data (128-dimensional conditional vector). The image is input to the convolutional layer, and the conditional vector is connected to the fully connected layer. Convolutional layer: The discriminator network contains 4 convolutional layers, each layer uses a 4×4 convolution kernel with a step size of 2, and extracts low-level to high-level features of the image in turn. The LeakyReLU activation function (α=0.2) is used, and batch normalization is added after each convolutional layer to avoid gradient vanishing and explosion. Fully connected layer: After the convolutional layer, the extracted features are flattened and connected to two fully connected layers. The input conditional vector is combined with the feature vector for further feature fusion. Output layer: Use the Sigmoid activation function to output a probability value, indicating whether the input design element image meets the user's personalized preferences. The output range is [0, 1]. The closer the value is to 1, the more it meets the user's preferences. Network parameter settings include: Optimizer: Adam optimizer, learning rate set to 0.0002, β1=0.5. Loss function: Use the binary cross-entropy loss function (Binary Cross-EntropyLoss) to measure the classification accuracy of the discriminator for real images and generated images. Generate a digital media design element optimization discriminator network to determine the match between the generated design element image and the user's preferences, and feed it back to the generator network for re-generation optimization adjustment.

[0115] Step S53: constructing an adversarial generation mapping relationship between digital media design elements and user-preferred digital media design elements based on the digital media design element optimization generator network and the digital media design element optimization discriminator network, and generating an initial digital media design element optimization generation model;

[0116] In the embodiment of the present invention, an adversarial generation mapping relationship between digital media design elements and optimized generation of digital media design elements preferred by users is constructed based on a digital media design element optimization generator network and a digital media design element optimization discriminator network. The generator and the discriminator update parameters alternately to form an adversarial relationship. The generator continuously optimizes the generation strategy to deceive the discriminator, and the discriminator continuously improves its discrimination ability, prompting the generator to generate design elements that are more in line with user preferences. The adversarial generation mapping relationship also sets the number of training cycles and the training cycle end conditions through a recurrent neural network, and finally obtains the initial digital media design element optimization generation model.

[0117] Step S54: transmitting the digital media design element association matrix to the initial digital media design element optimization generation model to perform personalized optimization generation training processing of the digital media design elements, and generating a personalized digital media design element optimization generation model.

[0118] In an embodiment of the present invention, the digital media design element association matrix is ​​used as an input condition and transmitted to the initial optimization generation model, and the generator is further optimized and trained to enhance the personalized effect. The digital media design element association matrix contains the association relationship between design elements (such as color and shape, layout, etc.). The association matrix is ​​set to an N×N matrix, where N is the number of design element categories. Each element (i, j) represents the association degree between the i-th design element and the j-th design element, ranging from 0 to 1. During the training process, the digital media design element association matrix is ​​fused with the personalized digital media design preference feature data and passed into the generator network as a conditional input, so that the generator generates design elements that meet the association relationship and user preferences. The training process is set to 10,000 iterations and the batch size is 64. The generator generates a design element image based on the association matrix and preference features, and the discriminator evaluates the generated image. The association degree of the association matrix is ​​used as an additional constraint to guide the generator to optimize the generation strategy to ensure that the generated design elements not only meet user preferences, but also follow the association relationship between design elements. Generate a personalized digital media design element optimization generation model, so that the model can be used to generate design element images that conform to specific styles and associations based on user preferences and design element associations.

[0119] Preferably, step S52 includes the following steps:

[0120] Perform optimization condition feature analysis of digital media design elements based on personalized digital media design preference feature data to generate optimization condition feature data of personalized digital media design elements;

[0121] Perform personalized digital media design element optimization discrimination decision design based on personalized digital media design element optimization condition characteristic data, and generate personalized digital media design element optimization discrimination decision;

[0122] A discriminator network for optimizing digital media design elements is established according to personalized digital media design element optimization discrimination decisions, and a digital media design element optimization discriminator network is generated.

[0123] The present invention performs optimization condition feature analysis on personalized digital media design preference feature data, and extracts feature conditions that are closely related to user preferences, such as color matching, shape selection, composition style, etc. Clear optimization conditions are provided for optimizing the generation model, which helps to guide the generation of design elements in the direction of user preferences. Ensure that the subsequent discriminator network can accurately evaluate according to the user's personalized needs when judging the generation results. This feature analysis helps to improve the accuracy of the judgment and make the optimization process more in line with the user's personalized design requirements. By using personalized optimization condition feature data for discriminant decision design, a set of evaluation criteria for user preferences is provided for the discriminator network. This discriminant decision design covers a variety of visual features, such as style consistency, element coordination, color matching, etc., which are used to identify which elements meet the user's design preferences and which need to be adjusted. In this way, the generator network can be continuously guided to adjust in the direction that meets user needs during the optimization process, thereby improving the quality of optimized generation. By establishing a discriminator network based on personalized discriminant decisions, the discriminator has the ability to discriminate generated design elements. The discriminator network compares the personalized discriminant decision with the actual generated result, evaluates its matching degree with the user's preference, provides effective feedback in the optimization process, and guides the generator to adjust the generation strategy. The established discriminator network can dynamically learn and adapt to the user's design preference characteristics, continuously optimize the evaluation criteria in the generation process, and improve the recognition and discrimination ability of design elements. The generated design element optimization discriminator network ensures the effectiveness of the adversarial generation process, making the generated design elements more in line with the user's personalized needs.

[0124] In an embodiment of the present invention, the personalized digital media design preference feature data is analyzed to extract the conditional features required by the user in the optimization of the design elements to guide the optimization discrimination strategy of the discriminator network. Multidimensional feature extraction is performed on the personalized digital media design preference feature data of the user, including color preference, shape preference, style tendency, etc. Each feature is represented by a specific numerical value, for example: a color histogram is used for encoding to generate a 256-dimensional feature vector, which represents the color distribution of the user's preference. The shape types and sizes commonly used by the user in the design are recorded and represented by a 10-dimensional vector, including common shapes (such as circles, rectangles, polygons) and their frequency of use. The user's design style is classified using a CNN model (such as ResNet-50) to generate an N-dimensional style vector, where N is the number of style categories, for example, N=8 represents 8 different design styles. The above features are combined to form a comprehensive conditional feature vector. Assuming that the user's color, shape, and style preferences are 256-dimensional, 10-dimensional, and 8-dimensional, respectively, the final personalized design element optimization conditional feature vector is 274-dimensional. The vector is normalized and the feature value range is normalized to [0, 1] to facilitate the use of the discriminator network in subsequent steps, and finally the optimized condition feature data of personalized digital media design elements is generated. Using the optimized condition feature data of personalized digital media design elements, a discriminant decision strategy is designed to evaluate whether the design elements meet the personalized preferences of users. The personalized condition feature data (274-dimensional vector) is matched with the design element features to set the discrimination criteria. For example, for color preference, the similarity threshold between color histograms is set to 0.8. Only when the color distribution of the generated design element is higher than the threshold, it is judged to meet the user's preference. For shape features, a preference matching threshold is set. For example, if the matching degree between the shape type contained in the generated element and the user's preferred type exceeds 70%, the discrimination condition is met. For style features, the cosine similarity is calculated using the style feature vector extracted by CNN, and the similarity threshold is set to 0.85. If it is higher than this value, it is judged to meet the user's style preference. The discrimination rules of each feature are combined into a comprehensive discrimination decision. The weighted average method is used to synthesize the discrimination results of each feature, and the weights of color, shape, and style are set to 0.4, 0.3, and 0.3 respectively to obtain the final discrimination result. Generate personalized digital media design element optimization discrimination decisions to provide discrimination criteria for the subsequent establishment of a discriminator network. Based on the personalized digital media design element optimization discrimination decision, a discriminator network is established to evaluate the matching degree between the generated design elements and the user preference features. The establishment of the user preference discriminator network for digital media design element optimization includes discriminator network structure analysis and network parameter setting. The discriminator network structure analysis includes: Input layer: The input is a design element image (256×256 pixels) and a personalized conditional feature vector (274 dimensions). Image data is input into the convolutional layer, and the conditional feature vector is input into the fully connected layer.Convolutional layer: 4 convolutional layers are designed, with a convolution kernel size of 4×4 and a stride of 2. Each layer uses the LeakyReLU activation function (α=0.2). The features in the image are extracted through the convolution operation, and finally a high-level feature representation is output. Fully connected layer: The image features output by the convolutional layer are fused with the personalized conditional feature vector in the fully connected layer. The fully connected layer contains 512 neurons, uses the LeakyReLU activation function, and batch normalization is used for stable training. Output layer: The Sigmoid activation function is used to output a probability value indicating whether the input image meets the user's personalized judgment decision. The output range is [0, 1]. The closer the value is to 1, the more it meets the user's preference. The network parameter settings include: Optimizer: Use the Adam optimizer, set the learning rate to 0.0002, and β1=0.5. Loss function: Use the binary cross-entropy loss function to measure the accuracy of the discriminator on the input data. Through iterative training, the network parameters are adjusted to improve the discriminator's ability to discriminate design elements. The digital media design element optimization discriminator network is generated to perform personalized discrimination on design elements and improve the optimization effect of the generated model.

[0125] Preferably, step S54 includes the following steps:

[0126] The digital media design element association data of each category in the digital media design element association matrix is ​​transmitted to the digital media design element optimization generator network of the initial digital media design element optimization generation model to perform the digital media design element generation training tasks of each category, and the digital media design element optimization discriminator network of the initial digital media design element optimization generation model is used to perform the discriminant optimization training processing of digital media design element generation on the digital media design element generation training tasks of each category to obtain a personalized digital media design element optimization generation model.

[0127] The present invention transmits the associated data of each category in the digital media design element association matrix to the generator network, provides the model with the associated features between the design elements of different categories, including the visual coordination, layout logic and style consistency between the elements, so that the generator network can learn and simulate the relationship between these elements during training, ensuring that the generator network can not only generate independent design elements, but also optimize according to the associated characteristics between the elements, thereby generating design works with overall coordination. By transmitting the associated data, the model can better capture the intrinsic connection between the multi-category design elements, and improve the quality and diversity of the optimized generation results. During the training process, the discriminator network evaluates the generated design elements of each category according to the personalized discrimination decision established before. The discriminator can identify whether the design elements generated by the generator conform to the characteristic relationship in the association matrix and the personalized preferences of the user, provide instant feedback to the generator, effectively guide the generator network to adjust the generation strategy, and continuously optimize the generation quality of the design elements of each category. By discriminating and optimizing the generation results, the model can generate design elements that are more in line with user needs and have coordination, and improve the accuracy and practicality of the personalized design element optimization generation model.

[0128] In an embodiment of the present invention, data in the digital media design element association matrix is ​​transmitted to the generator network to guide the generator to generate an image that conforms to the design element combination according to the association relationship. The association data that meets the threshold condition in the association matrix is ​​encoded as a conditional vector. For each training job, the generator network takes random noise and the conditional vector as input to generate a design element image that conforms to these association relationships. The random noise dimension is 100, and the dimension of the association matrix conditional vector is consistent with the matrix size. Perform design element generation training for each category. Conditional generation of the generator is performed according to different association relationships. For example, when the "blue-circle" association condition is transmitted, the generator will generate an image containing a blue circular element according to the condition. The generator network training parameter settings are: learning rate: 0.0002; batch size: 64; number of iterations: 10,000 times. Until the image generated by the generator can better reflect the association relationship set in the association matrix. The generator network learns to generate images that conform to the association of various design elements under the guidance of the association matrix, providing a basis for subsequent optimization training. The output of the generator is optimized and trained using the constructed discriminator network. The discriminator receives the image generated by the generator and discriminates based on the data in the association matrix, thereby adjusting the generator's generation strategy. The discriminator receives the image generated by the generator and the conditional vector of the association matrix, and outputs a probability value indicating whether the generated image meets the set association relationship (such as the association between color and shape). The discriminator network evaluates the generated image based on the actual association relationship (such as blue and circle). The closer the output probability is to 1, the more the generated image meets the design element association defined in the association matrix. Using the binary cross entropy loss function, the discriminator classifies the generated image based on its authenticity (whether it meets the conditions of the association matrix). During training, the real design element image is marked as 1, and the image generated by the generator is marked as 0. The discriminator continuously adjusts its parameters to improve the ability to distinguish images that meet the association conditions. The goal of the generator is to generate images that can "deceive" the discriminator, that is, to generate design element images that meet the conditions of the association matrix. The generator updates its parameters based on the feedback of the discriminator so that the generated results gradually approach the feature combination set in the association matrix. The training parameters of the discriminator network are set as follows: learning rate: both the discriminator and the generator are set to 0.0002, β1 = 0.5; batch size: 64; number of iterations: 20,000 times, until the average discrimination probability of the generator output image stabilizes above 0.85, indicating that the generated image meets the conditions set by the association matrix well. During the training process, the difference between the image generated by the generator and the real image is continuously evaluated. If the accuracy of the discriminator continues to improve, and the generated image gradually meets the conditions in the association matrix (such as the improvement of the matching degree of color and shape), the optimization training is considered to be successful. After the discriminant optimization training, a personalized digital media design element optimization generation model is generated.The model can generate design element images that conform to specific styles and associations based on the user's personalized preferences and the associations between design elements, thereby improving the personalization and diversity of the generated images.

[0129] This specification provides a digital media design element optimization system of AIGC technology, which is used to execute the digital media design element optimization method of AIGC technology as described above. The digital media design element optimization system of AIGC technology includes:

[0130] The digital media multi-source data acquisition module is used to collect and process digital media multi-source data using a preset digital media API call interface to generate digital media multi-source data;

[0131] The digital media design element analysis module is used to perform regional fitting digital media visual analysis based on digital media multi-source data to generate regional fitting digital media visual data; perform digital media design element analysis based on regional fitting digital media visual data to generate digital media design element data;

[0132] A digital media design element association analysis module is used to design an association matrix of each digital media design element type based on the digital media design element data, and generate a digital media design element association matrix;

[0133] The personalized digital media design preference feature analysis module is used to obtain user digital media design data; perform personalized digital media design preference feature analysis based on the user digital media design data to generate personalized digital media design preference feature data;

[0134] A design element optimization generation model construction module is used to establish a digital media design element optimization generation model of user preference based on a preset adversarial generative network model and personalized digital media design preference feature data, and generate an initial digital media design element optimization generation model; transmit the digital media design element association matrix to the initial digital media design element optimization generation model to perform personalized optimization generation training processing on the digital media design elements, and generate a personalized digital media design element optimization generation model;

[0135] The user digital media design element optimization generation module is used to obtain real-time user digital media design element data to be optimized; transmit the real-time user digital media design element data to be optimized to the personalized digital media design element optimization generation model for real-time user digital media design element optimization generation processing to generate real-time user digital media design element optimization data; and transmit the real-time user digital media design element optimization data to the terminal to execute the user digital media design element intelligent optimization operation.

[0136] The beneficial effect of the present application is that, by utilizing the preset digital media API call interface, multi-source data is collected and standardized conversion processing of heterogeneous data is performed, so that the generated digital media design elements can integrate the characteristics of multi-source data, avoid the problem of single data source in the prior art, and more comprehensively understand different digital media design elements, provide more diversified basic data for subsequent optimization generation, and improve the diversity and applicability of the optimization generation of digital media design elements. By analyzing the visual features of digital media to obtain digital media design element data, and analyzing the associated features of digital media design elements, in-depth mining of the associated features between each design element is achieved, and the understanding of different types of design elements is improved, ensuring that the optimized and generated design elements are more in line with user needs in terms of visual effects and overall relevance. By analyzing the user's digital media design content and interactive behavior, it is used to guide the optimized generation of design elements, ensuring that the generated content is more in line with the user's specific preferences, and based on the optimization generation model of the adversarial generation network, the personalized preference characteristics of the user and the design element associated data are fully utilized, and the intelligent and personalized optimization generation of digital media design elements is realized, the diversity of generated content and the matching degree of user preferences are enhanced, and the efficiency and quality of optimized generation are improved.

[0137] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0138] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A digital media design element optimization method based on AIGC technology, characterized in that: The following steps are involved: Step S1: using a preset digital media API calling interface to collect and process digital media multi-source data to generate digital media multi-source data; Step S2: performing regional fitting digital media visual analysis based on digital media multi-source data to generate regional fitting digital media visual data; performing digital media design element analysis based on the regional fitting digital media visual data to generate digital media design element data; Step S3: Design an association matrix of each type of digital media design element based on the digital media design element data to generate a digital media design element association matrix; Step S4: Obtain user digital media design data; Conduct personalized digital media design preference feature analysis based on user digital media design data to generate personalized digital media design preference feature data; Step S5: Establishing a digital media design element optimization generation model of user preference based on a preset generative adversarial network model and personalized digital media design preference feature data, and generating an initial digital media design element optimization generation model; The digital media design element association matrix is ​​transmitted to the initial digital media design element optimization generation model to perform personalized optimization generation training processing on the digital media design elements, thereby generating a personalized digital media design element optimization generation model; Step S6: Obtaining real-time user digital media design element data to be optimized; The real-time user digital media design element to-be-optimized data is transmitted to the personalized digital media design element optimization generation model for real-time user digital media design element optimization generation processing to generate real-time user digital media design element optimization data; According to the real-time user digital media design element optimization data is transmitted to the terminal to perform the user digital media design element intelligent optimization operation.

2. The digital media design element optimization method of AIGC technology according to claim 1, characterized in that: Step S1 includes the following steps: Step S11: using a preset digital media API calling interface to perform initial digital media multi-source data acquisition and processing to generate initial digital media multi-source data; Step S12: performing digital media information heterogeneous source type analysis based on the initial digital media multi-source data to generate digital media information heterogeneous source type data; Step S13: performing digital media information heterogeneous source structure analysis according to the digital media information heterogeneous source type data to generate digital media information heterogeneous source structure data; Step S14: Design a digital media multi-source heterogeneous conversion decision based on the digital media information heterogeneous source structure data, and perform heterogeneous data standardization conversion processing on the initial digital media multi-source data based on the digital media multi-source heterogeneous conversion decision to generate digital media multi-source data.

3. The digital media design element optimization method of AIGC technology according to claim 1, characterized in that: Step S2 includes the following steps: Step S21: extracting digital media local grid visual data from digital media multi-source data to generate digital media local grid visual data; Step S22: performing regional fitting processing of adjacent local grid visuals according to the digital media local grid visual data to generate regional fitting digital media visual data; Step S23: performing digital media design element analysis on the digital media multi-source data according to the regional fitting digital media visual data to generate digital media design element data.

4. The digital media design element optimization method of AIGC technology according to claim 3, characterized in that: Step S22 includes the following steps: Perform local grid visual descriptor analysis based on digital media local grid visual data to generate local grid descriptor data; Performing digital media local grid visual similarity analysis on digital media local grid visual data based on local grid descriptor data to generate digital media local grid visual similarity data; Based on the digital media local grid visual similarity data, the digital media local grid visual data is subjected to regional fitting processing of adjacent local grid visuals to generate regional fitting digital media visual data.

5. The digital media design element optimization method of AIGC technology according to claim 1, characterized in that: Step S3 includes the following steps: Step S31: performing digital media visual region abstract feature analysis based on the region fitting digital media visual data to generate digital media visual region abstract feature data; Step S32: using the digital media visual region abstract feature data as cluster labels to perform digital media visual region cluster analysis on the regional fitting digital media visual data to generate clustered digital media visual region data; Step S33: performing digital media design element correlation feature analysis based on the clustered digital media visual area data to generate digital media design element correlation feature data; Step S34: Design an association matrix for each type of digital media design element according to the digital media design element association feature data and the digital media design element data to generate a digital media design element association matrix.

6. The digital media design element optimization method of AIGC technology according to claim 1, characterized in that: Step S4 includes the following steps: Step S41: Obtaining user digital media design data; Step S42: performing user digital media design content extraction processing according to the user digital media design data to generate user digital media design content data; Step S43: using a preset convolutional neural network algorithm to perform digital media design content style analysis on the user digital media design content data to generate user digital media design content style data; Step S44: extracting user digital media design interaction behaviors according to the user digital media design data to generate user digital media design interaction behavior data; Step S45: performing a quantitative analysis of the user's digital media design interaction behavior preference based on the user's digital media design interaction behavior data to generate the user's digital media design interaction behavior preference quantitative data; Step S46: Map the user digital media design interaction behavior preference quantified data to the user digital media design content style data to perform personalized digital media design preference feature analysis to generate personalized digital media design preference feature data.

7. The digital media design element optimization method of AIGC technology according to claim 1, characterized in that: Step S5 includes the following steps: Step S51: Establishing a generator network for optimizing digital media design elements based on the fully connected layer and the deconvolution layer of the preset adversarial generative network model, and generating a generator network for optimizing digital media design elements; Step S52: establishing a user preference discriminator network for optimizing digital media design elements based on personalized digital media design preference feature data, and generating a digital media design element optimization discriminator network; Step S53: constructing an adversarial generation mapping relationship between digital media design elements and user-preferred digital media design elements based on the digital media design element optimization generator network and the digital media design element optimization discriminator network, and generating an initial digital media design element optimization generation model; Step S54: transmitting the digital media design element association matrix to the initial digital media design element optimization generation model to perform personalized optimization generation training processing of the digital media design elements, and generating a personalized digital media design element optimization generation model.

8. The digital media design element optimization method of AIGC technology according to claim 7, characterized in that: Step S52 includes the following steps: Perform optimization condition feature analysis of digital media design elements based on personalized digital media design preference feature data to generate optimization condition feature data of personalized digital media design elements; Perform personalized digital media design element optimization discrimination decision design based on personalized digital media design element optimization condition characteristic data, and generate personalized digital media design element optimization discrimination decision; A discriminator network for optimizing digital media design elements is established according to personalized digital media design element optimization discrimination decisions, and a digital media design element optimization discriminator network is generated.

9. The digital media design element optimization method of AIGC technology according to claim 7, characterized in that: Step S54 includes the following steps: The digital media design element association data of each category in the digital media design element association matrix is ​​transmitted to the digital media design element optimization generator network of the initial digital media design element optimization generation model to perform the digital media design element generation training tasks of each category, and the digital media design element optimization discriminator network of the initial digital media design element optimization generation model is used to perform the discriminant optimization training processing of digital media design element generation on the digital media design element generation training tasks of each category to obtain a personalized digital media design element optimization generation model.

10. A digital media design element optimization system based on AIGC technology, characterized in that: For executing the digital media design element optimization method of the AIGC technology according to claim 1, the digital media design element optimization system of the AIGC technology comprises: The digital media multi-source data acquisition module is used to collect and process digital media multi-source data using a preset digital media API call interface to generate digital media multi-source data; The digital media design element analysis module is used to perform regional fitting digital media visual analysis based on digital media multi-source data to generate regional fitting digital media visual data; perform digital media design element analysis based on regional fitting digital media visual data to generate digital media design element data; A digital media design element association analysis module is used to design an association matrix of each digital media design element type based on the digital media design element data, and generate a digital media design element association matrix; The personalized digital media design preference feature analysis module is used to obtain user digital media design data; perform personalized digital media design preference feature analysis based on the user digital media design data to generate personalized digital media design preference feature data; A design element optimization generation model construction module is used to establish a digital media design element optimization generation model of user preference based on a preset adversarial generative network model and personalized digital media design preference feature data, and generate an initial digital media design element optimization generation model; transmit the digital media design element association matrix to the initial digital media design element optimization generation model to perform personalized optimization generation training processing on the digital media design elements, and generate a personalized digital media design element optimization generation model; The user digital media design element optimization generation module is used to obtain real-time user digital media design element data to be optimized; transmit the real-time user digital media design element data to be optimized to the personalized digital media design element optimization generation model for real-time user digital media design element optimization generation processing to generate real-time user digital media design element optimization data; and transmit the real-time user digital media design element optimization data to the terminal to execute the user digital media design element intelligent optimization operation.

Images