Intelligent service system and method for customizing cultural and creative products

Abstract

The invention discloses an intelligent service system and method for customizing a cultural and creative product, and relates to the field of intelligent customization, and the method comprises the steps: extracting a product prototype image from a product prototype library as an initial material, and carrying out the semantic analysis of the prototype image of the cultural and creative product through employing an image processing technology based on deep learning, the method comprises the following steps: capturing key design elements in an image, and meanwhile, carrying out fine-grained cross-modal semantic interaction on a product customization demand and a product prototype image by combining the customization demand of a cultural and creative product input by a user, so as to realize customization feature modulation on the product prototype image according to the product customization demand; and on the basis, the customized effect image of the cultural and creative product is intelligently generated, so that the personalized customization requirement of the user can be quickly responded, the design period is shortened, the customization efficiency is improved, and the customization cost is reduced, thereby effectively improving the customization experience of the user.

Description

Technical Field

[0001] The present application relates to the field of intelligent customization, and more specifically, to an intelligent service system and method for customizing cultural and creative products. Background Art

[0002] With the rapid development of the cultural and creative industries, the market demand for cultural and creative products, as an important carrier of cultural expression and dissemination, is growing. Consumers are also increasingly demanding personalized customization of cultural and creative products. They are no longer satisfied with stereotyped goods, but prefer products that reflect personal style and unique taste. This trend has promoted the expansion of the cultural and creative product customization service market and technological innovation.

[0003] In the field of cultural and creative product customization, the traditional product customization model usually requires users and designers to conduct multiple in-depth communications, and designers make corresponding adjustments based on the user's explanation and needs. However, in the process of interaction between users and designers, due to the lack of efficient and practical tools and technical support, there may be a deviation in understanding between the two parties, which will cause the final product to be inconsistent with the user's expectations. In addition, the design cycle of this customization model is relatively long, and it is difficult to achieve rapid iteration, and it is impossible to respond efficiently to the user's immediate needs.

[0004] Therefore, an intelligent service system and method for customized cultural and creative products is needed to solve the above technical problems. Summary of the invention

[0005] In order to solve the above technical problems, this application is proposed.

[0006] According to one aspect of the present application, a method for intelligent service of customizing cultural and creative products is provided, which includes:

[0007] Extracting product prototype images from a product prototype library, and obtaining cultural and creative product customization requirements input by a user, wherein the cultural and creative product customization requirements include aesthetic preferences, cultural background, and descriptions of requirements for specific occasions;

[0008] Extracting image semantic features from the product prototype image to obtain a fine-grained semantic encoding feature map of the product prototype image;

[0009] Standardizing and semantically encoding the cultural and creative product customization requirements to obtain a semantically encoded feature vector of the cultural and creative product customization requirements;

[0010] Based on the cultural and creative product customization demand semantic coding feature vector, the product prototype image fine-grained semantic interactive optimization modulation is performed on the product prototype image fine-grained semantic coding feature map to obtain a customization demand modulation product image feature inversion feature map;

[0011] Based on the customization demand, the product image feature inversion feature map is modulated to generate a customization effect to obtain a customized effect image of the cultural and creative product.

[0012] Specifically, performing image semantic feature extraction on the product prototype image to obtain a fine-grained semantic encoding feature map of the product prototype image includes:

[0013] The product prototype image is input into a product prototype image feature extractor based on the ViT model to obtain a fine-grained semantic encoding feature map of the product prototype image.

[0014] Specifically, the cultural and creative product customization requirements are standardizedly expressed and semantically encoded to obtain a cultural and creative product customization requirement semantic encoding feature vector, including:

[0015] Inputting the cultural and creative product customization requirements into a customization requirement specification expression module based on a large language model to obtain a cultural and creative product customization requirement specification expression;

[0016] The cultural and creative product customization requirement specification expression is semantically encoded to obtain the cultural and creative product customization requirement semantic encoding feature vector.

[0017] Specifically, based on the cultural and creative product customization demand semantic coding feature vector, the product prototype image fine-grained semantic coding feature map is subjected to fine-grained semantic interaction optimization modulation to obtain a customization demand modulation product image feature inversion feature map, including:

[0018] Performing autocorrelation coding on the cultural and creative product customization demand semantic coding feature vector to obtain a cultural and creative product customization demand autocorrelation coding matrix;

[0019] Performing local fine-grained semantic interaction coding on the cultural and creative product customization demand autocorrelation coding matrix and the product prototype image fine-grained semantic coding feature map to obtain a set of customization demand-product prototype image cross-modal prompt information semantic coding matrices;

[0020] Based on the set of the customization demand-product prototype image cross-modal prompt information semantic coding matrices, the cultural and creative product customization demand autocorrelation coding matrix and the product prototype image fine-grained semantic coding feature map are guided to perform cross-modal interactive optimization to obtain the customization demand modulated product image feature inversion feature map.

[0021] Specifically, the cultural and creative product customization demand semantic coding feature vector is autocorrelatedly coded to obtain a cultural and creative product customization demand autocorrelation coding matrix, including:

[0022] The product of the cultural and creative product customization demand semantic coding feature vector and the transposed vector of the cultural and creative product customization demand semantic coding feature vector is calculated to obtain the cultural and creative product customization demand autocorrelation coding matrix.

[0023] Specifically, the cultural and creative product customization demand autocorrelation coding matrix and the product prototype image fine-grained semantic coding feature map are locally fine-grained semantically interactively coded to obtain a set of customization demand-product prototype image cross-modal prompt information semantic coding matrices, including:

[0024] Performing feature decoupling on the fine-grained semantic encoding feature map of the product prototype image to obtain a set of local semantic feature matrices of the product prototype image;

[0025] Performing a linear transformation on the cultural and creative product customization demand autocorrelation coding matrix to obtain a cultural and creative product customization demand query coding matrix and a cultural and creative product customization demand value coding matrix;

[0026] The local semantic feature matrices of each product prototype image in the set of the cultural and creative product customization demand query coding matrix, the cultural and creative product customization demand value coding matrix and the product prototype image local semantic feature matrix are respectively input into the cross-modal prompt information encoder based on the converter structure to obtain the set of the customization demand-product prototype image cross-modal prompt information semantic coding matrices.

[0027] Specifically, based on the set of the semantic coding matrix of the customization demand-product prototype image cross-modal prompt information, the cultural and creative product customization demand semantic coding feature vector and the product prototype image fine-grained semantic coding feature map are guided to perform cross-modal interactive optimization to obtain the customization demand modulated product image feature inversion feature map, including:

[0028] Inputting the set of the customization demand-product prototype image cross-modal prompt information semantic encoding matrices into a decoder-based information gating unit to obtain a set of customization demand-product prototype image cross-modal semantic interaction attention weights;

[0029] Inputting the set of the cultural and creative product customization demand autocorrelation coding matrix and the product prototype image local semantic feature matrix into a cross-modal interaction unit to obtain a set of customization demand-product prototype image cross-modal interaction local feature matrices;

[0030] The set of the customization demand-product prototype image cross-modal semantic interaction attention weights and the set of the customization demand-product prototype image cross-modal interaction local feature matrices are input into the cross-modal interaction optimization unit to obtain the customization demand modulated product image feature inversion feature map.

[0031] Specifically, based on the customization requirement, the product image feature is modulated to invert the feature map to generate a customization effect to obtain a customized effect image of a cultural and creative product, including:

[0032] The customized demand modulated product image feature inversion feature map is input into a customized effect generator based on a diffusion model to obtain the customized effect image of the cultural and creative product.

[0033] According to another aspect of the present application, there is provided a cultural and creative product customization intelligent service system, which includes:

[0034] A product image customization demand acquisition module is used to extract product prototype images from the product prototype library and simultaneously acquire cultural and creative product customization demands input by users, wherein the cultural and creative product customization demands include aesthetic preferences, cultural backgrounds, and descriptions of demands for specific occasions;

[0035] A product prototype image semantic feature extraction module, used to extract image semantic features from the product prototype image to obtain a fine-grained semantic encoding feature map of the product prototype image;

[0036] A customization requirement semantic coding module, used for standardizing and semantically coding the customization requirement of the cultural and creative product to obtain a semantic coding feature vector of the customization requirement of the cultural and creative product;

[0037] A customization demand modulation module is used to perform fine-grained semantic interaction optimization modulation on the fine-grained semantic coding feature map of the product prototype image based on the semantic coding feature vector of the customization demand of the cultural and creative product to obtain a customization demand modulation product image feature inversion feature map;

[0038] The customized effect generation module is used to modulate the product image feature inversion feature map based on the customized demand to generate customized effects to obtain a customized effect image of the cultural and creative product.

[0039] This application has at least the following technical effects:

[0040] Compared with the prior art, the present application provides an intelligent service system and method for customizing cultural and creative products. The system extracts product prototype images from a product prototype library as initial materials, and uses deep learning-based image processing technology to perform semantic analysis on the prototype images of cultural and creative products, thereby capturing key design elements in the images. At the same time, the system combines the customization needs of cultural and creative products input by users, performs fine-grained cross-modal semantic interaction between product customization needs and product prototype images, thereby realizing customized feature modulation of product prototype images according to product customization needs, and intelligently generates customized effect images of cultural and creative products based on this. The system can quickly respond to users' personalized customization needs, shorten the design cycle, improve customization efficiency, and reduce customization costs, thereby effectively improving users' customization experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0042] Figure 1 This is a flow chart of an intelligent service method for customizing cultural and creative products according to an embodiment of the present application.

[0043] Figure 2 The present invention is a flowchart for standardizing and semantically encoding the cultural and creative product customization requirements in the intelligent service method for cultural and creative product customization according to an embodiment of the present application to obtain a semantically encoded feature vector of the cultural and creative product customization requirements.

[0044] Figure 3 This is a flowchart of performing fine-grained semantic interactive optimization modulation on the fine-grained semantic coding feature map of the product prototype image based on the semantic coding feature vector of the cultural and creative product customization demand in the intelligent service method for cultural and creative product customization according to the embodiment of the present application to obtain a customization demand modulated product image feature inversion feature map.

[0045] Figure 4 This is a block diagram of an intelligent service system for customizing cultural and creative products according to an embodiment of the present application. DETAILED DESCRIPTION

[0046] The embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments described herein, but rather these embodiments are provided for a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0047] It should be understood that the various steps described in the method embodiments of the present disclosure may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present disclosure is not limited in this respect.

[0048] In the description of the embodiments of the present disclosure, the term "including" and similar terms should be understood as open inclusion, that is, "including but not limited to". The term "based on" should be understood as "based at least in part on". The term "one embodiment" or "the embodiment" should be understood as "at least one embodiment". The terms "first", "second", etc. may refer to different or the same objects. Other explicit and implicit definitions may also be included below.

[0049] It should be noted that the modifications of "one" and "plurality" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0050] Specifically, in this application, all actions to obtain signals, information or data are carried out in compliance with the relevant data protection laws and policies of the country where they are located, and with the authorization given by the owner of the corresponding device.

[0051] With the rapid development of the cultural and creative industries, the market demand for cultural and creative products, as an important carrier of cultural expression and dissemination, is growing. Consumers are also increasingly demanding personalized customization of cultural and creative products. They are no longer satisfied with stereotyped goods, but prefer products that reflect personal style and unique taste. This trend has promoted the expansion of the cultural and creative product customization service market and technological innovation.

[0052] In the field of cultural and creative product customization, the traditional product customization model usually requires users and designers to conduct multiple in-depth communications, and designers make corresponding adjustments based on the user's explanation and needs. However, in the process of interaction between users and designers, due to the lack of efficient and practical tools and technical support, there may be a deviation in understanding between the two parties, which will cause the final product to be inconsistent with the user's expectations. In addition, the design cycle of this customization model is relatively long, and it is difficult to achieve rapid iteration, and it is impossible to respond efficiently to the user's immediate needs.

[0053] Therefore, in response to the above technical problems, the technical concept of the present application is: extracting product prototype images from the product prototype library as initial materials, and using deep learning-based image processing technology to perform semantic analysis on the prototype images of cultural and creative products to capture the key design elements in the images. At the same time, combined with the cultural and creative product customization needs input by users, fine-grained cross-modal semantic interaction is performed between product customization needs and product prototype images, thereby realizing customized feature modulation of product prototype images according to product customization needs, and intelligently generating customized effect images of cultural and creative products based on this, which can quickly respond to users' personalized customization needs, shorten the design cycle, improve customization efficiency, and reduce customization costs, thereby effectively improving users' customization experience.

[0054] Figure 1 Flow chart of the intelligent service method for customizing cultural and creative products according to the embodiment of the present application. Figure 1As shown, according to the embodiment of the present application, the intelligent service method for customization of cultural and creative products includes: S110, extracting product prototype images from a product prototype library, and obtaining cultural and creative product customization requirements input by users, wherein the cultural and creative product customization requirements include aesthetic preferences, cultural backgrounds, and descriptions of requirements for specific occasions; S120, performing image semantic feature extraction on the product prototype images to obtain a fine-grained semantic coding feature map of the product prototype images; S130, performing standardized expression and semantic coding on the cultural and creative product customization requirements to obtain a semantic coding feature vector of the cultural and creative product customization requirements; S140, based on the semantic coding feature vector of the cultural and creative product customization requirements, performing fine-grained semantic interactive optimization modulation on the fine-grained semantic coding feature map of the product prototype image to obtain a feature inversion feature map of the product image modulated by the customization requirements; S150, performing customization effect generation based on the feature inversion feature map of the product image modulated by the customization requirements to obtain a customized effect image of the cultural and creative product.

[0055] In step S110, the product prototype image is extracted from the product prototype library, and the customization requirements of the cultural and creative products input by the user are obtained at the same time, and the customization requirements of the cultural and creative products include aesthetic preferences, cultural background and descriptions of specific occasion requirements. Among them, the user's aesthetic preferences, cultural background and specific occasion requirements are key factors that constitute the personalized characteristics of cultural and creative products. For example, the user may expect that specific cultural elements can be incorporated into the product design, or conform to a certain aesthetic style, or a special style designed for a specific occasion (such as a festival, celebration, etc.). By obtaining the customization requirements input by the user, it is helpful to accurately understand the user's expectations, thereby generating customized cultural and creative products that better meet the user's personalized needs. In addition, as the basis of creativity and design, the product prototype library provides a series of pre-designed basic templates or examples. By selecting the prototype closest to the customer's needs from the library as the design starting point, it can respond to the user's customization needs more quickly, thereby saving time and resources for creating designs from scratch.

[0056] Specifically, first, the system maintains a product prototype library containing various basic design elements and styles. These prototypes can be completed design templates, basic shapes or color schemes, etc., aiming to provide users with a diverse selection basis. When users access the customization service, they can browse the library to select the product that is closest to their imagination as a starting point. This step not only helps users quickly locate the design direction that meets their general ideas, but also provides a good start for subsequent more refined adjustments.

[0057] Next, after the user has selected a satisfactory product prototype, the system will guide them to a detailed customization requirement input interface. Among them, users are encouraged to describe in detail their aesthetic preferences, cultural background, and the needs of any specific occasion. For example, if you are preparing a gift to celebrate a festival or anniversary, the user may emphasize that the finished product should reflect the relevant theme; or if you want a product that represents your personal style and taste, then specific requirements on color matching, pattern selection, and even material texture are particularly important. By providing as much information as possible, we can better understand the user's expectations and make targeted design adjustments accordingly.

[0058] In order to ensure that the information collection process is both efficient and comprehensive, the system usually uses a structured approach to guide users to fill in relevant information. For example, a series of preset questions are set to allow users to choose appropriate options based on their own circumstances, while leaving open text boxes for additional instructions. This can not only ensure that the collected data has a certain degree of standardization, which is convenient for subsequent processing and analysis, but also fully respect the unique ideas of each customer. In addition, considering the possible differences in expression ability among different users, the system may also introduce natural language processing technology to assist in parsing unstructured text content, further improving the accuracy of understanding user intentions. In short, by effectively utilizing the product prototype library and accurately capturing user customization needs, it is possible to quickly respond to market changes while maintaining creative diversity and meet the growing personalized consumer needs.

[0059] In step S120, image semantic features are extracted from the product prototype image to obtain a fine-grained semantic coding feature map of the product prototype image. Specifically, in an embodiment of the present application, image semantic features are extracted from the product prototype image to obtain a fine-grained semantic coding feature map of the product prototype image, including: inputting the product prototype image into a product prototype image feature extractor based on the ViT model to obtain the fine-grained semantic coding feature map of the product prototype image. Accordingly, it is considered that the design of cultural and creative products involves not only tiny visual elements, but also the overall design style and product structure. Therefore, in order to improve the ability to understand the semantics of product prototype images, the present application adopts the ViT model to construct a product prototype image feature extractor, and performs fine-grained image semantic analysis on the product prototype image to obtain a fine-grained semantic coding feature map of the product prototype image. Those skilled in the art should be aware that, compared with traditional convolutional neural networks, the ViT model has significant advantages in processing long-distance dependencies and global context information. It divides the product prototype image into multiple small patches, linearly embeds and positionally encodes each small patch, and then uses a self-attention mechanism to capture the semantic dependencies between image patches. This can effectively capture the global information and detail features in the product prototype image, extract key design element features and deep semantic information in the product prototype image, and thus provide a solid feature foundation for subsequent customized feature modulation.

[0060] Specifically, the Vision Transformer (ViT) is a model based on the Transformer architecture, which directly inputs the image as a series of patches into the Transformer for processing. The introduction of the ViT model marks the beginning of the successful application of the Transformer architecture in the field of natural language processing to computer vision tasks, and has achieved or even surpassed the performance of traditional convolutional neural networks (CNNs) on multiple benchmarks. The following is a specific description of the processing process of inputting the product prototype image into the product prototype image feature extractor based on the ViT model:

[0061] In the ViT model, the product prototype image needs to be preprocessed first. Assuming there is an RGB color image of size 224x224 pixels as input, in order to convert it into a form suitable for Transformer processing, the image is usually divided into small blocks of fixed size or patches. For example, 16x16 pixels can be selected as the size of a patch, so that the entire image is divided into 14x14=196 such small blocks. Each patch is then flattened into a vector. If the original image is in RGB format, each patch can be flattened into a one-dimensional vector of length 16*16*3=768.

[0062] Next, all these flattened patch vectors are concatenated to form a sequence, which is actually the input of the Transformer model. Specifically, a special classification token [CLS] is added to the front of the sequence, which is similar to the [CLS] tag in NLP tasks and is used to summarize the information of the entire image. In addition, in order to provide position information to the model, each patch vector and [CLS] token is added with a learnable position encoding. The position encoding is obtained through training, and the purpose is to enable the model to understand the relative position relationship between different patches, which is very important for processing spatially structured data such as images.

[0063] After completing the above preparations, a tensor with a shape of (197,768) is obtained, where 197 represents the total number of patches including the [CLS] token, and 768 is the dimension of each patch vector. Next, this tensor will be fed into a multi-layer Transformer encoder. Each Transformer encoder layer consists of two sublayers: the first sublayer is a multi-head self-attention mechanism used to capture long-distance dependencies between different patches; the second sublayer is a feedforward network for further processing and enhancing feature expression capabilities. Residual connections and layer normalization are applied between these two sublayers and after the entire encoder layer to ensure stable gradient propagation and accelerate the convergence process.

[0064] A typical ViT model may contain multiple such Transformer encoder layers stacked together, and the specific number of layers depends on the needs of the actual application scenario. After all encoder layers are processed, the tensor of shape (196,768) is reshaped into a form that is closer to the spatial structure of the original image. For example, the original image is 224x224 pixels and is divided into 14x14 patches. Therefore, each 768-dimensional feature vector can be regarded as a 14x14x768 three-dimensional tensor, which is equivalent to rearranging the features of each patch into a two-dimensional grid to restore some of the spatial structure information of the original image.

[0065] In summary, the ViT model effectively achieves the goal of extracting rich semantic information from image data by decomposing the image into a series of ordered patches and leveraging the powerful Transformer architecture to handle the complex relationships between these patches.

[0066] In step S130, the cultural and creative product customization requirements are standardizedly expressed and semantically encoded to obtain a semantically encoded feature vector of the cultural and creative product customization requirements. Figure 2 The flowchart is a method for intelligent service of customizing cultural and creative products according to an embodiment of the present application, in which the cultural and creative product customization requirements are standardizedly expressed and semantically encoded to obtain a semantically encoded feature vector of the cultural and creative product customization requirements. Specifically, in the embodiment of the present application, Figure 2 As shown, the customization requirements of cultural and creative products are expressed in a standardized manner and semantically encoded to obtain a semantically encoded feature vector of the customization requirements of cultural and creative products, including: S210, inputting the customization requirements of cultural and creative products into a customization requirement standardized expression module based on a large language model to obtain a standardized expression of the customization requirements of cultural and creative products; S220, semantically encoding the standardized expression of the customization requirements of cultural and creative products to obtain a semantically encoded feature vector of the customization requirements of cultural and creative products.

[0067] Specifically, in step 210, the customization requirements of cultural and creative products are input into the customization requirements specification expression module based on the large language model to obtain the customization requirements specification expression of cultural and creative products. Among them, it is considered that the customization requirements of cultural and creative products input by the user may be unstructured, vague, or use colloquial or personalized language descriptions. Therefore, in order to reduce ambiguity and ensure the accurate expression of customization requirements, the present application further uses the customization requirements specification expression module based on the large language model to standardize the customization requirements of cultural and creative products, and converts them into a standard, clear, easy-to-understand and easy-to-execute language form. In a specific example of the present application, the large language model is a GPT-3 (Generative Pre-trained Transformer 3) model, which has a strong language understanding and generation capability through large-scale pre-training, can fully understand the semantic meaning of the customization requirements of cultural and creative products, and convert them into clear and standardized customization requirements descriptions, generate cultural and creative products Customization requirements specification expression, so as to ensure the accuracy of the subsequent cultural and creative product customization process.

[0068] Specifically, in step S220, the cultural and creative product customization requirement specification expression is semantically encoded to obtain the cultural and creative product customization requirement semantic encoding feature vector. Accordingly, in order to convert the textual product customization requirement description into a form understandable by the machine learning model, the present application further uses natural language processing technology to semantically encode the cultural and creative product customization requirement specification expression, map it to a high-dimensional semantic feature space, and deeply understand its contextual meaning, thereby generating a cultural and creative product customization requirement semantic encoding feature vector, so that the cultural and creative product customization requirement specification expression can effectively interact with the product prototype image across modal semantics. In a specific example of the present application, the BERT model is used to perform the above semantic encoding process.

[0069] Among them, in the process of customizing cultural and creative products, using the BERT model to semantically encode the cultural and creative product customization requirements input by users is one of the key steps to achieve efficient and accurate understanding of user intentions. The BERT model is known for its powerful two-way context understanding ability and has a wide range of applications in the field of natural language processing, including text classification, sentiment analysis, question-answering systems, etc. When applied to the customization of cultural and creative products, BERT can help the system deeply analyze the user's text description and convert it into a form that is easy for computers to process and rich in semantic information - that is, the semantic encoding feature vector of the cultural and creative product customization requirements. This process can not only improve the efficiency and quality of the subsequent design stage, but also ensure that the final product is closer to the real needs of users.

[0070] Specifically, the process of using the BERT model to semantically encode the standardized expression of the customization requirements of the cultural and creative products is as follows:

[0071] First, users submit customization requirements for cultural and creative products through a specific interface. These requirements usually include aesthetic preferences, cultural background, and descriptions of needs for specific occasions. In order for the BERT model to work effectively, the original text data needs to be preprocessed. This step mainly includes tokenization, adding special tags (such as [CLS] and [SEP]), and converting the text into a format acceptable to the model. Among them, tokenization refers to splitting a sentence into individual words or sub-word units, and adding special tags is to let the model know the start and end positions of a sequence. In addition, since BERT has a fixed input length limit, if the text is too long, it may need to be truncated or split.

[0072] After the above preprocessing, the text data is fed into the pre-trained BERT model. Here, BERT uses its multi-layer Transformer encoder structure to capture the contextual meaning of each word in the entire sentence. Specifically, each layer of the encoder contains two main parts: a self-attention mechanism and a feedforward neural network. The self-attention mechanism allows the model to simultaneously pay attention to information at different positions in the sentence and dynamically adjust the weight distribution according to the relationship between them; while the feedforward network further enhances the ability of feature representation. The entire encoding process is bidirectional, which means that the model can learn the meaning of the text from both left to right and right to left, thereby gaining a more comprehensive understanding.

[0073] As data flows through each layer of encoder, each word gradually forms a high-dimensional vector representation, which comprehensively reflects all relevant information about the word and its context. For the entire sentence, the vector corresponding to the [CLS] token at the beginning of the sequence is particularly important, because it brings together the main meaning of the entire sentence and is often used as input for downstream tasks. However, in the context of customized cultural and creative products, more attention is paid to how to transform the specific needs of users into a series of actionable design guidelines. Therefore, in addition to [CLS], the semantic encoding results of other key words or phrases need to be considered.

[0074] To achieve this, a common practice is to extract the hidden states of all words from the last layer of BERT output as preliminary semantic encoding features. Then, based on the actual application requirements, different strategies can be used to further refine these features. For example, the vectors of related words can be aggregated by weighted averaging, maximum pooling, etc. to generate a single feature vector representing the entire customization requirement. Another method is to directly select those key words related to the design (such as color, pattern style, etc.) and use their respective vectors to combine into the final feature representation. In this way, the system can quickly and accurately understand the needs of users and provide them with a more personalized service experience.

[0075] In step S140, based on the cultural and creative product customization demand semantic coding feature vector, the product prototype image fine-grained semantic coding feature map is fine-grained semantic interaction optimization modulated to obtain the customization demand modulated product image feature inversion feature map. Furthermore, in order to effectively use the semantic information of cultural and creative product customization demand to guide the customized feature modulation of the product prototype image, the present application proposes a cross-modal semantic interaction method based on prompt information, which uses the cultural and creative product customization demand semantic coding feature vector as prompt information to perform fine-grained semantic interaction optimization modulation on the product prototype image fine-grained semantic coding feature map, thereby achieving deep cross-modal semantic fusion, so that the customized modulated product image features can fully reflect the user's personalized needs.

[0076] Figure 3 The present invention is a flow chart of performing fine-grained semantic interaction optimization modulation on the fine-grained semantic coding feature map of the product prototype image based on the semantic coding feature vector of the cultural and creative product customization demand in the cultural and creative product customization intelligent service method according to the embodiment of the present application to obtain the customization demand modulation product image feature inversion feature map. Specifically, in the embodiment of the present application, Figure 3As shown, based on the cultural and creative product customization demand semantic coding feature vector, the product prototype image fine-grained semantic coding feature map is fine-grained semantic interactive optimization modulation to obtain the customization demand modulated product image feature inversion feature map, including: S310, autocorrelation coding the cultural and creative product customization demand semantic coding feature vector to obtain the cultural and creative product customization demand autocorrelation coding matrix; S320, local fine-grained semantic interactive coding is performed on the cultural and creative product customization demand autocorrelation coding matrix and the product prototype image fine-grained semantic coding feature map to obtain a set of customization demand-product prototype image cross-modal prompt information semantic coding matrices; S330, based on the set of customization demand-product prototype image cross-modal prompt information semantic coding matrices, guiding the cultural and creative product customization demand autocorrelation coding matrix and the product prototype image fine-grained semantic coding feature map to perform cross-modal interactive optimization to obtain the customization demand modulated product image feature inversion feature map.

[0077] Specifically, in step S310, the cultural and creative product customization demand semantic coding feature vector is autocorrelatedly coded to obtain the cultural and creative product customization demand autocorrelation coding matrix, including: calculating the product between the cultural and creative product customization demand semantic coding feature vector and the transposed vector of the cultural and creative product customization demand semantic coding feature vector to obtain the cultural and creative product customization demand autocorrelation coding matrix. That is, by calculating the product between the cultural and creative product customization demand semantic coding feature vector and its own transposed vector, the concept of outer product in matrix operation is used to reveal the autocorrelation of cultural and creative product customization demand, and obtain the cultural and creative product customization demand autocorrelation coding matrix.

[0078] The processing process of step S310 can be expressed by the following formula:

[0079]

[0080] Among them, V1 represents the semantic encoding feature vector of the customization demand of cultural and creative products, represents the matrix multiplication operation, (·) T Represents the transpose of a vector, M z Represents the autocorrelation coding matrix of customization demand for cultural and creative products.

[0081] More specifically, in step S320, local fine-grained semantic interaction encoding is performed on the cultural and creative product customization demand autocorrelation coding matrix and the product prototype image fine-grained semantic coding feature map to obtain a set of customization demand-product prototype image cross-modal prompt information semantic coding matrices, including: feature decoupling the product prototype image fine-grained semantic coding feature map to obtain a set of product prototype image local semantic feature matrices; linear transformation is performed on the cultural and creative product customization demand autocorrelation coding matrix to obtain a cultural and creative product customization demand query coding matrix and a cultural and creative product customization demand value coding matrix; each product prototype image local semantic feature matrix in the set of the cultural and creative product customization demand query coding matrix, the cultural and creative product customization demand value coding matrix and the product prototype image local semantic feature matrix is ​​respectively input into a cross-modal prompt information encoder based on a converter structure to obtain the set of customization demand-product prototype image cross-modal prompt information semantic coding matrices.

[0082] That is, by feature decoupling the fine-grained semantic coding feature map of the product prototype image, it is decomposed into multiple local semantic feature matrices of the product prototype image to improve the model's attention to the local details of the product prototype image and enhance the distinguishability of the features, thereby facilitating the execution of a more fine-grained cross-modal interaction analysis. Subsequently, the cultural and creative product customization demand autocorrelation coding matrix is ​​converted into a cultural and creative product customization demand query coding matrix and a cultural and creative product customization demand value coding matrix by means of linear mapping, so that it meets the input form requirements of the attention mechanism. Among them, the cultural and creative product customization demand query coding matrix is ​​used to query the correlation features with the product prototype image, while the cultural and creative product customization demand value coding matrix stores the original cultural and creative product customization demand information to be fused. Furthermore, based on the converter structure, the cultural and creative product customization demand query coding matrix, the cultural and creative product customization demand value coding matrix and the local semantic feature matrix of each product prototype image are cross-modally encoded. The self-attention mechanism of the converter structure is used to realize feature fusion and interactive analysis of cross-modal data, effectively capture the deep semantic connection between the cultural and creative product customization demand and the product prototype image features, generate a set of customization demand-product prototype image cross-modal prompt information semantic coding matrices, and use this as a basis to further guide subsequent feature interaction optimization.

[0083] The processing of step S320 can be expressed by the following formula:

[0084] Decouple(X)={M1,M2,...,M i ,...M n}

[0085]

[0086] Where X represents the fine-grained semantic encoding feature map of the product prototype image, Decouple(·) represents feature decoupling, M1, M2, M i and M n represents the first, second, i-th and n-th local semantic feature matrices of product prototype images in the set of local semantic feature matrices of product prototype images, respectively, n is the number of local semantic feature matrices of product prototype images, W q and W v denote the query embedding matrix and the value embedding matrix respectively, M q and M v They represent the cultural and creative product customization demand query coding matrix and the cultural and creative product customization demand value coding matrix, M i T It is M i The transposed matrix of , S is the characteristic scale value of the local semantic feature matrix of the product prototype image, softmax(·) represents the normalized exponential function, I i Represents the semantic encoding matrix of the cross-modal prompt information of the i-th customization requirement-product prototype image.

[0087] More specifically, in step S330, based on the set of semantic coding matrices of cross-modal prompt information of customization demand-product prototype image, the semantic coding feature vector of customization demand of cultural and creative products and the fine-grained semantic coding feature map of product prototype image are guided to perform cross-modal interaction optimization to obtain the feature inversion feature map of customization demand modulated product image, including: inputting the set of semantic coding matrices of cross-modal prompt information of customization demand-product prototype image into the decoder-based information gating unit to obtain the set of cross-modal semantic interaction attention weights of customization demand-product prototype image; inputting the set of autocorrelation coding matrices of customization demand of cultural and creative products and the set of local semantic feature matrices of product prototype image into the cross-modal interaction unit to obtain the set of cross-modal interaction local feature matrices of customization demand-product prototype image; inputting the set of cross-modal semantic interaction attention weights of customization demand-product prototype image and the set of cross-modal interaction local feature matrices of customization demand-product prototype image into the cross-modal interaction optimization unit to obtain the feature inversion feature map of customization demand modulated product image.

[0088] That is, the set of the semantic coding matrix of the cross-modal prompt information of the customization demand-product prototype image is further input into the information gating unit based on the decoder for information screening. The information gating unit learns the feature importance of each semantic coding matrix of the cross-modal prompt information of the customization demand-product prototype image through the decoder, and performs weight allocation accordingly to generate a set of attention weights of the cross-modal semantic interaction of the customization demand-product prototype image, so that in the subsequent feature interaction process, the important semantic connection between the product prototype image and the customization demand of the cultural and creative product can be effectively highlighted, while suppressing the interference of irrelevant information. Afterwards, the direct interaction between the set of the autocorrelation coding matrix of the customization demand of the cultural and creative product and the local semantic feature matrix of the product prototype image is performed by position point multiplication, so as to capture the correlation information of the two at the same feature space position, and the interaction features between the customization demand of the cultural and creative product and the product prototype image are weighted and aggregated with the attention weights generated in the above process, so as to enhance the expression ability of the features, restore the original image structure, and generate a product image feature representation incorporating the user's personalized demand information, that is, the customization demand modulated product image feature inversion feature map.

[0089] The processing of step S330 can be expressed by the following formula:

[0090] a i =softmax{decoder(I i ,W / )}

[0091] f0=couple{a1·M1⊙M z ,a2·M2⊙M z ,...,a n ·M n ⊙M z}

[0092] Among them, I i represents the semantic encoding matrix of the cross-modal prompt information of the i-th customization requirement-product prototype image, decoder(·) represents the decoder, and W / represents the decoding weight matrix, softmax(·) represents the normalized exponential function, a1, a2, a i and a n represents the cross-modal semantic interaction attention weights of the first, second, i-th and n-th customization requirement-product prototype images, ⊙ represents the point multiplication by position, couple{·} represents feature concatenation, and f0 represents the feature inversion feature map modulated by the customization requirement.

[0093] In step S150, based on the customization demand, the product image feature inversion feature map is modulated to generate a customized effect to obtain a customized effect image of a cultural and creative product. Specifically, in an embodiment of the present application, based on the customization demand, the product image feature inversion feature map is modulated to generate a customized effect to obtain a customized effect image of a cultural and creative product, including: inputting the customization demand modulated product image feature inversion feature map into a customized effect generator based on a diffusion model to obtain the customized effect image of the cultural and creative product. Among them, the diffusion model is a technology that has made significant progress in the field of image generation in recent years. Its working principle is similar to the diffusion process in the physical world: starting from an initial state, it gradually evolves to a more complex state until the final goal is reached. In the image generation task of the present application, the diffusion model receives the customization demand modulated product image feature inversion feature map as input, and through a series of preset diffusion steps, gradually increases the details and structure of the image until a clear cultural and creative product image that is highly matched with the user's customization needs is formed.

[0094] Specifically, the specific processing process of inputting the customized demand modulated product image feature inversion feature map into the customized effect generator based on the diffusion model is as follows:

[0095] First, it is necessary to prepare the above-obtained custom demand modulated product image feature inversion feature map as conditional input. This means that during the generation process, the diffusion model not only considers how to construct beautiful and meaningful images, but also pays special attention to how to make these images closely revolve around the user's custom needs. To this end, a series of carefully designed network layers are set up inside the diffusion model, including but not limited to encoder-decoder architecture, residual connection, attention mechanism and other components. These components work together to enable the model to incorporate new creative elements while retaining the essence of the original design.

[0096] In this process, the diffusion model uses a phased approach to image generation. In the initial stage, the model may generate a rough outline or framework based on the provided feature map, which is mainly used to determine the overall layout and general style. As the number of iterations increases, the model will continue to refine each part of the image, such as color matching, texture style, pattern design, etc., to make it richer and more delicate. Specifically, at each stage, the model will refer to the customized demand to modulate the product image feature inversion feature map to adjust the generation strategy to ensure that each step is moving in the direction expected by the user.

[0097] In addition, in order to improve the generation quality, the diffusion model also applies some advanced technical means, such as adaptive sampling, multi-scale fusion, etc. Adaptive sampling allows the model to dynamically adjust the parameter settings in subsequent steps according to the quality of the current generated image, thereby avoiding falling into the local optimal solution; while multi-scale fusion ensures the balance between detail expression and overall coordination in the final output by considering image information at different resolutions at the same time.

[0098] During the entire generation process, the system can also regularly display intermediate results to users so that feedback can be collected in a timely manner and corresponding adjustments can be made accordingly. This interactive design method not only improves user experience satisfaction, but also promotes the continuous optimization and improvement of creativity. Finally, when the model thinks that it has reached a satisfactory level, it will stop iterating and output the finished product, a rendering of a cultural and creative product made completely in accordance with the user's customized needs. This method not only greatly enriches the expression of cultural and creative products, but also provides consumers with unprecedented personalized choice space.

[0099] In particular, here, when the product prototype image fine-grained semantic coding feature map and the cultural and creative product customization demand semantic coding feature vector respectively represent the image fine-grained semantic coding features of the product prototype image and the textual semantic coding features expressed by the cultural and creative product customization demand specifications, considering the auxiliary differences in their cross-modal prompts, it is expected to improve the semantically consistent aggregation expression effect of the customization demand modulated product image feature inversion feature map obtained through cross-modal interaction optimization.

[0100] Preferably, inputting the customized demand modulated product image feature inversion feature map into a customized effect generator based on a diffusion model to obtain the customized effect image of the cultural and creative product includes:

[0101] The feature set of the customized demand modulated product image feature inversion feature map is subjected to feature clustering to obtain a customized demand modulated product image feature inversion class feature set and a customized demand modulated product image feature inversion class feature set, that is:

[0102]

[0103] in, is the feature set within the category of the customized demand modulation product image feature inversion, f 1i is the feature value of each position in the feature set of the custom demand modulation product image feature inversion class, f 24 is the feature value of each position in the out-of-class feature set of the customized demand modulation product image feature inversion;

[0104] The ratio of the number of eigenvalues ​​in the feature set of the custom demand modulated product image feature inversion class to the number of eigenvalues ​​in the feature set of the custom demand modulated product image feature inversion feature map is calculated, that is:

[0105]

[0106] Wherein, k is the number of eigenvalues ​​in the feature set of the custom demand modulated product image feature inversion class, m is the number of eigenvalues ​​in the feature set of the custom demand modulated product image feature inversion feature map, and λ is the ratio of k to m;

[0107] The ratio of the λth power of the sum of the absolute values ​​of all eigenvalues ​​in the feature set of the customized demand modulated product image feature inversion class to the λth power of the sum of the absolute values ​​of all eigenvalues ​​in the feature set of the customized demand modulated product image feature inversion feature map is calculated to obtain the customized demand modulated product image feature inversion modulation weight, that is:

[0108]

[0109] Wherein, w1 is the inversion modulation weight of the image feature of the customized demand modulation product;

[0110] Calculate the ratio of the λ / 2 power of the sum of the squares of all eigenvalues ​​in the feature set of the customized demand modulated product image feature inversion class to the λ / 2 power of the sum of the squares of all eigenvalues ​​in the feature set of the customized demand modulated product image feature inversion feature map to obtain the customized demand modulated product image feature inversion harmonic weight, that is:

[0111]

[0112] Wherein, w2 is the inversion and harmonization weight of the image feature of the customized demand modulated product;

[0113] For each feature value in the feature set within the custom demand modulation product image feature inversion class, calculate the product of the feature value and the custom demand modulation product image feature inversion harmonic weight value and add the custom demand modulation product image feature inversion modulation weight value to obtain the optimized feature value, that is:

[0114] f′ 1i =w2×f 1i +w1

[0115] Among them, f′ 1i is the optimized feature value of each position of the feature set within the image feature inversion class of the customized demand modulation product;

[0116] For each feature value in the out-of-class feature set of the customized demand modulated product image feature inversion, the product of the feature value and the weight of the customized demand modulated product image feature inversion modulation value is calculated to obtain the optimized feature value, that is:

[0117] f 24 ′=w1×f 24

[0118] Among them, f 24 ' is the optimized feature value of each position of the out-of-class feature set of the customized demand modulation product image feature inversion;

[0119] The optimized customized demand modulated product image feature inversion feature map composed of the optimized feature values ​​of the feature set within the customized demand modulated product image feature inversion class and the feature set outside the customized demand modulated product image feature inversion class is input into the customized effect generator based on the diffusion model to obtain the customized effect image of the cultural and creative product.

[0120] Therefore, while performing feature clustering on the customized demand modulated product image feature inversion feature map, the interactive description of the key feature information of the customized demand modulated product image feature inversion feature map in the clustering process is carried out, and the geometric equivariant topology of the features is constructed by low-rank harmonic modulation of the features on the equivariant basis of the clustering features and the features of the whole of the customized demand modulated product image feature inversion feature map, so as to obtain the graphical distribution translation and rotation symmetry of the clustering features of the customized demand modulated product image feature inversion feature map relative to the features of the whole, thereby introducing geometric message passing into the feature expression of the customized demand modulated product image feature inversion feature map, and realizing the cluster mapping symmetry of the customized demand modulated product image feature inversion feature map through manipulation of irreducible low-rank order coefficients, thereby improving the consistency of the clustering-based feature representation of the customized demand modulated product image feature inversion feature map, thereby improving the image quality of the customized effect image of the cultural and creative product obtained by inputting the customized demand modulated product image feature inversion feature map into the customized effect generator based on the diffusion model.

[0121] In summary, the intelligent service method for customized cultural and creative products based on the embodiment of the present application is explained, which extracts product prototype images from the product prototype library as initial materials, and uses deep learning-based image processing technology to perform semantic analysis on the prototype images of cultural and creative products, and captures the key design elements in the image. At the same time, combined with the customization needs of cultural and creative products input by users, through fine-grained cross-modal semantic interaction between product customization needs and product prototype images, customized feature modulation of product prototype images is achieved according to product customization needs, and customized effect images of cultural and creative products are intelligently generated based on this. In this way, users' personalized customization needs can be quickly responded to, the design cycle can be shortened, customization efficiency can be improved, and customization costs can be reduced, thereby effectively improving users' customization experience.

[0122] Figure 4 FIG. 1 is a block diagram of an intelligent service system for customizing cultural and creative products according to an embodiment of the present application. Figure 4 As shown, according to the embodiment of the present application, the intelligent service system 100 for customizing cultural and creative products includes: a product image customization demand acquisition module 110, which is used to extract product prototype images from a product prototype library, and at the same time obtain cultural and creative product customization demands input by users, wherein the cultural and creative product customization demands include aesthetic preferences, cultural backgrounds, and descriptions of demands for specific occasions; a product prototype image semantic feature extraction module 120, which is used to perform image semantic feature extraction on the product prototype image to obtain a fine-grained semantic coding feature map of the product prototype image; a customization demand semantic coding module 130, which is used to perform standardized expression and semantic coding on the cultural and creative product customization demands to obtain a semantic coding feature vector of the cultural and creative product customization demands; a customization demand modulation module 140, which is used to perform fine-grained semantic interactive optimization modulation on the fine-grained semantic coding feature map of the product prototype image based on the semantic coding feature vector of the cultural and creative product customization demands to obtain a customization demand modulated product image feature inversion feature map; and a customization effect generation module 150, which is used to generate customization effects based on the customization demand modulated product image feature inversion feature map to obtain a customized effect image of the cultural and creative product.

[0123] The specific operations of each step in the above-mentioned cultural and creative product customization intelligent service system have been referenced above. Figures 1 to 3 The invention has been introduced in detail in the description of the intelligent service method for customization of cultural and creative products, and therefore, its repeated description will be omitted.

[0124] As described above, the intelligent service system 100 for customization of cultural and creative products according to the embodiment of the present disclosure can be implemented in various wireless terminals, such as a server with an intelligent service algorithm for customization of cultural and creative products. In a possible implementation, the intelligent service system 100 for customization of cultural and creative products according to the embodiment of the present disclosure can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the intelligent service system 100 for customization of cultural and creative products can be a software module in the operating system of the wireless terminal, or can be an application developed for the wireless terminal; of course, the intelligent service system 100 for customization of cultural and creative products can also be one of the many hardware modules of the wireless terminal.

[0125] Alternatively, in another example, the cultural and creative product customization intelligent service system 100 and the wireless terminal may also be separate devices, and the cultural and creative product customization intelligent service system 100 may be connected to the wireless terminal via a wired and / or wireless network, and transmit interactive information in accordance with an agreed data format.

[0126] The above is merely an example of the principles of the present disclosure, and various modifications may be made by those skilled in the art without departing from the scope of the present disclosure. The above embodiments are presented for the purpose of illustration rather than limitation. The present disclosure may also take many forms other than those explicitly described herein.

Claims

1. An intelligent service method for customizing cultural and creative products, characterized in that: include: Extracting product prototype images from a product prototype library, and obtaining cultural and creative product customization requirements input by a user, wherein the cultural and creative product customization requirements include aesthetic preferences, cultural background, and descriptions of requirements for specific occasions; Extracting image semantic features from the product prototype image to obtain a fine-grained semantic encoding feature map of the product prototype image; Standardizing and semantically encoding the cultural and creative product customization requirements to obtain a semantically encoded feature vector of the cultural and creative product customization requirements; Based on the cultural and creative product customization demand semantic coding feature vector, the product prototype image fine-grained semantic interactive optimization modulation is performed on the product prototype image fine-grained semantic coding feature map to obtain a customization demand modulation product image feature inversion feature map; Based on the customization demand, the product image feature inversion feature map is modulated to generate a customization effect to obtain a customized effect image of the cultural and creative product.

2. The intelligent service method for customizing cultural and creative products according to claim 1, characterized in that: Extracting image semantic features from the product prototype image to obtain a fine-grained semantic encoding feature map of the product prototype image includes: The product prototype image is input into a product prototype image feature extractor based on the ViT model to obtain a fine-grained semantic encoding feature map of the product prototype image.

3. The intelligent service method for customizing cultural and creative products according to claim 2, characterized in that: The cultural and creative product customization requirements are expressed in a standardized manner and semantically encoded to obtain a cultural and creative product customization requirement semantic encoding feature vector, including: Inputting the cultural and creative product customization requirements into a customization requirement specification expression module based on a large language model to obtain a cultural and creative product customization requirement specification expression; The cultural and creative product customization requirement specification expression is semantically encoded to obtain the cultural and creative product customization requirement semantic encoding feature vector.

4. The intelligent service method for customizing cultural and creative products according to claim 3, characterized in that: Based on the cultural and creative product customization demand semantic coding feature vector, the product prototype image fine-grained semantic coding feature map is subjected to fine-grained semantic interactive optimization modulation to obtain a customization demand modulation product image feature inversion feature map, including: Performing autocorrelation coding on the cultural and creative product customization demand semantic coding feature vector to obtain a cultural and creative product customization demand autocorrelation coding matrix; Performing local fine-grained semantic interaction coding on the cultural and creative product customization demand autocorrelation coding matrix and the product prototype image fine-grained semantic coding feature map to obtain a set of customization demand-product prototype image cross-modal prompt information semantic coding matrices; Based on the set of the customization demand-product prototype image cross-modal prompt information semantic coding matrices, the cultural and creative product customization demand autocorrelation coding matrix and the product prototype image fine-grained semantic coding feature map are guided to perform cross-modal interactive optimization to obtain the customization demand modulated product image feature inversion feature map.

5. The intelligent service method for customizing cultural and creative products according to claim 4, characterized in that: The cultural and creative product customization demand semantic coding feature vector is autocorrelatedly coded to obtain a cultural and creative product customization demand autocorrelation coding matrix, including: The product of the cultural and creative product customization demand semantic coding feature vector and the transposed vector of the cultural and creative product customization demand semantic coding feature vector is calculated to obtain the cultural and creative product customization demand autocorrelation coding matrix.

6. The intelligent service method for customizing cultural and creative products according to claim 5, characterized in that: The cultural and creative product customization demand autocorrelation coding matrix and the product prototype image fine-grained semantic coding feature map are locally fine-grained semantically interactively coded to obtain a set of customization demand-product prototype image cross-modal prompt information semantic coding matrices, including: Performing feature decoupling on the fine-grained semantic encoding feature map of the product prototype image to obtain a set of local semantic feature matrices of the product prototype image; Performing a linear transformation on the cultural and creative product customization demand autocorrelation coding matrix to obtain a cultural and creative product customization demand query coding matrix and a cultural and creative product customization demand value coding matrix; The local semantic feature matrices of each product prototype image in the set of the cultural and creative product customization demand query coding matrix, the cultural and creative product customization demand value coding matrix and the product prototype image local semantic feature matrix are respectively input into the cross-modal prompt information encoder based on the converter structure to obtain the set of the customization demand-product prototype image cross-modal prompt information semantic coding matrices.

7. The intelligent service method for customizing cultural and creative products according to claim 6, characterized in that: Based on the set of the semantic coding matrix of the cross-modal prompt information of the customization demand-product prototype image, the semantic coding feature vector of the customization demand of the cultural and creative product and the fine-grained semantic coding feature map of the product prototype image are guided to perform cross-modal interactive optimization to obtain the feature inversion feature map of the customization demand modulated product image feature, including: Inputting the set of the customization demand-product prototype image cross-modal prompt information semantic encoding matrices into a decoder-based information gating unit to obtain a set of customization demand-product prototype image cross-modal semantic interaction attention weights; Inputting the set of the cultural and creative product customization demand autocorrelation coding matrix and the product prototype image local semantic feature matrix into a cross-modal interaction unit to obtain a set of customization demand-product prototype image cross-modal interaction local feature matrices; The set of the customization demand-product prototype image cross-modal semantic interaction attention weights and the set of the customization demand-product prototype image cross-modal interaction local feature matrices are input into the cross-modal interaction optimization unit to obtain the customization demand modulated product image feature inversion feature map.

8. The intelligent service method for customizing cultural and creative products according to claim 7, characterized in that: Based on the customization requirement, the product image feature inversion feature map is modulated to generate a customization effect to obtain a customized effect image of a cultural and creative product, including: The customized demand modulated product image feature inversion feature map is input into a customized effect generator based on a diffusion model to obtain the customized effect image of the cultural and creative product.

9. The intelligent service system for customized cultural and creative products is characterized by: include: A product image customization demand acquisition module is used to extract product prototype images from the product prototype library and simultaneously acquire cultural and creative product customization demands input by users, wherein the cultural and creative product customization demands include aesthetic preferences, cultural backgrounds, and descriptions of demands for specific occasions; A product prototype image semantic feature extraction module, used to extract image semantic features from the product prototype image to obtain a fine-grained semantic encoding feature map of the product prototype image; A customization requirement semantic coding module, used for standardizing and semantically coding the customization requirement of the cultural and creative product to obtain a semantic coding feature vector of the customization requirement of the cultural and creative product; A customization demand modulation module is used to perform fine-grained semantic interaction optimization modulation on the fine-grained semantic coding feature map of the product prototype image based on the semantic coding feature vector of the customization demand of the cultural and creative product to obtain a customization demand modulation product image feature inversion feature map; The customized effect generation module is used to modulate the product image feature inversion feature map based on the customized demand to generate customized effects to obtain a customized effect image of the cultural and creative product.

10. The intelligent service system for customizing cultural and creative products according to claim 9, characterized in that: The customized effect generation module is used to: The customized demand modulated product image feature inversion feature map is input into a customized effect generator based on a diffusion model to obtain the customized effect image of the cultural and creative product.

Images