Generative artificial intelligence product display method and system based on image processing

Abstract

The invention provides a generative artificial intelligence product display method and system based on image processing. The method comprises the following steps: firstly, receiving display demand information of a user; secondly, automatically identifying life cycle stages of corresponding types of products based on shelf time and sales data of the products; thirdly, constructing a condition vector according to the display demand information, the life cycle stage and the user interaction feedback data of the user; then, inputting the condition vector into a conditional generative adversarial network model, and generating a display image corresponding to the display demand by a generator based on the condition vector; and finally, displaying the generated display image. According to the method, the conditional vectors are input into the conditional generative adversarial network model, and the generator generates the display image corresponding to the display demand based on the conditional vectors, so that the accuracy and the automation degree of the product generation image required by the user are greatly improved, and the user experience is greatly improved.

Description

Technical Field

[0001] This invention relates to the field of automated production processing technology, specifically to a generative artificial intelligence product display method and system based on image processing. Background Technology

[0002] In the digital economy era, consumers are increasingly demanding personalized and precise product displays. Traditional product display methods rely heavily on manual design, static images, or videos, lacking the ability to generate displays in real-time and automatically adapt to user needs. Especially in e-commerce and advertising, with the diversification of product types and the multifaceted nature of user demands, intelligently generating product display images that meet specific display requirements has become a crucial direction for businesses to improve user experience and conversion rates. However, most current display solutions suffer from problems such as limited content generation, slow response times, and difficulty in automatically adapting to market conditions, failing to meet the requirements of the modern business environment.

[0003] Generative artificial intelligence (GAN) technology, especially Generative Adversarial Networks (GANs), has demonstrated powerful capabilities in automatically generating multimedia content such as images and audio. GANs, through the interplay between a generator and a discriminator, can generate high-quality, realistic image content. Conditional Generative Adversarial Networks (cGANs) further incorporate conditional vectors, enabling the generator to produce content that meets specific requirements based on given conditions, thus achieving customized image generation. This technology has wide applications in areas such as ad generation and personalized content recommendation. However, current cGAN applications often rely on single conditions for content generation, lacking comprehensive consideration of multi-dimensional conditions such as product lifecycle stages and user feedback data, resulting in a low degree of matching between generated content and display needs.

[0004] In e-commerce, products typically go through multiple lifecycle stages, such as pre-launch, official launch, peak sales period, price reduction promotions, and clearance countdown. Different stages require different display methods to better attract user attention. For example, images displayed during the pre-launch stage might emphasize the product's uniqueness, while those during the peak sales period might highlight its popularity, and clearance countdowns need to convey a sense of urgency. Currently, product display images often cannot be adjusted according to these dynamic lifecycle stages, relying instead on static content or manual design, making it difficult to quickly respond to market changes.

[0005] On the other hand, users' personalized needs and feedback are increasingly becoming key factors in product presentation. By analyzing data such as click-through rates, browsing time, purchase history, and user reviews, we can understand users' preferred presentation styles and content types. However, traditional presentation methods often struggle to utilize this personalized data to generate customized content. Even when some technical solutions attempt to incorporate user feedback data, the actual results are often unsatisfactory, lacking a systematic solution for multi-condition, multi-stage presentation needs.

[0006] Furthermore, existing product display methods suffer from insufficient automation, low production efficiency, and unstable quality. Current technologies fail to consider product lifecycle stages during generator processing, cannot accurately capture product characteristics at different lifecycle stages, and cannot ensure content aligns well with the current market environment. Moreover, they do not incorporate user interaction feedback data into conditional vectors to generate more personalized display images that improve user experience and conversion rates. They also lack consideration of product lifecycle stages. Additionally, current technologies do not employ conditional generative adversarial networks (GANs), relying on static matching from scene libraries rather than dynamic content generation through non-linear mapping, resulting in low flexibility and automation. To improve product display efficiency, enhance image generation quality, reduce manual reliance, and address diverse customer order demands, a new, refined, automated, real-time, and targeted product image generation method is urgently needed to address the limitations of existing methods and improve efficiency and accuracy, ultimately enhancing user satisfaction. Summary of the Invention

[0007] To address the aforementioned problems in existing technologies, this invention provides a generative artificial intelligence product display method and system based on image processing. The method first receives user display request information; second, it automatically identifies the lifecycle stage of the corresponding product category based on the product's listing time and sales data; third, it constructs a condition vector based on the user's display request information, lifecycle stage, and user interaction feedback data; then, it inputs the condition vector into a conditional generative adversarial network (GAN) model, and the generator generates a display image corresponding to the display request based on the condition vector; finally, it displays the generated image. This application, by inputting the condition vector into a GAN model and having the generator generate a display image corresponding to the display request based on the condition vector, significantly improves the accuracy and automation of generating images of user-requested products, greatly enhancing the user experience.

[0008] This application provides a generative artificial intelligence product demonstration method based on image processing, including the following steps:

[0009] S1: Receive user display requirements information, including product category, display style, and applicable scenario type;

[0010] S2: Automatically identify the life cycle stage of the corresponding product category based on the product's listing time and sales data. The life cycle stage includes the pre-heating, official release, hot-selling period, price reduction promotion, and clearance countdown stage.

[0011] S3: Construct a condition vector based on the user's display needs, lifecycle stage, and user interaction feedback data;

[0012] S4: Input the conditional vector into the conditional generative adversarial network model, and the generator generates the display image corresponding to the display requirements based on the conditional vector;

[0013] S5: Display the generated image.

[0014] Preferably, the user interaction feedback data includes historical click-through rates, browsing history, browsing dwell time, purchase history, and user reviews; the product categories include: clothing, home furnishings, electronic products, food, cosmetics, jewelry, sports equipment, and office supplies; the display styles include: minimalist style, retro style, modern style, luxury style, natural style, technological style, childlike style, and festive style; the applicable scene types include: indoor scenes, outdoor scenes, shopping mall scenes, home environment, office environment, exhibition environment, catering environment, and gym environment.

[0015] Preferably, the automatic identification of the product category's lifecycle stage based on the product's listing time and sales data includes:

[0016] S21: Construct real-time feature data for product categories and construct an input feature vector X. The feature data includes the listing time, sales volume, click volume, and inventory of products in the corresponding category.

[0017] S22: Calculate the probability P(y=k|X) that the product belongs to each stage of its life cycle. The calculation formula is as follows:

[0018]

[0019] Where P(y=k|X) represents the probability that the product belongs to the k-th life cycle stage, X is the input feature vector, k represents the specific life cycle stage number, and K is the total number of life cycle stages; The weight vector β represents the weight vector corresponding to lifecycle stage k. T The dot product of the input feature vector X and exp is the natural exponential function, T is the transpose of the vector; j is a natural number, j∈(1,K); y is a parameter.

[0020] S23: Select the lifecycle stage with the highest probability value as the lifecycle stage corresponding to the current product.

[0021] Preferably, the step of constructing a condition vector based on the user's display needs information, lifecycle stage, and user interaction feedback data includes:

[0022] S31: Encode or embed words for product category, display style, applicable scenario type, and product life cycle stage respectively to generate category vector, style vector, scenario vector, and life cycle stage vector;

[0023] S32: Numericalize or normalize the click-through rate, browsing dwell time, probability values ​​of the current product's corresponding lifecycle stage, and purchase records to generate user feedback vectors.

[0024] S33: The category vector, style vector, scene vector, lifecycle stage vector, and user feedback vector are concatenated end to end to form a complete conditional vector, which serves as the input to the generative adversarial network model.

[0025] Preferably, the step of inputting the conditional vector into the conditional generative adversarial network model, and the generator generating a display image corresponding to the display requirements based on the conditional vector, includes:

[0026] S41: Input the constructed conditional vector U and random noise vector z into the trained generator G;

[0027] S42: The generator automatically generates a display image G(U,z) based on the condition vector U and the random noise vector z, including: the generator transforms the random noise based on the vector information of the condition vector U and maps it to a display image G(U,z) with conditional features; the discriminator D receives the image G(U,z) and the condition vector U generated by the generator and outputs a probability value to determine whether the generated image meets the set threshold to meet the requirements of the condition vector.

[0028] The generator's hidden layer uses a modified Sigmoid activation function, as shown below:

[0029]

[0030] Where, α k =P(y=k|X) represents the probability value of the current life cycle stage k, which serves as a dynamic factor for the Sigmoid activation function; X is the input feature vector for construction.

[0031] S43: When the output probability value reaches the set threshold, output the generated display image.

[0032] This application also provides a generative artificial intelligence product display system based on image processing, including:

[0033] The user display requirement input module receives the user's display requirement information, including product category, display style, and applicable scenario type;

[0034] The product lifecycle stage identification module automatically identifies the lifecycle stage of the corresponding product category based on the product's listing time and sales data. The lifecycle stages include the pre-heating, official release, hot-selling period, price reduction promotion, and clearance countdown stages.

[0035] The condition vector construction module constructs condition vectors based on user display requirements, lifecycle stages, and user interaction feedback data.

[0036] The image generation module takes conditional vectors as input to the conditional generative adversarial network model, and the generator generates display images corresponding to the display requirements based on the conditional vectors.

[0037] The output module displays the generated images.

[0038] Preferably, the user interaction feedback data includes historical click-through rates, browsing history, browsing dwell time, purchase history, and user reviews; the product categories include: clothing, home furnishings, electronic products, food, cosmetics, jewelry, sports equipment, and office supplies; the display styles include: minimalist style, retro style, modern style, luxury style, natural style, technological style, childlike style, and festive style; the applicable scene types include: indoor scenes, outdoor scenes, shopping mall scenes, home environment, office environment, exhibition environment, catering environment, and gym environment.

[0039] Preferably, the automatic identification of the product category's lifecycle stage based on the product's listing time and sales data includes:

[0040] The module for constructing the input feature vector X: constructs real-time feature data for product categories and constructs the input feature vector X. The feature data includes the listing time, sales volume, click volume, and inventory of products in the corresponding category.

[0041] Probability Calculation Module: Calculates the probability P(y=k|X) of a product belonging to each stage of its life cycle. The calculation formula is as follows:

[0042]

[0043] Where P(y=k|X) represents the probability that the product belongs to the k-th life cycle stage, X is the input feature vector, k represents the specific life cycle stage number, and K is the total number of life cycle stages; The weight vector β represents the weight vector corresponding to lifecycle stage k. TThe dot product of the input feature vector X and exp is the natural exponential function, T is the transpose of the vector; j is a natural number, j∈(1,K); y is a parameter.

[0044] The module for determining the current product's lifecycle stage: Select the lifecycle stage with the highest probability value as the current product's lifecycle stage.

[0045] Preferably, the step of constructing a condition vector based on the user's display needs information, lifecycle stage, and user interaction feedback data includes:

[0046] Stage Vector Generation Module: Encodes or embeds words for product category, display style, applicable scenario type, and product lifecycle stage respectively, generating category vector, style vector, scenario vector, and lifecycle stage vector;

[0047] User feedback vector generation module: Converts or normalizes the probability values ​​of click-through rate, browsing dwell time, and purchase records corresponding to the current product's lifecycle stage to generate user feedback vectors.

[0048] Vector concatenation module: The category vector, style vector, scene vector, lifecycle stage vector, and user feedback vector are concatenated end to end to form a complete conditional vector, which serves as the input to the generative adversarial network model.

[0049] Preferably, the step of inputting the conditional vector into the conditional generative adversarial network model, and the generator generating a display image corresponding to the display requirements based on the conditional vector, includes:

[0050] Input module: Inputs the constructed conditional vector U and random noise vector z into the trained generator G;

[0051] Image construction module: The generator automatically generates a display image G(U,z) based on the condition vector U and the random noise vector z. This includes: the generator transforms the random noise based on the vector information of the condition vector U and maps it to a display image G(U,z) with conditional features; the discriminator D receives the image G(U,z) and the condition vector U generated by the generator and outputs a probability value to determine whether the generated image meets the set threshold to satisfy the requirements of the condition vector.

[0052] The generator's hidden layer uses a modified Sigmoid activation function, as shown below:

[0053]

[0054] Where, α k =P(y=k|X) represents the probability value of the current life cycle stage k, which serves as a dynamic factor for the Sigmoid activation function; X is the input feature vector for construction.

[0055] Conditional judgment output module: When the output probability value reaches the set threshold, the generated display image is output.

[0056] This invention provides a generative artificial intelligence product display method and system based on image processing, which can achieve the following beneficial technical effects:

[0057] 1. This application first receives the user's display requirement information; second, it automatically identifies the lifecycle stage of the corresponding product category based on the product's listing time and sales data; third, it constructs a condition vector based on the user's display requirement information, lifecycle stage, and user interaction feedback data; then, it inputs the condition vector into a conditional generative adversarial network model, and the generator generates a display image corresponding to the display requirement based on the condition vector; finally, it displays the generated display image. This application, by inputting the condition vector into a conditional generative adversarial network model and having the generator generate a display image corresponding to the display requirement based on the condition vector, greatly improves the accuracy and automation of generating images for user-required products, significantly enhancing the user experience.

[0058] 2. This invention firstly achieves precise capture of the characteristics of different lifecycle stages by incorporating probability values ​​of product lifecycle stages into the generator's activation function, making the displayed content more aligned with the current market environment. By calculating the probability values ​​of products at different lifecycle stages and introducing them as dynamic factors into the generator's activation function, the generator can automatically adjust the features of the output image to accurately adapt to the needs of different product lifecycle stages. This allows the displayed images generated during the pre-sale phase to highlight the product's uniqueness, demonstrate attractiveness during the peak sales period, and emphasize the urgency of the promotion during the clearance countdown phase, thereby effectively increasing user attention and product market performance. By combining user interaction feedback data into the conditional vector, more matching displayed images can be generated based on users' personalized needs, thereby improving user experience and product conversion rates. Furthermore, incorporating the probability values ​​of the lifecycle stage into the generator's activation function makes the generated displayed images more vivid and layered. For example, displayed images generated during the peak sales period may have higher visual appeal, encouraging users to buy, while displayed images generated during the clearance countdown phase have a sense of urgency, thus better stimulating users' purchasing decisions. This solution employs a conditional generative adversarial network model, which eliminates the need for static matching of scene libraries during the generation process. Instead, it dynamically generates content through a nonlinear mapping process of the generator, thereby greatly improving the flexibility and automation of content generation.

[0059] 3. This invention utilizes user interaction feedback data (such as click-through rate, browsing dwell time, purchase history, etc.) to construct conditional vectors, thereby generating display images that better meet user needs based on their preferences. The style and content of the generated images are highly matched with user preferences, significantly improving the personalization of the user experience and the accuracy of the display effect, thus increasing user retention and conversion rates. It eliminates the need for a static scene library, enhancing the automation and flexibility of display content: Unlike traditional display methods that rely on static matching based on a scene library, this application uses a conditional generative adversarial network generator to directly generate display images based on conditional vectors, avoiding the limitations of static content and greatly improving the flexibility of content generation. During the generation process, the generator dynamically generates content based on the probability of the input lifecycle stage and the user's personalized needs, automatically adapting to changes in market and user demands, significantly improving the automation and generation efficiency of product display. Attached Figure Description

[0060] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0061] Figure 1 This is a schematic diagram illustrating the steps of a generative artificial intelligence product display method based on image processing according to the present invention;

[0062] Figure 2 This is a schematic diagram of a generative artificial intelligence product display system based on image processing according to the present invention. Detailed Implementation

[0063] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0064] Example 1:

[0065] In view of the aforementioned problems mentioned in the prior art, and in order to solve the above technical problems, as shown in the appendix. Figure 1 As shown: This application provides a generative artificial intelligence product demonstration method based on image processing, including the following steps:

[0066] S1: Receive user display requirement information, including product category, display style, and applicable scenario type; the e-commerce platform wants to generate personalized display images for newly released smart home products. Through step S1 of this application's method, user display requirement information can be received. The specific implementation process is as follows: The user inputs the product category. The user can select the "Product Category" option, and the platform displays different categories of products for the user to choose from. In this embodiment, the user selects "Smart Home," including smart speakers, smart light bulbs, smart door locks, and other smart home subcategories. Product category information can be used to control the background layout and atmosphere of the display image, making the generated display image more in line with the technological feel of smart home products or the needs of home scenarios. The user selects the display style. The platform provides multiple display styles for the user to choose from, including minimalist, retro, modern, luxurious, and technological styles. In this embodiment, the user selects a technological style, meaning the generated display image should have technological style characteristics such as cool colors and simple lines to highlight the product's innovation and futuristic feel. This display style information will be used as a conditional input, making the image generated by the generator visually meet the requirements of a technological feel. Users can select the applicable scenario type for the displayed images. The platform offers a variety of common scenario types for users to choose from, such as indoor scenarios, outdoor scenarios, shopping mall scenarios, home environments, office environments, and exhibition environments. In this embodiment, the user selects a home environment, intending to showcase the application scenario of smart home products in a real home environment. The applicable scenario type information allows the generated display image to be closer to a home scene setting; for example, the background can include typical home furniture and decorations, making it easier for consumers to understand the product's application scenario. The generated condition vector is based on the user's input information: "Product Category = Smart Home," "Display Style = Technological Style," and "Applicable Scenario Type = Home Environment." The system generates a condition vector X for subsequent image generation. The product category, display style, and applicable scenario type in the condition vector will serve as one of the generator's inputs, enabling the generated image to visually simultaneously reflect the technological feel of the smart home product, the scene setting of the home environment, and the specific display style. Guided by the condition vector, the generator produces a display image with a technological style while showcasing the application effect of smart home products in a home environment. The generated display images can be directly applied to product display pages on e-commerce platforms, providing consumers with more attractive and contextualized product display effects, thereby increasing users' willingness to purchase.

[0067] S2: The system automatically identifies the product lifecycle stage based on the product's listing time and sales data. The lifecycle stages include pre-launch, official release, peak sales period, price reduction promotion, and clearance countdown. In one embodiment, a smart speaker product is listed on an e-commerce platform. The system automatically identifies the product's lifecycle stage by analyzing its listing time and sales data (such as sales volume, inventory changes, and user clicks). The product lifecycle stages include five phases: "pre-launch," "official release," "peak sales period," "price reduction promotion," and "clearance countdown." The specific identification conditions and explanations for each phase are as follows: Pre-launch stage identification: The product has just been listed, and the listing time is within two weeks, with limited sales data. The system monitors the product's listing time and determines whether it is within the pre-launch period (e.g., listing time less than two weeks and clicks not reaching a certain threshold). At this time, sales are low, and inventory has almost no change. When the pre-launch stage conditions are met, the system marks the product as "pre-launch stage," providing "pre-launch" as the lifecycle stage input for the generator's display image. The generated image will highlight the product's features in an eye-catching way, generating initial interest in the product among consumers. The "Official Launch Phase" identifies products that have been listed for more than two weeks beyond the pre-launch period, but with low sales and stable market feedback. When a product has exceeded the pre-launch period and click-through rates have increased, but sales growth is slow, the system automatically identifies it as part of the "Official Launch Phase." At this stage, users have some understanding of the product, but purchasing behavior has not yet shown significant growth. After identifying the product as being in the "Official Launch Phase," the system marks it as such and provides corresponding input to the display image generator. The generated images will focus on detailed product features to enhance user understanding and purchase intention. The "Hot Selling Period Phase" identifies products that experience a significant increase in sales, with both click-through rates and purchase conversion rates reaching high levels. When the system detects rapid sales growth, fast inventory depletion, and high click-through rates, it automatically identifies the product as being in the "Hot Selling Period." At this stage, consumer attention is high, and conversion rates are significantly improved. The system uses the "Hot Selling Period" as the lifecycle stage input for the display image generator. The generated display images will enhance the product's appeal, for example, by showcasing user reviews and sales figures to create a sense of urgency and encourage users to purchase quickly. Price Reduction / Promotion Phase Recognition: After a product has been on the market for some time, sales growth slows, inventory is sufficient, but click-through rates and purchase conversion rates decline. When the system detects a slowdown in sales growth but still a large inventory, it automatically identifies the product as being in a "price reduction / promotion phase." In this phase, the product needs price adjustments or promotional activities to attract more consumers. The system uses "price reduction / promotion" as input to the generator's lifecycle phase, and the generated display images will highlight discounts or promotional information to attract user purchase interest. Clearance Countdown Phase Recognition: The product has been on the market for a considerable time, inventory is nearing depletion, and it is about to be removed from the shelves.During the identification process, when the system detects that the inventory level is approaching the low inventory threshold (e.g., less than 5 units) and the product's sales and click-through rates are no longer increasing significantly, it automatically identifies the product as being in the "clearance countdown stage." At this point, the system will generate a strong sense of urgency effect for the product's display image, prompting users that the product is about to sell out. The system uses the "clearance countdown" as input to the generator's lifecycle stage, and the generated display image will highlight information such as "limited-time offer" and "clearance sale" to attract users to purchase in the final stage.

[0068] In the above embodiments, the system monitors the product's listing time and sales data (such as sales volume, clicks, and inventory) and automatically identifies the product's current lifecycle stage according to predetermined rules. The identified stage information serves as input for generating display images, enabling the generator to produce display images that conform to the characteristics of each stage: Pre-launch stage: showcasing product features to arouse user interest; Official launch stage: detailing product functions to promote user understanding; Hot-selling period: enhancing the sense of scarcity to incentivize user purchases; Price reduction and promotion stage: displaying promotional information to attract user attention; Clearance countdown stage: creating a sense of urgency to encourage users to purchase as soon as possible.

[0069] S3: Construct a condition vector based on the user's display requirements, product lifecycle stage, and user interaction feedback data; A user on an e-commerce platform wants to display a smart speaker product. The system needs to construct a condition vector X based on the user's display requirements, product lifecycle stage, and user interaction feedback data to input into the generator for image generation. The process of constructing the condition vector is as follows: 1. Obtain the user's display requirements information. The user inputs or selects the product's display requirements information on the platform, including the following: Product category: The user selects the "Smart Home" category (e.g., smart speaker), and the system encodes this category as a vector. Display style: The user selects "Technological style," indicating that the display image needs to have a cool color scheme and a simple style. Applicable scenario type: The user selects "Home environment," meaning that the display image should include home elements to make the product look suitable for a home setting. The system converts the display requirements information into corresponding vectors, specifically: Category vector U: The "Smart Home" category is encoded as a fixed numerical vector, such as [1,0,0] (corresponding to the identifier of the smart home category). Style U: The "Technological style" is encoded as a specific numerical vector, such as [0,1,0] (representing a technological feel). Scene Vector, UScene: Encodes "Home Environment" as a scene vector, for example [0,0,1] (representing home environment). The final vectorized form of the displayed demand information is: UDemand = [UCategory, UStyle, UScene]. Product Lifecycle Stage Information: The system identifies the current product lifecycle stage based on the product's listing time and sales data, identifying it as "Hot Selling Period". The information of the "Hot Selling Period" lifecycle stage is also encoded as a vector to represent the current market status of the product, so that the generated display image can reflect the stage characteristics (such as scarcity). Lifecycle Stage Vector, ULifecycle: Encodes "Hot Selling Period" as a specific numerical vector, for example [0,1,0,0,0], indicating that the product is in the "Hot Selling Period" stage. User Interaction Feedback Data: To make the display image more personalized, the system will collect user interaction feedback data, including the following indicators: Click-through Rate: The user click-through rate for this product is 30%. This data is standardized to [0.3]. Browsing Dwell Time: The average dwell time of users browsing this product page is 45 seconds. The dwell time is normalized to [0.45]. Purchase history: Whether the user has purchased similar products. If the user has previously purchased other smart home devices, the purchase history will be coded as [1]. User evaluation sentiment score: The system performs sentiment analysis on user evaluations and calculates the sentiment score. The normalized sentiment score is [0.8], which indicates that the user's evaluation of the product is relatively positive.The vectorized form of user interaction feedback data is: Ufeedback = [0.3, 0.45, 1, 0.8]. Construct a condition vector by concatenating the above three parts of information to form a complete condition vector X = [Udemand, Ulifecycle, Ufeedback]. The final combination of the vectors is as follows: X = [1, 0, 0, 0, 1, 0, 0, 0, 1, 0.3, 0.45, 1, 0.8].

[0070] The system uses conditional vectors to generate display images. The constructed conditional vector X is input into the generator of the conditional generative adversarial network (GAN) model. The generator produces the display image based on various information from the conditional vectors: the category vector controls the main subject, reflecting the appearance and functional characteristics of a smart speaker; the style vector controls the display style, generating a technologically advanced visual effect; the scene vector controls the background scene, presenting the product in a home environment; the lifecycle stage vector enhances the image's sense of urgency, aligning with the product's current sales status; and user interaction feedback data adjusts details, such as highlighting key areas, making the image more appealing to users. The final generated display image presents the smart speaker with a technological feel in a home environment, while emphasizing the visual appeal of a best-selling product to meet specific user display needs.

[0071] S4: Input the conditional vector into the conditional generative adversarial network model. The generator produces a display image corresponding to the display requirements based on the conditional vector. Each conditional vector X contains user display requirement information (product category, display style, applicable scenario type), product lifecycle stage (e.g., peak sales period), and user interaction feedback data (e.g., click-through rate, browsing dwell time). The system uses this conditional vector to control the content and style of the generated image to meet the user's specific display needs for the smart speaker product.

[0072] The conditional vector X is input into the Conditional Generative Adversarial Network (cGAN) model. The system uses the constructed conditional vector X as input, passing it to the generator part of the cGAN model. The conditional vector X contains multiple dimensions of display information, such as: product category: smart home; display style: technological style; scene type: home environment; life cycle stage: peak sales period; user interaction feedback data: click-through rate, browsing dwell time, etc. This conditional vector controls the content and style of the images generated by the generator, enabling them to reflect the user's specific display needs and the current market status of the product.

[0073] The generator's nonlinear mapping process involves the generator G receiving a conditional vector X and a random noise vector z as input. The generator's role is to transform the conditional vector and random noise into a suitable display image through a nonlinear mapping of the neural network. The generation process includes the following steps: Initial Input Layer: The generator concatenates the conditional vector X and the random noise vector z to form a combined vector, which serves as the generator's initial input. This combined vector contains features such as product category, display style, scene type, lifecycle stage, and user feedback information, serving as the basic conditions for image generation. Layer-by-Layer Processing of Hidden Layers: The combined vector [X, z] passes through multiple hidden layers of the generator, with each layer progressively mapping to the image space using a nonlinear activation function (such as ReLU, Sigmoid, etc.). Activation functions help the generator capture complex image features, making the generated image more technologically advanced and personalized. For example, if the generator uses a Sigmoid dynamic activation function, which includes the probability value of the product lifecycle stage as a dynamic factor, the generator's output at each layer will adaptively adjust according to the lifecycle stage, reflecting the scarcity of products during their "hot-selling" period. Image Generation at the Output Layer: After multiple layers of processing, the generator outputs the final image. The content of this image will reflect the requirements in the conditional vector: for example, a background environment of a home scene, a technological display style, and a visual effect highlighting its hot-selling status. Example of generated image content: The generated image G(X,z) will reflect the following: Smart Home Product Category: The main subject of the generated image is a smart speaker product, whose appearance and product features meet the requirements of the smart home category, such as a simple and modern design style. Technological Display Style: The overall style of the image leans towards a technological feel, using a cool-toned background and simple line design, making the product appear modern and technologically advanced, matching the user's chosen display style. Home Environment Scene: The background of the generated image is set in a home environment, and the image may contain some furniture or home decoration elements, allowing users to intuitively understand the product's application effect in a home setting. Hot-Selling Period Sensation: When generating the image, the generator incorporates the characteristics of a hot-selling period into the image, such as appropriately increasing brightness and contrast to highlight the product's visual appeal and hot-selling status, creating a sense of urgency and stimulating users' desire to buy. Personalized User Interaction Feedback Data: Combining user interaction feedback data, such as click-through rate and browsing dwell time, the generated image details are closer to user preferences. For example, if a user has spent a considerable amount of time browsing a product detail page, the generated display image will further highlight the smart speaker's key features, such as sound quality and smart functions, attracting the user to learn more. The final output display image, G(X,z), directly showcases the various features requested by the user and can be applied to product detail pages or advertising placements on e-commerce platforms.The specific features of the final displayed images include: a technologically advanced presentation of the smart speaker product in a home environment, a visual effect emphasizing the product's popularity, personalized designs that attract users, and designs that align with user preferences and interests.

[0074] Conditional Generative Adversarial Networks (cGANs) are an extension of Generative Adversarial Networks (GANs) that incorporate specific conditional information during the generation process, ensuring that the generated images or content meet the input conditional requirements. A cGAN mainly consists of a generator and a discriminator, which interact and optimize each other during training to generate high-quality images that meet the conditions.

[0075] The cGAN model structure includes a generator, G, which is a neural network responsible for generating images that meet the requirements based on input conditions. It receives a condition vector X (e.g., display requirement information, lifecycle stage, user feedback data, etc.) and a random noise vector z as input. The generator transforms the input into the final display image through a series of hidden layers. The generator's goal is to generate images that are as realistic as possible and satisfy the input conditions, making it indistinguishable from real images by the discriminator.

[0076] The discriminator, D, is also a neural network used to distinguish whether the input image is a real image or an image generated by the generator. The discriminator receives both a conditional vector X and the image as input, and outputs a probability value indicating whether the image meets the input conditions. The discriminator's goal is to accurately distinguish between real and generated images and determine whether the generated image meets the input conditions. The cGAN's working principle and training process involve adversarial training between the generator and discriminator, progressively optimizing the generator's output to ensure that the generated images are not only realistic but also meet the requirements in the conditional vector. Both the generator and discriminator receive the conditional vector X as input. The conditional vector can contain information such as display requirements, product lifecycle stages, and user feedback data, used to control the content and style of the generated image to meet specific needs. A random noise vector z serves as another input to the generator, used to introduce diversity into the image.

[0077] The generator process involves the generator G receiving a combined input [X,z] and performing a series of nonlinear mappings through a neural network to progressively transform the combined input into an output image. In the generator's hidden layers, nonlinear activation functions (such as ReLU or Sigmoid) are used to process the output of each layer, enabling the generator to learn complex features. The generator's goal is to generate an image G(X,z) that best meets the desired effect and satisfies the requirements of the condition vector X. For example, when the condition vector is specified as "hot-selling period," the generated image will highlight the product's attractiveness or scarcity. The discriminator process involves the discriminator D receiving either the generated image G(X,z) or a real image, along with the corresponding condition vector X. The discriminator calculates the features of the image and the condition vector, classifying the image as real or generated, and evaluating whether the generated image meets the condition X. The discriminator's output is a probability value indicating whether the input image satisfies the condition. A high probability value indicates that the image is realistic and meets the condition, while a low probability value indicates that the image is not realistic or does not meet the condition. Loss Functions of the Generator and Discriminator: The discriminator's loss function is optimized by minimizing the loss function; the generator's loss function is optimized by minimizing the loss function. Training Process: In each training iteration, the generator and discriminator are updated alternately. First, the discriminator updates its own parameters to improve its ability to distinguish between real and generated images; then, the generator updates its own parameters to generate more realistic and suitable images. The ultimate goal of training is to reach a Nash equilibrium between the discriminator and the generator, where the discriminator cannot distinguish between real and generated images, and the generator's output image has the highest quality and meets the input conditions.

[0078] The workflow of the conditional generation process is as follows: During generation, the system inputs a conditional vector X into a trained generator. The generator generates a display image that meets the conditions through the following steps: It receives the conditional vector and random noise. The input conditional vector X contains information such as product category, display style, applicable scenario type, and lifecycle stage. The generator concatenates the conditional vector X and the random noise vector z into a combined vector [X,z], which is then used as input. Layer-by-layer nonlinear mapping: The generator inputs the combined vector [X,z] into the first layer of a neural network, processing it layer by layer through a series of hidden layers. The activation function in each layer is adjusted based on the conditional vector; for example, a conditionally driven sigmoid activation function is used to amplify or reduce the influence of a specific lifecycle stage. As the number of layers increases, the generator gradually captures image features, ensuring that the generated display image conforms to the information in the conditional vector. The output layer generates the display image; finally, the generator outputs a display image G(X,z) that meets the conditions. For example, if the conditional vector specifies "smart speaker" as the product category, "tech style" as the display style, "home environment" as the scene type, and the current lifecycle stage is "hot selling period", then the generated image will show the smart speaker's tech-savvy appearance in a home environment, while highlighting the product's hot selling status.

[0079] S5: Display the generated showcase image. After the showcase image is output, the system optimizes and adjusts details such as brightness, contrast, and color to ensure consistent visual effects across different devices. Based on user feedback, the image can be further adjusted to maintain clarity across different resolutions and screen sizes. The optimized showcase image can be applied across multiple channels, including e-commerce platforms, social media, and advertising systems, to enhance the product's market influence. For example, on e-commerce platforms, the showcase image will be used on product detail pages or recommendation sections to increase user purchase interest through attractive displays. If users have personalized needs, the system will personalize the generated image display on different pages based on their browsing history or preferences. For example, through a recommendation system, customized showcase images will be prioritized for users who may be interested in the product.

[0080] In some embodiments, the user interaction feedback data includes historical click-through rates, browsing history, browsing dwell time, purchase history, and user reviews; the product categories include: clothing, home furnishings, electronic products, food, cosmetics, jewelry, sports equipment, and office supplies; the display styles include: minimalist style, retro style, modern style, luxury style, natural style, technological style, childlike style, and festive style; the applicable scene types include: indoor scenes, outdoor scenes, shopping mall scenes, home environment, office environment, exhibition environment, catering environment, and gym environment.

[0081] In some embodiments, the automatic identification of the product category's lifecycle stage based on the product's listing time and sales data includes:

[0082] S21: Construct real-time feature data for product categories and construct an input feature vector X. The feature data includes the listing time, sales volume, click volume, and inventory of products in the corresponding category.

[0083] S22: Calculate the probability P(y=k|X) that the product belongs to each stage of its life cycle. The calculation formula is as follows:

[0084]

[0085] Where P(y=k|X) represents the probability that the product belongs to the k-th life cycle stage, X is the input feature vector, k represents the specific life cycle stage number, and K is the total number of life cycle stages; The weight vector β represents the weight vector corresponding to lifecycle stage k. T The dot product of the input feature vector X and exp is the natural exponential function, T is the transpose of the vector; j is a natural number, j∈(1,K); y is a parameter.

[0086] S23: Select the lifecycle stage with the highest probability value as the lifecycle stage corresponding to the current product.

[0087] In one embodiment, an e-commerce platform lists a smart speaker product. The system identifies the product's lifecycle stage by collecting real-time data, including listing time, sales volume, click volume, and inventory. The product lifecycle stages are divided into five phases: pre-launch, official release, peak sales period, price reduction promotion, and clearance countdown. Step S21: Construct real-time feature data. Listing time: The smart speaker product has been listed for more than a month. Sales volume: In the past week, the product's sales volume has grown rapidly, with a significant increase in daily sales. Click volume: The product page maintains a high click volume, indicating very high user attention. Inventory: Inventory is being depleted quickly, and the platform has replenished stock multiple times to meet market demand. Based on the above real-time data, the system constructs these data into an input feature vector for probability calculation in subsequent stages. Step S22: Calculate the probability of the product belonging to each lifecycle stage. The system calculates the probability of the product belonging to different lifecycle stages based on the above feature data. The following is the judgment logic for each stage: Pre-launch stage: Generally applicable to newly listed products, with relatively low sales and click volume. Official Launch Phase: Applicable to products where sales growth is not yet significant, clicks have increased to some extent but sales are changing slowly. Peak Sales Period: Sales and clicks both increase significantly, inventory continues to decline, requiring frequent restocking. Price Reduction Promotion: Sales growth slows, but inventory remains high; the platform may adopt a price reduction promotion strategy. Clearance Countdown: Product inventory is nearing depletion, and sales and clicks have decreased significantly. In this embodiment, based on the analysis of real-time feature data, the system finds that the peak sales period has the highest probability value because the product's sales, clicks, and inventory depletion rate all meet the typical characteristics of a peak sales phase. Step S23: Select the lifecycle stage with the highest probability value. The system compares the probability values ​​of each stage and selects the "peak sales period" with the highest probability as the current lifecycle stage of the product, marking it as "peak sales period." This result will be used to generate personalized display images, highlighting the scarcity and attractiveness of peak-selling products, further promoting sales. By analyzing the product's listing time, sales volume, clicks, and inventory, the system can accurately identify the current lifecycle stage. In this embodiment, the system automatically identifies that the smart speaker product is in a "hot-selling period," and thus incorporates the characteristics of the hot-selling period when generating the display image, such as highlighting sales volume and scarcity, to enhance users' desire to buy.

[0088] In some embodiments, constructing a condition vector based on the user's display needs information, lifecycle stage, and user interaction feedback data includes:

[0089] S31: Encode or embed words for product category, display style, applicable scenario type, and product life cycle stage respectively to generate category vector, style vector, scenario vector, and life cycle stage vector;

[0090] S32: Numericalize or normalize the click-through rate, browsing dwell time, probability values ​​of the current product's corresponding lifecycle stage, and purchase records to generate user feedback vectors.

[0091] S33: The category vector, style vector, scene vector, lifecycle stage vector, and user feedback vector are concatenated to form a complete conditional vector, which serves as the input to the generative adversarial network model. The system encodes or embeds words into the display requirements and lifecycle stages. For example, if the product category is "smart home," the system encodes the "smart home" category, generating a category vector. The encoded vector for the smart home category is [1,0,0], indicating that the product is a smart home product. For the display style, if the user selects a "tech-savvy" style, the system encodes the style, generating a style vector. For example, the vector for the tech-savvy style is [0,1,0], emphasizing the tech-savvy display effect. For the applicable scene type, if the user selects a "home environment," the system encodes the home environment, generating a scene vector. For example, the scene vector for the home environment is [0,0,1], helping the generator understand the application scenario of the displayed image. For the lifecycle stage, if the product is identified as being in its "hot-selling period," the system encodes the "hot-selling period" lifecycle stage, generating a lifecycle stage vector. For example, the vector representation of the peak sales period is [0,1,0,0,0]. After coding, the vector representations of demand information and life cycle stages are as follows: Category vector: [1,0,0], Style vector: [0,1,0], Scene vector: [0,0,1], Life cycle stage vector: [0,1,0,0,0].

[0092] Step S32: Numericalize or normalize the user interaction feedback data. Click-through rate: The click-through rate for this product is 30%. The system normalizes the click-through rate and represents it as [0.3], which is used as part of the user feedback vector. Browsing dwell time: The average dwell time of users browsing the product page is 45 seconds. The system normalizes the dwell time to [0.45], which is used in the user feedback vector. Purchase history: The user has previously purchased other smart home devices. The system numericalizes the purchase history as [1], indicating that the user has a preference for smart home products. Lifecycle stage probability value: The system calculates the probability value of this product being in the "hot-selling period" to be 0.85, indicating that there is an 85% probability that the product is in the hot-selling stage. This value is normalized and added to the user feedback vector, represented as [0.85]. The final user feedback vector is: User feedback vector: [0.3, 0.45, 1, 0.85].

[0093] Step S33: Generate a complete conditional vector by concatenating the category vector, style vector, scene vector, lifecycle stage vector, and user feedback vector to form a complete conditional vector X = [1,0,0,0,1,0,0,0,1,0,1,0,0,0,0.3,0.45,1,0.85]. The generated conditional vector is applied to the generative adversarial network model. The system takes the complete conditional vector X as input and passes it to the generator part of the conditional generative adversarial network model. The generator generates a display image that meets the following characteristics based on the conditional vector: Product category characteristics (smart home): highlighting the smart home functions of the product; Display style (tech style): the image design has a cool color tone and a modern feel; Scene type (home environment): the background is set to a home scene; Lifecycle stage characteristics (hot sales period): enhancing users' desire to buy by highlighting sales volume and scarcity; User interaction feedback data: displaying more product details based on user browsing preferences. This embodiment encodes or normalizes user display needs, lifecycle stages, and interaction feedback data to construct a complete conditional vector X, which is used to generate personalized display images that meet the user's requirements. The resulting display images accurately match user needs in terms of content and style, improving product display effectiveness and user experience.

[0094] In some embodiments, the step of inputting conditional vectors into a conditional generative adversarial network model, and the generator generating a display image corresponding to the display requirements based on the conditional vectors, includes:

[0095] S41: Input the constructed conditional vector U and random noise vector z into the trained generator G;

[0096] S42: The generator automatically generates a display image G(U,z) based on the condition vector U and the random noise vector z, including: the generator transforms the random noise based on the vector information of the condition vector U and maps it to a display image G(U,z) with conditional features; the discriminator D receives the image G(U,z) and the condition vector U generated by the generator and outputs a probability value to determine whether the generated image meets the set threshold to meet the requirements of the condition vector.

[0097] The generator's hidden layer uses a modified Sigmoid activation function, as shown below:

[0098]

[0099] Where, α k =P(y=k|X) represents the probability value of the current life cycle stage k, which serves as a dynamic factor for the Sigmoid activation function; X is the input feature vector for construction.

[0100] S43: When the output probability value reaches the set threshold, output the generated display image.

[0101] In one embodiment, an e-commerce platform lists a smart speaker product. The system needs to generate display images based on the product's display requirements, lifecycle stage, and user feedback data to better attract users. The following is a detailed description of the entire generation process. Step S41: Input a conditional vector and random noise into the generator. The system has already constructed a conditional vector U, which includes display requirement information (product category, display style, scene type), the product's lifecycle stage (e.g., peak sales period), and user interaction feedback data (e.g., click-through rate, browsing dwell time, etc.). The conditional vector U represents all control information when generating the display image, such as the "smart home" product category, the "tech-savvy" display style, the "home environment" scene, and the "peak sales period" stage. Input a random noise vector. The system generates a random noise vector z to introduce diversity into the display image. The random noise vector allows the generated image to have different details while meeting the conditions. For example, by introducing random noise, the generator can generate images with different angles or lighting effects, making the displayed content less monotonous.

[0102] Step S42: The generator generates a display image. The generator uses a non-linear mapping. After receiving the combined input [U,z], it passes it layer by layer, processing the data through hidden layers in the neural network. In the hidden layers, the generator uses a modified Sigmoid activation function. In the activation function of each hidden layer, the probability value of the product lifecycle stage (e.g., the probability value of "hot-selling period") is used as a dynamic factor. For example, when the product is in its hot-selling period, the activation function increases the weight of specific features, making the generated image more prominent in terms of scarcity and hot-selling characteristics. This dynamic factor controls the amplification or reduction of different features in the activation function. For example, if the probability of a hot-selling period is high, the generated display image may be more vivid, highlighting the product's attractiveness. The generator outputs a display image G(U,z) after multi-layer processing, which conforms to the display requirement information in the conditional vector. The generated display images may include the following characteristics: Product category characteristics: showcasing the appearance and features of the smart speaker; Display style characteristics: the image features a cool color scheme and a technological design; Scene type characteristics: the image background is a home scene, showcasing the product's application in the home environment; Lifecycle characteristics: the image shows a high demand for the product, creating a hot-selling atmosphere.

[0103] The discriminator's judgment process involves generating an image G(U,z). After generating the image, the discriminator D receives the generated image and the condition vector U, and determines whether the generated image meets the condition requirements. The discriminator's goal is to output a probability value to judge the quality and compliance of the generated image. If the generated image is of high quality and meets the condition requirements, the discriminator will output a higher probability value; if the generated image does not meet the requirements or is of low quality, the discriminator's probability value will be lower. For example, if the discriminator believes that the generated image can well demonstrate the product's popularity and technological feel, the output probability value will be higher.

[0104] Step S43: Output a display image that meets the requirements. Check the output probability value of the discriminator. If the output probability value of the discriminator reaches a set threshold (e.g., 0.8), the system considers the generated image quality to meet the requirements. If the generated display image meets the conditions, the system will output the image as the final display image and directly apply it to the e-commerce platform. Otherwise, the generator and discriminator will continue iterative training to improve the quality of the generated image. The display image that meets the conditions will be applied to the product details page or advertising display position on the e-commerce platform. The generated display image has the technological feel of a smart speaker, presented in a home scene, and the visual effect highlights the hot-selling status of the product, attracting users' attention and purchasing interest. In this embodiment, through a conditional generative adversarial network model, the generator receives a conditional vector and random noise to generate an image that meets the display requirements. The generator uses a dynamic activation function, combined with the probability of the life cycle stage, to generate a display image with hot-selling period characteristics. The discriminator judges the generated image, and outputs the display image when it meets the requirements, realizing personalized and scenario-based product display.

[0105] This application also provides a generative artificial intelligence product display system based on image processing, such as... Figure 2 The image processing-based generative artificial intelligence product display system described in this application consists of various hardware components, which are interconnected via a network or bus to support the entire process from data acquisition and image generation to display. The following is a detailed description of the system's main hardware components and connection methods.

[0106] The main server is responsible for the core computing and management functions of the system, primarily including training the conditional generative adversarial network (GAN) model and inference for the generator and discriminator. The main server is equipped with high-performance CPUs, GPUs (such as NVIDIA Tesla or A100), and large-capacity memory to handle the complex computations of the deep learning model. The data server stores product data (such as listing time, sales data, click-through rates, etc.), user interaction data (such as click-through rate, dwell time, etc.), and generated display images. The data server is equipped with large-capacity storage space and high-speed read / write capabilities (such as SSD arrays) to ensure efficient data reading and writing. The main server and data server are connected via a high-speed local area network (such as gigabit Ethernet) to ensure data transmission speed and stability, enabling rapid data acquisition for model training and image generation. Smartphones, tablets, and PCs: Users can access the e-commerce platform or display system interface and view personalized product display images through various terminal devices. These devices typically include displays, network modules (Wi-Fi or cellular networks), and touchscreens (in smartphones and tablets) for easy user interaction. Front-end user devices connect to the server cluster via the Internet, and user browsing behavior and feedback data are transmitted to the server in real time for system feedback and optimization.

[0107] User behavior tracking device: Used to collect user browsing behavior on the platform, such as click-through rate, dwell time, and purchase history. This device is typically an embedded software system that can be directly embedded into the front-end application. Sales data acquisition system: Collects data such as product sales volume and inventory changes in real time. This system is usually an ERP system (such as SAP or Oracle) that can record product sales in real time. The data acquisition module transmits real-time data to the data server through interfaces or APIs, ensuring that the system can update user behavior and sales data in real time.

[0108] High-performance GPU accelerator cards (such as NVIDIA Tesla or A100) configured on the main server are used to perform image generation tasks for the Conditional Generative Adversarial Network (CGN) model. The GPU accelerates the model's inference process through high-parallel computing capabilities, generating high-quality display images. The processor (CPU) is responsible for scheduling tasks, inputting conditional vectors into the generator, and managing data transmission and the image generation process. The GPU and CPU are connected via a PCIe bus; the GPU handles model inference and image generation, while the CPU manages tasks and schedules data.

[0109] The main server and data server are equipped with SSD storage for rapid storage of generated display images. Storage devices hold generated images, user feedback data, and parameters of the training model, ensuring efficient data storage and retrieval. Cloud storage serves as backup storage, storing historical image data, user data, product display models, etc., providing higher redundancy and security for the system. Local storage connects to the server via SATA or NVMe interfaces, while cloud storage is synchronized and updated via the internet or a high-speed local area network. Display control devices and Content Delivery Network (CDN) nodes distribute the generated display images to various front-end user terminals, ensuring low latency when users access display images from different locations. CDN nodes can store personalized images generated by the system and load them quickly. The display device controller controls the display interface at the system's front end, displaying the generated display images, product information, and other content to users.

[0110] CDN nodes connect to the main server via the internet and to front-end user terminal devices for rapid image distribution. Display device controllers connect directly to the displays of front-end user terminal devices to display and switch images. The network module includes routers and switches to connect the various hardware components within a local area network (LAN), ensuring stable communication between server nodes, data acquisition modules, and image generation and processing units. Firewalls and security devices protect system data security, preventing unauthorized access and ensuring the security of user data and generated images. The network module connects all servers and terminal devices, providing a high-speed, low-latency network communication environment to guarantee secure and rapid data transmission. The server cluster (including the main server and data servers) is connected via a LAN. The main server handles model inference and control, while the data servers handle data storage and management. Front-end user devices access the system via the internet to request data and view displayed images. The data acquisition module connects to the data server via an API interface to transmit user behavior and sales data in real time. The GPU and CPU in the image generation and processing unit are connected via a PCIe bus. The generated display images are stored in local storage and distributed to front-end user devices via CDN nodes. The network module connects the entire system, ensuring stable communication between servers, terminal devices, and storage devices, and guaranteeing the overall efficient operation of the system. The system's hardware architecture achieves a complete closed loop from data acquisition and image generation to multi-platform display. Server clusters, high-performance GPUs, CDN nodes, and the network module together form a highly efficient generative AI product display system, supporting personalized and real-time product demonstrations.

[0111] The user display requirement input module receives the user's display requirement information, including product category, display style, and applicable scenario type;

[0112] The product lifecycle stage identification module automatically identifies the lifecycle stage of the corresponding product category based on the product's listing time and sales data. The lifecycle stages include the pre-heating, official release, hot-selling period, price reduction promotion, and clearance countdown stages.

[0113] The condition vector construction module constructs condition vectors based on user display requirements, lifecycle stages, and user interaction feedback data.

[0114] The image generation module takes conditional vectors as input to the conditional generative adversarial network model, and the generator generates display images corresponding to the display requirements based on the conditional vectors.

[0115] The output module displays the generated images.

[0116] In some embodiments, the user interaction feedback data includes historical click-through rates, browsing history, browsing dwell time, purchase history, and user reviews; the product categories include: clothing, home furnishings, electronic products, food, cosmetics, jewelry, sports equipment, and office supplies; the display styles include: minimalist style, retro style, modern style, luxury style, natural style, technological style, childlike style, and festive style; the applicable scene types include: indoor scenes, outdoor scenes, shopping mall scenes, home environment, office environment, exhibition environment, catering environment, and gym environment.

[0117] In some embodiments, the automatic identification of the product category's lifecycle stage based on the product's listing time and sales data includes:

[0118] The module for constructing the input feature vector X: constructs real-time feature data for product categories and constructs the input feature vector X. The feature data includes the listing time, sales volume, click volume, and inventory of products in the corresponding category.

[0119] Probability Calculation Module: Calculates the probability P(y=k|X) of a product belonging to each stage of its life cycle. The calculation formula is as follows:

[0120]

[0121] Where P(y=k|X) represents the probability that the product belongs to the k-th life cycle stage, X is the input feature vector, k represents the specific life cycle stage number, and K is the total number of life cycle stages; The weight vector β represents the weight vector corresponding to lifecycle stage k. T The dot product of the input feature vector X and exp is the natural exponential function, T is the transpose of the vector; j is a natural number, j∈(1,K); y is a parameter.

[0122] The module for determining the current product's lifecycle stage: Select the lifecycle stage with the highest probability value as the current product's lifecycle stage.

[0123] In some embodiments, constructing a condition vector based on the user's display needs information, lifecycle stage, and user interaction feedback data includes:

[0124] Stage Vector Generation Module: Encodes or embeds words for product category, display style, applicable scenario type, and product lifecycle stage respectively, generating category vector, style vector, scenario vector, and lifecycle stage vector;

[0125] User feedback vector generation module: Converts or normalizes the probability values ​​of click-through rate, browsing dwell time, and purchase records corresponding to the current product's lifecycle stage to generate user feedback vectors.

[0126] Vector concatenation module: The category vector, style vector, scene vector, lifecycle stage vector, and user feedback vector are concatenated end to end to form a complete conditional vector, which serves as the input to the generative adversarial network model.

[0127] In some embodiments, the step of inputting conditional vectors into a conditional generative adversarial network model, and the generator generating a display image corresponding to the display requirements based on the conditional vectors, includes:

[0128] Input module: Inputs the constructed conditional vector U and random noise vector z into the trained generator G;

[0129] Image construction module: The generator automatically generates a display image G(U,z) based on the condition vector U and the random noise vector z. This includes: the generator transforms the random noise based on the vector information of the condition vector U and maps it to a display image G(U,z) with conditional features; the discriminator D receives the image G(U,z) and the condition vector U generated by the generator and outputs a probability value to determine whether the generated image meets the set threshold to satisfy the requirements of the condition vector.

[0130] The generator's hidden layer uses a modified Sigmoid activation function, as shown below:

[0131]

[0132] Where, α k =P(y=k|X) represents the probability value of the current life cycle stage k, which serves as a dynamic factor for the Sigmoid activation function; X is the input feature vector for construction.

[0133] Conditional judgment output module: When the output probability value reaches the set threshold, the generated display image is output.

[0134] This invention provides a generative artificial intelligence product display method and system based on image processing, which can achieve the following beneficial technical effects:

[0135] 1. This application first receives the user's display requirement information; second, it automatically identifies the lifecycle stage of the corresponding product category based on the product's listing time and sales data; third, it constructs a condition vector based on the user's display requirement information, lifecycle stage, and user interaction feedback data; then, it inputs the condition vector into a conditional generative adversarial network model, and the generator generates a display image corresponding to the display requirement based on the condition vector; finally, it displays the generated display image. This application, by inputting the condition vector into a conditional generative adversarial network model and having the generator generate a display image corresponding to the display requirement based on the condition vector, greatly improves the accuracy and automation of generating images for user-required products, significantly enhancing the user experience.

[0136] 2. This invention firstly achieves precise capture of the characteristics of different lifecycle stages by incorporating probability values ​​of product lifecycle stages into the generator's activation function, making the displayed content more aligned with the current market environment. By calculating the probability values ​​of products at different lifecycle stages and introducing them as dynamic factors into the generator's activation function, the generator can automatically adjust the features of the output image to accurately adapt to the needs of different product lifecycle stages. This allows the displayed images generated during the pre-sale phase to highlight the product's uniqueness, demonstrate attractiveness during the peak sales period, and emphasize the urgency of the promotion during the clearance countdown phase, thereby effectively increasing user attention and product market performance. By combining user interaction feedback data into the conditional vector, more matching displayed images can be generated based on users' personalized needs, thereby improving user experience and product conversion rates. Furthermore, incorporating the probability values ​​of the lifecycle stage into the generator's activation function makes the generated displayed images more vivid and layered. For example, displayed images generated during the peak sales period may have higher visual appeal, encouraging users to buy, while displayed images generated during the clearance countdown phase have a sense of urgency, thus better stimulating users' purchasing decisions. This solution employs a conditional generative adversarial network model, which eliminates the need for static matching of scene libraries during the generation process. Instead, it dynamically generates content through a nonlinear mapping process of the generator, thereby greatly improving the flexibility and automation of content generation.

[0137] 3. This invention utilizes user interaction feedback data (such as click-through rate, browsing dwell time, purchase history, etc.) to construct conditional vectors, thereby generating display images that better meet user needs based on their preferences. The style and content of the generated images are highly matched with user preferences, significantly improving the personalization of the user experience and the accuracy of the display effect, thus increasing user retention and conversion rates. It eliminates the need for a static scene library, enhancing the automation and flexibility of display content: Unlike traditional display methods that rely on static matching based on a scene library, this application uses a conditional generative adversarial network generator to directly generate display images based on conditional vectors, avoiding the limitations of static content and greatly improving the flexibility of content generation. During the generation process, the generator dynamically generates content based on the probability of the input lifecycle stage and the user's personalized needs, automatically adapting to changes in market and user demands, significantly improving the automation and generation efficiency of product display.

[0138] The foregoing has provided a detailed description of a generative artificial intelligence product display method and system based on image processing. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas and methods of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A generative artificial intelligence product display method based on image processing, characterized in that, Including the following steps: S1: Receive user display requirements information, including product category, display style, and applicable scenario type; S2: Automatically identify the life cycle stage of the corresponding product category based on the product's listing time and sales data. The life cycle stage includes the pre-heating, official release, hot-selling period, price reduction promotion, and clearance countdown stage. S3: Construct a condition vector based on the user's display needs, lifecycle stage, and user interaction feedback data; S4: Input the conditional vector into the conditional generative adversarial network model, and the generator generates the display image corresponding to the display requirements based on the conditional vector; S5: Display the generated image.

2. The generative artificial intelligence product display method based on image processing as described in claim 1, characterized in that, The user interaction feedback data includes historical click-through rate, browsing history, browsing time, purchase history, and user reviews; The product categories include: clothing, home furnishings, electronic products, food, cosmetics, jewelry, sports equipment, and office supplies; the display styles include: minimalist, retro, modern, luxury, natural, technological, childlike, and festive; the applicable scene types include: indoor scenes, outdoor scenes, shopping mall scenes, home environments, office environments, exhibition environments, catering environments, and gym environments.

3. The generative artificial intelligence product display method based on image processing as described in claim 1, characterized in that, The automatic identification of the product category's lifecycle stage based on the product's listing time and sales data includes; S21: Construct real-time feature data for product categories and construct an input feature vector X. The feature data includes the listing time, sales volume, click volume, and inventory of products in the corresponding category. S22: Calculate the probability P(y=k|X) that the product belongs to each stage of its life cycle. The calculation formula is as follows: Where P(y=k|X) represents the probability that the product belongs to the k-th life cycle stage, X is the input feature vector, k represents the specific life cycle stage number, and K is the total number of life cycle stages; The weight vector β represents the weight vector corresponding to lifecycle stage k. T The dot product of the input feature vector X and exp is the natural exponential function, T is the transpose of the vector; j is a natural number, j∈(1,K); y is a parameter. S23: Select the lifecycle stage with the highest probability value as the lifecycle stage corresponding to the current product.

4. The generative artificial intelligence product display method based on image processing as described in claim 3, characterized in that, The construction of a condition vector based on user display needs, lifecycle stage, and user interaction feedback data includes: S31: Encode or embed words for product category, display style, applicable scenario type, and product life cycle stage respectively to generate category vector, style vector, scenario vector, and life cycle stage vector; S32: Numericalize or normalize the click-through rate, browsing dwell time, probability values ​​of the current product's corresponding lifecycle stage, and purchase records to generate user feedback vectors. S33: The category vector, style vector, scene vector, lifecycle stage vector, and user feedback vector are concatenated end to end to form a complete conditional vector, which serves as the input to the generative adversarial network model.

5. The generative artificial intelligence product display method based on image processing as described in claim 1, characterized in that, The step of inputting conditional vectors into a conditional generative adversarial network model, and the generator generating display images corresponding to the display requirements based on the conditional vectors, includes: S41: Input the constructed conditional vector U and random noise vector z into the trained generator G; S42: The generator automatically generates a display image G(U,z) based on the condition vector U and the random noise vector z, including: the generator transforms the random noise based on the vector information of the condition vector U and maps it to a display image G(U,z) with conditional features; the discriminator D receives the image G(U,z) and the condition vector U generated by the generator and outputs a probability value to determine whether the generated image meets the set threshold to meet the requirements of the condition vector. The generator's hidden layer uses a modified Sigmoid activation function, as shown below: Where, α k =P(y=k|X) represents the probability value of the current life cycle stage k, which serves as a dynamic factor for the Sigmoid activation function; X is the input feature vector for construction. S43: When the output probability value reaches the set threshold, output the generated display image.

6. A generative artificial intelligence product display system based on image processing, characterized in that, include: The user display requirement input module receives the user's display requirement information, including product category, display style, and applicable scenario type; The product lifecycle stage identification module automatically identifies the lifecycle stage of the corresponding product category based on the product's listing time and sales data. The lifecycle stages include the pre-heating, official release, hot-selling period, price reduction promotion, and clearance countdown stages. The condition vector construction module constructs condition vectors based on user display requirements, lifecycle stages, and user interaction feedback data. The image generation module takes conditional vectors as input to the conditional generative adversarial network model, and the generator generates display images corresponding to the display requirements based on the conditional vectors. The output module displays the generated images.

7. The generative artificial intelligence product display system based on image processing as described in claim 6, characterized in that, The user interaction feedback data includes historical click-through rate, browsing history, browsing time, purchase history, and user reviews; The product categories include: clothing, home furnishings, electronic products, food, cosmetics, jewelry, sports equipment, and office supplies; the display styles include: minimalist, retro, modern, luxury, natural, technological, childlike, and festive; the applicable scene types include: indoor scenes, outdoor scenes, shopping mall scenes, home environments, office environments, exhibition environments, catering environments, and gym environments.

8. The generative artificial intelligence product display system based on image processing as described in claim 6, characterized in that, The automatic identification of the product category's lifecycle stage based on the product's listing time and sales data includes; The module for constructing the input feature vector X: constructs real-time feature data for product categories and constructs the input feature vector X. The feature data includes the listing time, sales volume, click volume, and inventory of products in the corresponding category. Probability Calculation Module: Calculates the probability P(y=k|X) of a product belonging to each stage of its life cycle. The calculation formula is as follows: Where P(y=k|X) represents the probability that the product belongs to the k-th life cycle stage, X is the input feature vector, k represents the specific life cycle stage number, and K is the total number of life cycle stages; The weight vector β represents the weight vector corresponding to lifecycle stage k. T The dot product of the input feature vector X and exp is the natural exponential function, T is the transpose of the vector; j is a natural number, j∈(1,K); y is a parameter. The module for determining the current product's lifecycle stage: Select the lifecycle stage with the highest probability value as the current product's lifecycle stage.

9. The generative artificial intelligence product display system based on image processing as described in claim 8, characterized in that, The construction of a condition vector based on user display needs, lifecycle stage, and user interaction feedback data includes: Stage Vector Generation Module: Encodes or embeds words for product category, display style, applicable scenario type, and product lifecycle stage respectively, generating category vector, style vector, scenario vector, and lifecycle stage vector; User feedback vector generation module: Converts or normalizes the probability values ​​of click-through rate, browsing dwell time, and purchase records corresponding to the current product's lifecycle stage to generate user feedback vectors. Vector concatenation module: The category vector, style vector, scene vector, lifecycle stage vector, and user feedback vector are concatenated end to end to form a complete conditional vector, which serves as the input to the generative adversarial network model.

10. The generative artificial intelligence product display system based on image processing as described in claim 6, characterized in that, The step of inputting conditional vectors into a conditional generative adversarial network model, and the generator generating display images corresponding to the display requirements based on the conditional vectors, includes: Input module: Inputs the constructed conditional vector U and random noise vector z into the trained generator G; Image construction module: The generator automatically generates a display image G(U,z) based on the condition vector U and the random noise vector z. This includes: the generator transforms the random noise based on the vector information of the condition vector U and maps it to a display image G(U,z) with conditional features; the discriminator D receives the image G(U,z) and the condition vector U generated by the generator and outputs a probability value to determine whether the generated image meets the set threshold to satisfy the requirements of the condition vector. The generator's hidden layer uses a modified Sigmoid activation function, as shown below: Where, α k =P(y=k|X) represents the probability value of the current life cycle stage k, which serves as a dynamic factor for the Sigmoid activation function; X is the input feature vector for construction. Conditional judgment output module: When the output probability value reaches the set threshold, the generated display image is output.

Images