An e-commerce multi-category commodity data expansion method, device, equipment and medium
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- XIAMEN ZIXUN INFORMATION TECHNOLOGY CO LTD
- Filing Date
- 2026-01-31
- Publication Date
- 2026-06-02
Smart Images
Figure CN122135134A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of image data processing technology, and in particular to a method, apparatus, equipment, and medium for expanding e-commerce multi-category product data. Background Technology
[0002] In e-commerce, the diversity and richness of product image data have a significant impact on product display, recommendation systems, and user purchasing decisions. However, many e-commerce categories face the problem of scarce product image data, especially newly launched or niche categories. Traditional data augmentation methods, such as image rotation and flipping, can only increase the amount of data within a limited scope and cannot generate new category images. The existing technology has the following problems: Insufficient data diversity: Traditional data augmentation methods cannot generate entirely new product images; they can only perform simple transformations on existing images. Poor product category adaptability: General image generation models struggle to generate images that meet specific requirements for different e-commerce product categories. Lack of quality assessment standards: The lack of an effective assessment mechanism to screen high-quality generated images results in inconsistent quality of generated data. Text rendering issues: In e-commerce product images, text content such as product labels and price information often appears blurry, affecting the product display effect. Summary of the Invention
[0003] The technical problem to be solved by the present invention is to provide a method, device, equipment and medium for expanding e-commerce multi-category product data, which realizes high-quality and diversified e-commerce product image expansion through multi-category adaptation mechanism and multi-dimensional scoring mechanism. In a first aspect, the present invention provides a method for expanding e-commerce multi-category product data, comprising the following steps: S10. Obtain product category information; S20. Construct a prompt word template based on the product category information; S30. Input the prompt word template into the Qwen-Image model to generate a product image; S40. Perform multi-dimensional scoring on the generated product image and output a comprehensive score; S50. Based on the comprehensive score, high-quality images are obtained as supplementary data. Secondly, the present invention provides an e-commerce multi-category product data expansion device, comprising the following modules: The information acquisition module retrieves product category information; The prompt word module constructs prompt word templates based on the product category information; The image generation module inputs the prompt word template into the Qwen-Image model to generate a product image; The scoring module performs multi-dimensional scoring on the generated product images and outputs a comprehensive score. The output filtering module filters images based on the comprehensive score to obtain high-quality images as supplementary data. Thirdly, the present invention provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described in the first aspect. Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in the first aspect. One or more technical solutions provided by this invention have at least the following technical effects or advantages: Multi-category adaptation: Through targeted prompt word template design, it can adapt to multiple categories of e-commerce and generate product images that conform to the characteristics of the categories. High-quality generation: Based on the Qwen-Image large model, the high-quality image generation capability ensures the clarity and realism of the generated images, especially the text rendering quality. Effective evaluation: Through a multi-dimensional scoring mechanism, the quality of generated images is scientifically evaluated, and high-quality data is selected. High data richness: The generated images cover a variety of styles, layouts and scenes, greatly enriching e-commerce product data. Text rendering optimization: By utilizing Qwen-Image's MSRoPE technology, the rendering effect of text in images has been specially optimized, solving the problem of blurry and unclear text in e-commerce product images. The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description The present invention will be further described below with reference to the accompanying drawings and embodiments. Figure 1 This is a flowchart of the method in Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the device in Embodiment 2 of the present invention. Detailed Implementation This application provides a method, apparatus, device, and medium for expanding e-commerce multi-category product data. Through a multi-category adaptation mechanism and a multi-dimensional scoring mechanism, it achieves high-quality and diversified e-commerce product image expansion. Example 1 like Figure 1 As shown in the figure, this embodiment provides a method for expanding e-commerce multi-category product data, including the following steps: S10. Obtain product category information; The product category information includes: category name, category characteristics, target user group, and set display scenario; The prompt word template is used to convert the product category information into a description that can be understood by the Qwen-Image model; S20. Construct a prompt word template based on the product category information; S30. Input the prompt word template into the Qwen-Image model to generate a product image; S40. Perform multi-dimensional scoring on the generated product image and output a comprehensive score; The multi-dimensional scoring specifically includes: Image sharpness score S1: S1=1-(1 / √(1 + 0.5*σ²)), where σ is the degree of blurriness of the generated product image; the method for obtaining σ² is: calculate the gradient magnitude of the generated product image, calculate the statistical variance of the gradient magnitude, and obtain σ². Category matching score S2: S2=1-|(P - P0)| / P0, where P is the matching degree of category features in the generated product image, and P0 is the preset ideal value of category features; the method of obtaining P is: using an image recognition model to extract feature vectors of the generated product image, calculating the similarity between the feature vectors and the set reference vectors, and the similarity is the matching degree of category features; Visual appeal score S3: S3 = 0.3H1 + 0.4C + 0.3*S7, where H1 is the color saturation of the image, C is the color contrast, and S7 is the image composition score; the method for obtaining S7 is to use a deep learning model to evaluate the composition quality of the generated product image. Background interference score S4: S4=1-(B / (W*H2)), where B is the number of pixels in the background interference area, W is the width of the generated product image, and H2 is the height of the generated product image; Product subject visibility score S5: S5=(AI) / A, where A is the area of the product subject in the generated product image, and I is the area of the occluded product subject in the generated product image; Text clarity score S6: S6 = 1-(1 / √(1+0.3*E²)), where E is the accuracy of the text in the image; the method for obtaining E is: using an OCR tool to extract the text in the generated product image, comparing it with the set expected text, and calculating the accuracy of the extracted text in the generated product image. Calculate the overall score S: S = 0.2S1 + 0.25S2 + 0.25S3 + 0.1S4 + 0.1S5 + 0.1S6; S50. Based on the comprehensive score, high-quality images are selected as supplementary data. Specifically, the generated product images with a comprehensive score greater than or equal to a set threshold are selected as high-quality images, and these high-quality images are used as supplementary data. Based on the same inventive concept, this application also provides an apparatus corresponding to the method in Embodiment 1, as detailed in Embodiment 2. Example 2 like Figure 2 As shown in the figure, this embodiment provides an e-commerce multi-category product data expansion device, including the following modules: The information acquisition module retrieves product category information; The prompt word module constructs prompt word templates based on the product category information; The image generation module inputs the prompt word template into the Qwen-Image model to generate a product image; The scoring module performs multi-dimensional scoring on the generated product images and outputs a comprehensive score. The output filtering module filters images based on the comprehensive score to obtain high-quality images as supplementary data. In the information acquisition module: The product category information includes: category name, category characteristics, target user group, and set display scenario; The prompt word template is used to convert the product category information into a description that can be understood by the Qwen-Image model; The multi-dimensional scoring in the scoring module specifically includes: Image sharpness score S1: S1=1-(1 / √(1 + 0.5*σ²)), where σ is the degree of blurriness of the generated product image; the method for obtaining σ² is: calculate the gradient magnitude of the generated product image, calculate the statistical variance of the gradient magnitude, and obtain σ². Category matching score S2: S2=1-|(P - P0)| / P0, where P is the matching degree of category features in the generated product image, and P0 is the preset ideal value of category features; the method of obtaining P is: using an image recognition model to extract feature vectors of the generated product image, calculating the similarity between the feature vectors and the set reference vectors, and the similarity is the matching degree of category features; Visual appeal score S3: S3 = 0.3H1 + 0.4C + 0.3*S7, where H1 is the color saturation of the image, C is the color contrast, and S7 is the image composition score; the method for obtaining S7 is to use a deep learning model to evaluate the composition quality of the generated product image. Background interference score S4: S4=1-(B / (W*H2)), where B is the number of pixels in the background interference area, W is the width of the generated product image, and H2 is the height of the generated product image; Product subject visibility score S5: S5=(AI) / A, where A is the area of the product subject in the generated product image, and I is the area of the occluded product subject in the generated product image; Text clarity score S6: S6 = 1-(1 / √(1+0.3*E²)), where E is the accuracy of the text in the image; the method for obtaining E is: using an OCR tool to extract the text in the generated product image, comparing it with the set expected text, and calculating the accuracy of the extracted text in the generated product image. Calculate the overall score S: S = 0.2S1 + 0.25S2 + 0.25S3 + 0.1S4 + 0.1S5 + 0.1S6; The filtering output module is specifically: The generated product images with a comprehensive score greater than or equal to a set threshold are considered high-quality images, and these high-quality images are used as supplementary data. Since the apparatus described in Embodiment 2 of the present invention is an apparatus used to implement the method of Embodiment 1 of the present invention, those skilled in the art can understand the specific structure and variations of the apparatus based on the method described in Embodiment 1 of the present invention, and therefore will not be described again here. All apparatuses used in the method of Embodiment 1 of the present invention fall within the scope of protection of the present invention. Example 3 The overall concept of the technical solution in Embodiment 3 of this application is as follows: This invention provides a method for expanding e-commerce multi-category product data based on Qwen-Image, including: acquiring product category information; constructing specific prompt word templates based on the product category information; generating product images using the Qwen-Image model; evaluating the quality of the generated images based on a multi-dimensional scoring mechanism; and selecting high-quality images as expansion data based on the scoring results. This invention provides an e-commerce multi-category product data expansion system based on Qwen-Image, including: a product category information acquisition module, a prompt word template construction module, a Qwen-Image image generation module, a multi-dimensional scoring module, and a data filtering module. The overall approach is as follows: Product category information acquisition → prompt word template construction → Qwen-Image image generation → multi-dimensional scoring → data filtering → e-commerce product data expansion; The specific steps include: (1) Obtaining product category information Obtain category information for e-commerce products, including category name, category characteristics, target user group, and common display scenarios. For example, the characteristics of the "Women's Clothing - Dresses" category include "elegant, fashionable, and suitable for spring," the target user is "women aged 25-35," and the common display scenarios are "outdoors and offices." (2) Construction of prompt word templates Based on product category information, construct a prompt word template that includes the following elements: Category characteristics: Describes the core features of this category. Style Description: Describe the desired image style (e.g., minimalist, retro). Product description: Describe the specific features of the product (e.g., "dress length reaches the knee, lace embellishment"). Background elements: Describe background elements (such as "spring garden" or "city street"). Technical specifications: Describe the image requirements (e.g., "1080×1080 pixels, high resolution") Quality requirements: Describe the image quality requirements (e.g., "sharp edges, no blurry text") For example, the prompt template for the category "Women's Clothing - Dresses" is: "Spring fashion dress, elegant and simple style, knee-length, lace embellishment, spring garden background, 1080×1080 pixels, high resolution, sharp edges, no blurry text." The prompt word template is the result of expanding and specifying product category information (especially category characteristics, target user groups and common display scenarios), transforming abstract category information into specific descriptions that can be understood by the Qwen-Image model. (3) Qwen-Image image generation Input the prompt template into the Qwen-Image model to generate product images. Based on its "three-in-one" architecture (MLLM, enhanced VAE, and MMDit) and MSRoPE technology, the Qwen-Image model can understand complex prompts and generate high-quality images, with particularly optimized text rendering effects. (4) Multidimensional scoring The generated images are scored in multiple dimensions, including: Image sharpness score: S1 = 1-(1 / √(1 + 0.5*σ²)), where σ is the degree of blur in the image (calculated from the image gradient).
[0004] The σ is calculated through image gradient: the gradient magnitude of the image is calculated using an image gradient operator, and the variance of the gradient magnitude is calculated to obtain σ²; the gradient operator with required parameters is obtained from multiple inputs such as the original product image, operator size, and gradient direction fusion; the calculation formula is:
[0005] Category matching score: S2 = 1 - |(P - P0)| / P0, where P is the matching degree of category features in the generated image, and P0 is the ideal value of category features. Visual appeal score: S3 = 0.3H + 0.4C + 0.3*S, where H is the color saturation of the image, C is the color contrast, and S is the image composition score. Background interference score: S4 = 1 - (B / (W*H)), where B is the number of pixels in the background interference area, and W and H are the image width and height. Product visibility score: S5 = (A - I) / A, where A is the area of the product's main body and I is the area of the product's main body that is obscured. Text clarity score: S6 = 1 - (1 / √(1 + 0.3*E²)), where E is the degree of blurriness of the text in the image, specifically optimized for Qwen-Image's MSRoPE technology. (5) Data filtering Calculate the overall score: S = 0.2S1 + 0.25S2 + 0.25S3 + 0.1S4 + 0.1S5 + 0.1S6; Set a threshold T (e.g., 0.75), and use images with a comprehensive score ≥ T as high-quality augmentation data. The multi-dimensional scoring is as follows: 1. Image sharpness score (S1) formula:
[0006] Parameter description: σ: The degree of blur in the image, calculated from the image gradient. Calculation method: The gradient magnitude of an image is calculated using the Laplacian gradient operator. The variance of the gradient magnitude is calculated to obtain σ² (the larger the variance, the blurrier the image). Example: If σ² = 4, then
[0007] formula: Parameter description: P: The actual matching degree of category features in the generated image (calculated by the image recognition model).
[0008] P0: The ideal value for category characteristics (usually set to 1, indicating a perfect match).
[0009] Calculation method: Image recognition models (CNNs) are used to extract feature vectors from generated images. CNNs do not directly output P; they only generate the raw materials (feature vectors) for calculating P. P is the final result calculated by comparing the similarity between the feature vectors and the reference vectors. Calculate the actual matching degree P If P = 0.8, then (for example, by measuring the similarity between image features and target category features using cosine similarity).
[0010] Example:
[0011] If the actual matching degree 3. Visual Appeal Rating (S3) formula:
[0012] Parameter description: H: Color saturation (0~1). Calculation method: Convert the image to the HSV color space and calculate the average value of the V channel (luminance). C: Color contrast (0~1). Calculation method: Calculate the brightness difference between bright colors (high brightness areas) and dark colors (low brightness areas) in the image. S: Image composition score (0~1). Computational method: Deep learning models (such as CNN) are used to evaluate the quality of the graph.
[0013] If H = 0.9, C = 0.8, S = 0.7, then:
[0014] 4. Background Interference Score (S4) formula:
[0015] Parameter description: B: Pixel count of background interference areas (background interference areas are those background parts in e-commerce product images that are similar to product features and affect product recognition) W and H: The width and height of the image (in pixels). Calculation method: Use semantic segmentation models (such as SAM) to distinguish between the product subject and the background. The number of pixels B in the interference region of the statistical background Example: If the image resolution is 1080×1080, and the number of pixels in the background interference area is B=100000, then
[0016] 5. Product visibility score (S5) formula:
[0017] Parameter description: A: Total area (in pixels) of the main product area I: Number of pixels in the area where the product is obscured Calculation method: Use object detection models (such as YOLO) to identify the main body area of the product; The number of pixels I that are occluded in the main area is counted; Example: If the area of the main product area A = 500,000 and the obscured area I = 50,000, then
[0018] 6. Text clarity score (S6) formula:
[0019] Parameter description: E: The degree of blurriness of the text in the image (calculated using OCR or blur detection algorithms). Calculation method: Use OCR tools (such as Tesseract) to extract text from images. E is calculated based on character recognition accuracy. E is not obtained by directly inputting an image; rather, it requires OCR to extract text and compare it with reference text for calculation. Character recognition accuracy is an evaluation metric after OCR processing, not a direct output function of OCR. Product image → OCR text extraction → Comparison with expected text → Calculation of accuracy → Obtaining E; For example: Referring to "Qwen-Image", recognizing "Qwen-Imge" → 9 out of 10 characters are correct → accuracy = 0.9; Example: If E = 2, then
[0020] The S6 rating system is specifically adapted to the characteristics of text generated by MSRoPE technology from three aspects: "fuzziness calculation, formula coefficient, and evaluation dimensions". This ensures that the clarity of e-commerce product text generated by this technology can be accurately measured, neither underestimating its text quality advantages nor ignoring real fuzziness issues. Comprehensive score calculation Formula: S = 0.2S1 + 0.25S2 + 0.25S3 + 0.1S4 + 0.1S5 + 0.1S6; The overall score is: Assume the scores are as follows: S1=0.423, S2=0.8, S3=0.8, S4=0.915, S5=0.9, S6=0.326;
[0021] (6) Data application The selected high-quality images are applied to scenarios such as e-commerce product display and recommendation system training. Detailed explanation of the multi-dimensional scoring mechanism: Image sharpness rating: The degree of blurriness is assessed by calculating the gradient magnitude of the image; the larger the σ, the more blurred the image. The formula S1 = 1 - (1 / √(1 + 0.5*σ²)) ensures that the higher the degree of blurriness, the lower the rating. Category matching score: By comparing the similarity between the generated image and the category features, P0 is the ideal value of the category features (usually 1), and P is the actual matching degree (calculated by the image recognition model). Visual appeal score: Combining color saturation, contrast and composition scores, the weighting takes into account the visual appeal elements of e-commerce images. Background interference score: assesses the degree of interference of background elements with the main product. B is the number of pixels in the background interference area, and W*H is the total number of pixels in the image. Product visibility score: assesses whether the product is obscured. A is the area of the product's main area, and I is the area of the obscured area. Text clarity score: Specifically designed for text information in e-commerce product images, E represents the degree of text blur, S6 = 1 - (1 / √(1 + 0.3*E²)), ensuring that images with high text clarity receive higher scores. For the "Home Appliances - Smartwatch" category, a prompt template was created: "Smartwatch, fashionable and technological style, round dial, metal strap, outdoor sports scene, 1080×1080 pixels, high resolution, clear edges, and no blurry text". After generating images, they were scored, and images with a comprehensive score of ≥0.8 were selected as supplementary data. For the "beauty - lipstick" category, a prompt template was constructed: "lipstick, exquisite and elegant style, red, lipstick texture, dressing table background, 1080×1080 pixels, high resolution, clear edges, and no blurry text". After generating images, they were scored, and images with a comprehensive score of ≥0.75 were selected as supplementary data. For the "food-chocolate" category, a prompt template was constructed: "chocolate, exquisite dessert style, handmade, gift box packaging, coffee shop background, 1080×1080 pixels, high resolution, clear edges, and no blurry text". After generating images, they were scored, and images with a comprehensive score of ≥0.8 were selected as supplementary data. Based on the same inventive concept, this application provides an electronic device embodiment corresponding to Embodiment 1, as detailed in Embodiment 4. Example 4 This embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it can implement any of the implementation methods in Embodiment 1. Since the electronic device described in this embodiment is the device used to implement the method in Embodiment 1 of this application, those skilled in the art can understand the specific implementation method and various variations of the electronic device in this embodiment based on the method described in Embodiment 1 of this application. Therefore, how the electronic device implements the method in the embodiment of this application will not be described in detail here. Any device used by those skilled in the art to implement the method in the embodiment of this application falls within the scope of protection of this application. Based on the same inventive concept, this application provides a storage medium corresponding to Embodiment 1, as detailed in Embodiment 5. Example 5 This embodiment provides a computer-readable storage medium storing a computer program thereon. When the computer program is executed by a processor, it can implement any of the implementation methods in Embodiment 1. The technical solutions provided in this application embodiment have at least the following technical effects or advantages: 1. Using Qwen-Image as the core generation model, and leveraging its MSRoPE technology to achieve higher quality text rendering, it is particularly suitable for text information in e-commerce product images. 2. A method for constructing prompt word templates for multiple product categories in e-commerce was designed, which can generate high-quality images based on the characteristics of different product categories. 3. A multi-dimensional scoring mechanism including text clarity scoring is proposed to ensure the quality of generated images. 4. Innovatively combining Qwen-Image's MSRoPE technology with e-commerce product image generation improves the clarity and readability of text in product images. Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes. While specific embodiments of the present invention have been described above, those skilled in the art should understand that the specific embodiments described are merely illustrative and not intended to limit the scope of the present invention. Equivalent modifications and variations made by those skilled in the art in accordance with the spirit of the present invention should be covered within the scope of protection of the claims of the present invention.
Claims
1. A method for expanding e-commerce multi-category product data, characterized in that: Includes the following steps: S10. Obtain product category information; S20. Construct a prompt word template based on the product category information; S30. Input the prompt word template into the Qwen-Image model to generate a product image; S40. Perform multi-dimensional scoring on the generated product image and output a comprehensive score; S50. Based on the comprehensive score, high-quality images are obtained as supplementary data.
2. The method for expanding e-commerce multi-category product data according to claim 1, characterized in that: In S10: The product category information includes: category name, category characteristics, target user group, and set display scenario; The prompt word template is used to transform the product category information into a description that can be understood by the Qwen-Image model.
3. The method for expanding e-commerce multi-category product data according to claim 1, characterized in that: The multi-dimensional scoring in S40 specifically includes: Image sharpness score S1: S1=1-(1 / √(1 + 0.5*σ²)), where σ is the degree of blurriness of the generated product image; the method for obtaining σ² is: calculate the gradient magnitude of the generated product image, calculate the statistical variance of the gradient magnitude, and obtain σ². Category matching score S2: S2=1-|(P - P0)| / P0, where P is the matching degree of category features in the generated product image, and P0 is the preset ideal value of category features; the method of obtaining P is: using an image recognition model to extract feature vectors of the generated product image, calculating the similarity between the feature vectors and the set reference vectors, and the similarity is the matching degree of category features; Visual appeal score S3: S3 = 0.3H1 + 0.4C + 0.3*S7, where H1 is the color saturation of the image, C is the color contrast, and S7 is the image composition score; the method for obtaining S7 is to use a deep learning model to evaluate the composition quality of the generated product image. Background interference score S4: S4=1-(B / (W*H2)), where B is the number of pixels in the background interference area, W is the width of the generated product image, and H2 is the height of the generated product image; Product subject visibility score S5: S5=(AI) / A, where A is the area of the product subject in the generated product image, and I is the area of the occluded product subject in the generated product image; Text clarity score S6: S6 = 1-(1 / √(1+0.3*E²)), where E is the accuracy of the text in the image; the method for obtaining E is: using an OCR tool to extract the text in the generated product image, comparing it with the set expected text, and calculating the accuracy of the extracted text in the generated product image. Calculate the overall score S: S = 0.2S1 + 0.25S2 + 0.25S3 + 0.1S4 + 0.1S5 + 0.1S6.
4. The method for expanding e-commerce multi-category product data according to claim 1, characterized in that: Specifically, S50 is: The generated product images with a comprehensive score greater than or equal to a set threshold are considered high-quality images, and these high-quality images are used as supplementary data.
5. A data expansion device for multi-category e-commerce products, characterized in that: Includes the following modules: The information acquisition module retrieves product category information; The prompt word module constructs prompt word templates based on the product category information; The image generation module inputs the prompt word template into the Qwen-Image model to generate a product image; The scoring module performs multi-dimensional scoring on the generated product images and outputs a comprehensive score. The output filtering module filters images based on the comprehensive score to obtain high-quality images as supplementary data.
6. The e-commerce multi-category product data expansion device according to claim 5, characterized in that: In the information acquisition module: The product category information includes: category name, category characteristics, target user group, and set display scenario; The prompt word template is used to transform the product category information into a description that can be understood by the Qwen-Image model.
7. The e-commerce multi-category product data expansion device according to claim 5, characterized in that: The multi-dimensional scoring in the scoring module specifically includes: Image sharpness score S1: S1=1-(1 / √(1 + 0.5*σ²)), where σ is the degree of blurriness of the generated product image; the method for obtaining σ² is: calculate the gradient magnitude of the generated product image, calculate the statistical variance of the gradient magnitude, and obtain σ². Category matching score S2: S2=1-|(P - P0)| / P0, where P is the matching degree of category features in the generated product image, and P0 is the preset ideal value of category features; the method of obtaining P is: using an image recognition model to extract feature vectors of the generated product image, calculating the similarity between the feature vectors and the set reference vectors, and the similarity is the matching degree of category features; Visual appeal score S3: S3 = 0.3H1 + 0.4C + 0.3*S7, where H1 is the color saturation of the image, C is the color contrast, and S7 is the image composition score; the method for obtaining S7 is to use a deep learning model to evaluate the composition quality of the generated product image. Background interference score S4: S4=1-(B / (W*H2)), where B is the number of pixels in the background interference area, W is the width of the generated product image, and H2 is the height of the generated product image; Product subject visibility score S5: S5=(AI) / A, where A is the area of the product subject in the generated product image, and I is the area of the occluded product subject in the generated product image; Text clarity score S6: S6 = 1-(1 / √(1+0.3*E²)), where E is the accuracy of the text in the image; the method for obtaining E is: using an OCR tool to extract the text in the generated product image, comparing it with the set expected text, and calculating the accuracy of the extracted text in the generated product image. Calculate the overall score S: S = 0.2S1 + 0.25S2 + 0.25S3 + 0.1S4 + 0.1S5 + 0.1S6.
8. The e-commerce multi-category product data expansion device according to claim 5, characterized in that: The filtering output module is specifically: The generated product images with a comprehensive score greater than or equal to a set threshold are considered high-quality images, and these high-quality images are used as supplementary data.
9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 4.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 4.