Package design optimization system and method based on big data analysis

Abstract

The invention relates to the technical field of big data analysis, and more particularly discloses a package design optimization system and method based on big data analysis, which analyze and optimize package design by using a product image and a package design image. The attractive force and the spreading effect of product packaging are improved in the modes of color adjustment, pattern optimization, information transmission improvement and the like, so that the high requirement of consumers for visual experience is met, and a packaging design scheme with better attractive force and information transmission effect is provided for the fields of beauty makeup commodities and the like.

Description

Technical Field

[0001] The present application relates to the technical field of big data analysis, and more specifically, to a packaging design optimization system and method based on big data analysis. Background Art

[0002] With the rapid development of mobile networks, people's demands for a higher quality of life are becoming increasingly stringent. Product packaging on beauty shelves must capture consumers' attention, enhance their focus, and enhance their perception. This approach maximizes the use of simple graphics and clean color blocks, ensuring a clear and distinct packaging style, optimizing the consumer visual experience and effectively conveying product information in a short period of time. This design concept, which prioritizes function over form, plays a crucial role in packaging design. It emphasizes function over form and plays a significant role in packaging design. Due to its long-standing application in packaging design, it has gradually become a prominent feature in the field of visual design. Its unique style and artistic appeal attract young people through its intuitive communication, inspiring designers to create more valuable products and providing valuable insights for commercial product promotion.

[0003] Therefore, a packaging design optimization system and method based on big data analysis is desired. Summary of the Invention

[0004] This application is proposed to solve the above technical problems. The embodiments of this application provide a packaging design optimization system and method based on big data analysis, which analyzes and adjusts product packaging design to enhance packaging appeal and promotional effects, meet consumer needs, and promote product promotion.

[0005] Accordingly, according to one aspect of the present application, a packaging design optimization system based on big data analysis is provided, which includes:

[0006] Packaging design image acquisition module, used to acquire product images and packaging design images;

[0007] a packaging design image processing module for extracting a packaging design image block global-local feature vector and a product image block feature vector from the packaging design image and the packaging design image, respectively;

[0008] A packaging design image fusion module is used to fuse and modulate the global-local feature vector of the packaging design image block and the feature vector of the product image block to obtain an optimized product packaging design analysis feature vector;

[0009] The packaging design image analysis module is used to pass the optimized product packaging design analysis feature vector through a classifier to obtain a classification result, and the classification result is used to indicate whether the product packaging design meets the standards.

[0010] According to another aspect of the present application, a packaging design optimization method based on big data analysis is provided, which includes:

[0011] Get product images and packaging design images;

[0012] extracting a package design image block global-local feature vector and a product image block feature vector from the package design image and the package design image, respectively;

[0013] fusing and modulating the global-local feature vector of the packaging design image block and the feature vector of the product image block to obtain an optimized product packaging design analysis feature vector;

[0014] The optimized product packaging design analysis feature vector is passed through a classifier to obtain a classification result, and the classification result is used to indicate whether the product packaging design meets the standards.

[0015] Compared with the existing technology, the present application provides a packaging design optimization system and method based on big data analysis, which uses product images and packaging design images to analyze and optimize packaging design, and enhances the attractiveness and communication effect of product packaging by adjusting colors, optimizing patterns, improving information communication, etc., so as to meet consumers' high requirements for visual experience and provide more attractive and information-communicating packaging design solutions for beauty products and other fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are intended to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0017] Figure 1 Schematic diagram of a packaging design optimization system based on big data analysis according to an embodiment of the present application.

[0018] Figure 2 This is a block diagram of a packaging design image processing module in a packaging design optimization system based on big data analysis according to an embodiment of the present application.

[0019] Figure 3 Schematic diagram of a block diagram of a packaging image processing unit in a packaging design optimization system based on big data analysis according to an embodiment of the present application.

[0020] Figure 4 Schematic diagram of a block diagram of a product image processing unit in a packaging design optimization system based on big data analysis according to an embodiment of the present application.

[0021] Figure 5 Flowchart of a packaging design optimization method based on big data analysis according to an embodiment of the present application. DETAILED DESCRIPTION

[0022] Various exemplary embodiments, features, and aspects of the present application will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise indicated.

[0023] The word “exemplary” is used exclusively herein to mean “serving as an example, example, or illustration.” Any embodiment described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments.

[0024] In addition, numerous specific details are provided in the following detailed description to better illustrate the present application. Those skilled in the art will appreciate that the present application can be practiced without certain specific details. In some instances, methods, means, components, and circuits well known to those skilled in the art are not described in detail in order to highlight the main purpose of the present application.

[0025] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. Throughout the description of this application, "plurality" means two or more, unless otherwise specifically defined.

[0026] Figure 1 FIG2 shows a block diagram of a packaging design optimization system based on big data analysis according to an embodiment of the present application. Figure 1 As shown, according to an embodiment of the present application, a packaging design optimization system 100 based on big data analysis includes: a packaging design image acquisition module 110, used to acquire product images and packaging design images; a packaging design image processing module 120, used to extract packaging design image block global-local feature vectors and product image block feature vectors from the packaging design image and the packaging design image, respectively; a packaging design image fusion module 130, used to fuse and modulate the packaging design image block global-local feature vector and the product image block feature vector to obtain an optimized product packaging design analysis feature vector; a packaging design image analysis module 140, used to pass the optimized product packaging design analysis feature vector through a classifier to obtain a classification result, and the classification result is used to indicate whether the product packaging design meets the standard.

[0027] In an embodiment of the present application, the packaging design image acquisition module 110 is used to obtain product images and packaging design images. It should be understood that the packaging design drawing shows the appearance and design features of the product packaging, including the color, pattern, text information, etc. of the packaging. Packaging design greatly affects consumers' first impression of the product, can convey the brand image, positioning and characteristics of the product, and arouse consumers' interest and desire to buy. By combining product images and packaging design drawings, the overall effect of product packaging can be more comprehensively evaluated. Product images provide information about the product itself, while packaging design drawings show the appearance and design features of the product packaging. Combining the two can help analyze whether the packaging design matches the product itself, whether it can attract the target consumer group, and whether it is consistent with the brand image and market positioning. By comprehensively analyzing product images and packaging design drawings, potential problems in packaging design can be discovered and improvement suggestions can be made to better meet consumer needs and improve the effectiveness of product promotion. Optimizing packaging design can enhance the market competitiveness of products, increase consumers' desire to buy, and promote product sales and brand promotion.

[0028] In this embodiment of the present application, the packaging design image processing module 120 is configured to extract global-local feature vectors of the packaging design image block and feature vectors of the product image block from the packaging design image and the packaging design image, respectively. It should be understood that extracting global-local feature vectors of the packaging design image block can help analyze the overall structure and local details of the packaging design. Global feature vectors can capture information such as overall layout, color matching, and theme, while local feature vectors can capture features such as local patterns, details, and text. By combining global and local feature vectors, a more comprehensive assessment of the visual effects and design characteristics of the packaging design can be achieved. Extracting feature vectors of the product image block can help analyze the characteristics of the product itself, including appearance, shape, color, and other information. The feature vectors of the product image block can reflect characteristics such as product quality, design style, and features, helping to assess the product's appeal and market competitiveness. By extracting feature vectors of the packaging design image and product image, respectively, a more in-depth analysis and comparison of the packaging design and the product itself can be performed. Combining these feature vectors can reveal the correlation between the packaging design and the product, assessing whether the packaging design matches the product, whether it can attract the target consumer group, and whether it meets the brand image and market demand. The comprehensive use of feature vectors extracted from packaging design images and product images can provide more accurate guidance and improvement suggestions for packaging design and product optimization to meet consumer needs and improve product promotion effects.

[0029] Specifically, in one embodiment of the present application, Figure 2 FIG2 is a block diagram of a packaging design image processing module in a packaging design optimization system based on big data analysis according to an embodiment of the present application. Figure 2As shown, in the above-mentioned packaging design optimization system 100 based on big data analysis, the packaging design image processing module 120 includes: a packaging image processing unit 121, which is used to perform image block processing on the packaging design image and then extract features to obtain the global-local feature vector of the packaging design image block; a product image processing unit 122, which is used to perform convolution encoding on the product image to obtain the feature vector of the product image block.

[0030] Accordingly, in a specific example of the present application, the packaging image processing unit 121 is used to perform image segmentation processing on the packaging design image and then extract features to obtain the global-local feature vector of the packaging design image block. It should be understood that segmenting the packaging design image can decompose the entire image into multiple small blocks, each of which can provide more specific local information. By extracting the features of each image block, the details and structural features of the image can be captured more meticulously. Packaging design usually contains many details and local features, such as patterns, text, etc. By extracting the local feature vectors of the image blocks, these local information can be better captured, which helps to analyze the details and design features of the image. In addition to local features, the overall layout and color matching of the packaging design are also very important. By extracting the global feature vectors of the image blocks, the overall structure and layout features can be captured, which helps to analyze the overall style and visual effect of the packaging design. Combining global and local feature vectors can provide a more comprehensive perspective, allowing the deep learning algorithm to comprehensively consider the overall and local features of the packaging design, thereby more accurately evaluating the effects and characteristics of the packaging design. Therefore, by segmenting the packaging design image and extracting the global-local feature vector, richer information can be provided for the deep learning algorithm, helping to better understand and analyze the packaging design image, and providing strong support for improving and optimizing the packaging design.

[0031] further, Figure 3 FIG2 is a block diagram of a packaging image processing unit in a packaging design optimization system based on big data analysis according to an embodiment of the present application. Figure 3As shown, in the packaging design image processing module 120 of the above-mentioned packaging design optimization system 100 based on big data analysis, the packaging image processing unit 121 includes: an image blocking subunit 1211, used to perform image blocking processing on the packaging design image to obtain a packaging design image block sequence; a first convolutional coding subunit 1212, used to pass the packaging design image block sequence through a first convolutional neural network model as a filter to obtain a plurality of packaging design image block feature vectors; a packaging image feature extraction subunit 1213, used to extract the packaging design image block global semantic feature vector and the packaging design image block local association feature vector from the plurality of packaging design image block feature vectors; a global-local fusion subunit 1214, used to fuse the packaging design image block global semantic feature vector and the packaging design image block local association feature vector to obtain a packaging design image block global-local feature vector.

[0032] Specifically, the image segmentation subunit 1211 is configured to perform image segmentation processing on the packaging design image to obtain a sequence of packaging design image segments. It should be understood that by decomposing the packaging design image into small segments, local features within the image can be better captured. Each image segment represents a local region of the image, which facilitates the extraction of local features such as texture, pattern, and color. Dividing the image segments into sequences helps preserve spatial structural information within the image. This serialized representation better reflects the relative positions and relationships between different regions within the image, facilitating more detailed analysis of image content. Image segmentation can also be used as a data augmentation technique. By generating different image segmentation sequences, the diversity of training data can be increased, improving the generalization ability of the model. Image segmentation allows each image segment to be processed individually, which may be necessary for certain image processing tasks. For example, in object detection or image segmentation, segmenting the image into segments can better locate and identify objects. Image segmentation can reduce the size of each segment, thereby reducing computational complexity. This is particularly important when processing large images, effectively improving processing efficiency. Therefore, by performing image segmentation processing on the packaging design image to obtain a packaging design image block sequence, local features can be better extracted, spatial structure information can be preserved, data diversity can be increased, adaptation to different task requirements can be achieved, and computational complexity can be reduced, thereby providing more advantages and possibilities for the training and application of deep learning models. Accordingly, the image segmentation subunit is configured to uniformly segment the packaging design image to obtain the packaging design image block sequence.

[0033] Specifically, the first convolutional encoding subunit 1212 is configured to pass the package design image block sequence through a first convolutional neural network model acting as a filter to obtain multiple package design image block feature vectors. It should be understood that a convolutional neural network is a deep learning model that can effectively extract image features. By inputting the image block sequence into a convolutional neural network, the convolutional neural network's powerful feature extraction capabilities can be leveraged to extract rich feature information from each image block. The convolutional neural network learns local image features in the convolutional layers, which are crucial for each image block in the package design image block sequence. By processing the image block sequence in the first convolutional layer, the local features of each image block can be better captured. While learning features, the convolutional neural network also preserves the spatial structure of the input image. This is crucial for package design image block sequences, as the spatial relationships between image blocks help the model better understand the overall image content. Features learned at different levels by the convolutional neural network have different scales and levels of abstraction. By using multiple convolutional layers in the convolutional neural network, feature representations at different levels can be obtained, thereby more comprehensively describing the features of the package design image block sequence. The parameter sharing mechanism in convolutional neural networks can effectively reduce the number of model parameters, improving training efficiency and generalization capabilities. This is particularly important for processing large sequences of image patches. Therefore, by passing the sequence of packaging design image patches through the first convolutional neural network model as a filter, the advantages of convolutional neural networks in feature extraction, local feature learning, spatial information preservation, multi-scale feature learning, and parameter sharing can be fully utilized. This allows for better acquisition of feature vectors for multiple packaging design image patches, providing richer information and better performance for subsequent image processing and analysis tasks.

[0034] Correspondingly, the first convolution encoding subunit includes: a packaging design image block convolution subunit, which is used to perform convolution processing on the input data to obtain a convolution feature map; a packaging design image block pooling subunit, which is used to pool the various feature matrices of the convolution feature map along the channel dimension to obtain a pooled feature map; and a packaging design image block activation subunit, which is used to perform nonlinear activation on the pooled feature map to obtain an activation feature map; wherein, the output of the last layer of the first convolution neural network model is the multiple packaging design image block feature vectors, and the input of the first layer of the first convolution neural network model is each packaging design image block in the packaging design image block sequence.

[0035] Specifically, the packaging image feature extraction subunit 1213 is configured to extract global semantic feature vectors and local correlation feature vectors for the packaging design image blocks from the multiple packaging design image block feature vectors. It should be understood that the global semantic features of a packaging design image generally represent the overall image content and semantic information. By integrating the feature vectors of multiple image blocks, a global semantic feature vector representing the entire packaging design image can be extracted, thereby better capturing the overall image information. Different image blocks in a packaging design image may have certain correlations and interactions. By extracting local correlation feature vectors for packaging design image blocks, the correlation information between image blocks can be better captured, helping to understand the connections and importance between different regions in the image. The global semantic feature vector and the local correlation feature vector can provide image information at different levels and angles. Comprehensively considering global and local information enables the model to more comprehensively understand and analyze packaging design images, thereby improving its ability to grasp and comprehend image content. Combining the global semantic feature vector and the local correlation feature vector enables the fusion and learning of features at different levels. This helps improve the representational capabilities of features, enabling the model to better distinguish between different packaging design images and enhance the performance of subsequent tasks. In many image processing and analysis tasks, it is crucial to consider both global and local information. For example, in tasks such as object detection, image classification, and image generation, both global semantic information and local association information can have a significant impact on the accuracy and effectiveness of the task. Therefore, extracting the global semantic feature vector and local association feature vector of the packaging design image block from multiple packaging design image block feature vectors can better capture the overall semantic information and local association of the image, improve the model's ability to understand and represent image content, and provide stronger support for subsequent image processing and analysis tasks.

[0036] Correspondingly, the packaging image feature extraction subunit includes: a packaging design feature cascade secondary subunit, which is used to cascade the multiple packaging design image block feature vectors to obtain a global semantic feature vector of the packaging design image block; and a packaging design feature two-dimensional arrangement secondary subunit, which is used to two-dimensionally arrange the multiple packaging design image block feature vectors into a packaging design image block feature matrix and then pass it through an image feature extractor based on a second convolutional neural network model to obtain a local correlation feature vector of the packaging design image block.

[0037] Furthermore, the packaging design feature cascade secondary subunit is configured to concatenate the feature vectors of the multiple packaging design image blocks to obtain a global semantic feature vector for the packaging design image block. It should be understood that concatenating the feature vectors of multiple image blocks integrates their feature dimensions to form a more comprehensive and complete feature vector. This helps maintain the consistency and integrity of the overall features and improves the representational capabilities of the features. Concatenating the feature vectors of multiple image blocks simultaneously considers the information of all image blocks, thereby better capturing the global semantic information of the entire packaging design image. This helps comprehensively consider the features of different regions, thereby better understanding the overall image content. Concatenated feature vectors help the model capture the semantic relevance between different image blocks. By integrating the features of different image blocks, the relationships and importance between them can be better expressed, helping to improve the model's understanding of the overall semantics of the image. Concatenating feature vectors achieves feature fusion, combining feature information from different image blocks. This helps increase the diversity and richness of features, enabling the model to better distinguish between different packaging design images and improving the performance of subsequent tasks. By concatenating feature vectors, subsequent processing steps can be simplified. Concatenating multiple feature vectors into a single global feature vector reduces the complexity of subsequent processing, improving computational efficiency and model interpretability. Therefore, concatenating feature vectors from multiple packaging design image patches to obtain a global semantic feature vector for the patch helps comprehensively consider information from the entire image, capturing global semantics and relevance. This improves feature representation and model performance, providing stronger support for subsequent image processing and analysis tasks.

[0038] Furthermore, the second-level subunit for two-dimensional arrangement of packaging design features is configured to two-dimensionally arrange the multiple packaging design image block feature vectors into a packaging design image block feature matrix, which is then passed through an image feature extractor based on a second convolutional neural network model to obtain local correlation feature vectors for the packaging design image blocks. It should be understood that arranging the image block feature vectors into a feature matrix preserves the relative spatial relationships of the image blocks. This helps capture the local correlation and spatial structure between image blocks, improving the understanding and representation of local information. The image feature extractor based on the second convolutional neural network has excellent feature extraction capabilities when processing image data. By inputting the image block feature matrix into such a feature extractor, local correlation features between image blocks can be effectively learned, improving feature representation capabilities. The feature extractor based on a convolutional neural network can learn the correlation and importance between different regions in the image. This helps capture local correlation features between image blocks, helping the model better understand the connections and interactions between image blocks, and improving its ability to grasp local information. The convolutional neural network uses parameter sharing, which effectively reduces the number of model parameters and improves the model's generalization and efficiency. This is very important for processing large-scale image data and can better adapt to the features of image blocks of different sizes and complexities. Through the feature extractor based on the second convolutional neural network, the feature information between different image blocks can be effectively fused together. This helps to improve the diversity and richness of features, allowing the model to better capture the local correlation features between image blocks. Therefore, after arranging the feature vectors of multiple packaging design image blocks in two dimensions into a packaging design image block feature matrix, the image feature extractor based on the second convolutional neural network model can effectively extract the local correlation feature vectors of the packaging design image blocks, helping the model to better understand and analyze the correlation between image blocks, and improve the performance of image processing and analysis tasks.

[0039] Specifically, the global-local fusion subunit 1214 is configured to fuse the global semantic feature vector of the packaging design image block with the local correlation feature vector of the packaging design image block to obtain a global-local feature vector for the packaging design image block. It should be understood that the global semantic feature vector captures the overall semantic information of the entire packaging design image, while the local correlation feature vector captures the local correlations between image blocks. Fusion of these two features can better comprehensively consider the global semantics and local correlations of the image, resulting in a more comprehensive and accurate feature representation. Global semantic features and local correlation features often contain information at different levels and types. Fusion of these two features allows the model to simultaneously consider both global and local information, resulting in richer and more diverse features and improving the model's ability to understand the image. Fusion of global semantic features and local correlation features can enhance feature representation capabilities. Such a global-local feature vector not only better represents the overall semantic information of the image but also captures the correlations between different regions within the image, helping to improve the performance of subsequent tasks. By fusing global and local features, information at different levels and scales within the image can be comprehensively considered. This helps the model gain a more comprehensive understanding of image content, thereby improving its ability to analyze, recognize, and reason about images. Fusion of global and local features can improve feature robustness, making the model more adaptable to changes in scale, posture, and environment, thereby improving the model's generalization and robustness. Therefore, fusing the global semantic feature vector of a packaging design image patch with the local association feature vector of the packaging design image patch to obtain a global-local feature vector of the packaging design image patch can comprehensively consider the global and local information of the image, improve the feature representation capability and model performance, and provide a more effective feature representation for subsequent image processing, analysis, and recognition tasks.

[0040] Accordingly, the fusion global-local subunit is used to fuse the global semantic feature vector of the packaging design image block and the local association feature vector of the packaging design image block using the following fusion formula to obtain the global-local feature vector of the packaging design image block, wherein the fusion formula is:

[0041] V1=αV s +βV d

[0042] Among them, V1 is the global-local feature vector of the packaging design image block, V s Design the global semantic feature vector of the image block for the packaging, V dis the local association feature vector of the packaging design image block, "+" represents the addition of the elements at corresponding positions of the global semantic feature vector of the packaging design image block and the local association feature vector of the packaging design image block, and α and β are weighting parameters for controlling the balance between the global semantic feature vector of the packaging design image block and the local association feature vector of the packaging design image block in the global-local feature vector of the packaging design image block.

[0043] Accordingly, in a specific example of the present application, the product image processing unit 122 is configured to perform convolutional encoding on the product image to obtain the product image block feature vector. It should be understood that convolutional neural networks (CNNs) have demonstrated outstanding performance in image processing and are capable of effectively extracting features from images. Through convolutional encoding, the CNN architecture can be utilized to extract various features from product images, including texture, shape, and color, thereby converting image information into more representative feature vectors. CNNs preserve the spatial structure of images when processing them, which is particularly important for product images. Through convolution operations, the network effectively captures the correlations between different regions in the image, thereby generating image block feature vectors with more regional and local characteristics. CNNs utilize parameter sharing, which reduces the number of network parameters and improves the model's generalization and efficiency. This is crucial for processing large-scale product image data and is more adaptable to product images of varying sizes and complexities. CNNs have a multi-layered structure that can learn feature representations at different levels of image abstraction. Through convolutional encoding, high-level semantic features of the image can be gradually extracted, resulting in more abstract and meaningful product image block feature vectors. Convolutional neural networks can automatically learn the feature representations that best distinguish different categories, eliminating the need for manually designed feature extractors. This allows for better adaptation to diverse product image data types and styles, improving the discriminative and generalizable capabilities of features. Therefore, converting product images into product image block feature vectors through convolutional coding fully leverages the advantages of convolutional neural networks in image processing, extracting rich feature information while preserving spatial structure. This allows for hierarchical feature representation, improving the discriminative and generalizable capabilities of features, and providing more effective feature representations for subsequent product image analysis, recognition, and processing tasks.

[0044] further, Figure 4 FIG2 is a block diagram of a product image processing unit in a packaging design optimization system based on big data analysis according to an embodiment of the present application. Figure 4As shown, in the packaging design image processing module 120 of the above-mentioned packaging design optimization system 100 based on big data analysis, the product image processing unit 122 includes: a product image multi-feature extraction subunit 1221, which is used to pass the product image through a convolutional neural network model with multi-feature extraction capability to obtain multiple product image block feature maps; a product image feature fusion subunit 1222, which is used to fuse the multiple product image block feature maps to obtain a product image block feature map; and a product image feature dimensionality reduction subunit 1223, which is used to perform global mean pooling on each feature matrix of the product image block feature map along the channel dimension to obtain the product image block feature vector.

[0045] Specifically, the product image multi-feature extraction subunit 1221 is configured to pass the product image through a convolutional neural network model with multi-feature extraction capabilities to obtain multiple product image patch feature maps. It should be understood that convolutional neural network models have diverse feature extraction capabilities at different levels and depths. By extracting features at different levels, feature representations at multiple levels of abstraction can be obtained, ranging from low-level edges and textures to high-level semantic information, helping to more comprehensively describe the content of the product image. Convolutional neural network models can effectively capture local features in images. By applying convolution operations at different locations and scales, multiple product image patch feature maps can be obtained. Each feature map corresponds to a local feature at a different location and scale in the image, helping to better understand the structure and content of the image. Convolutional neural networks can preserve the spatial structure of images when processing them. By extracting multiple product image patch feature maps, the spatial relationships between different regions in the image can be preserved, helping to better capture local patterns and global structure in the image. By obtaining multiple product image patch feature maps, a rich and diverse feature representation can be obtained. These feature maps can be used for feature fusion, combining feature information at different scales and levels of abstraction to obtain a more comprehensive and rich feature representation of the product image. The fusion of feature maps from multiple product image patches can enhance feature representation capabilities. Combining different feature maps can provide more diverse and rich feature representations, helping to improve the model's understanding and representation of product images. Therefore, extracting multiple product image patch feature maps using a convolutional neural network model with multi-feature extraction capabilities can yield rich and diverse feature representations, including feature information at different levels of abstraction and scales. This facilitates a more comprehensive understanding of the content of product images, improves feature representation capabilities and model performance, and provides more effective feature representations for subsequent product image analysis, recognition, and processing tasks.

[0046] Specifically, the product image feature fusion subunit 1222 is configured to fuse the multiple product image block feature maps to obtain a product image block feature map. It should be understood that different product image block feature maps can capture different local information and features of an image. Fusion of these feature maps can provide a richer and more comprehensive feature representation. By fusing different feature maps, the advantages of each feature map can be integrated to better describe the content of the product image. Fusion of multiple product image block feature maps can enhance the expressive power of features. The combination of different feature maps can provide more diverse and comprehensive feature information, helping to improve the distinguishability and representation of features, making the final product image block feature map more representative. The fusion process can avoid information loss. Different feature maps may contain complementary information. By fusing this information, important features in the image can be better preserved, avoiding information loss caused by a single feature map. Fusion of multiple product image block feature maps can improve the robustness and stability of features. By fusing feature information from multiple sources, the impact of noise can be reduced, making the final feature representation more stable and reliable. The fusion process can also reduce dimensionality and feature complexity. By fusing multiple feature maps into a single product image patch feature map, we can simplify feature representation, reduce computational complexity, and improve model efficiency and performance. Therefore, fusing multiple product image patch feature maps to obtain a product image patch feature map can provide a richer and more comprehensive feature representation, enhance the expressiveness and discrimination of features, reduce information loss, and improve the robustness and stability of features. It can also reduce feature dimensionality and simplify feature representation, providing a more effective and efficient feature representation for subsequent product image analysis, recognition, and processing tasks.

[0047] Specifically, the product image feature dimensionality reduction subunit 1223 is configured to perform global mean pooling on each feature matrix along the channel dimension of the product image patch feature map to obtain the product image patch feature vector. It should be understood that global mean pooling can combine the spatial information in each feature matrix into a single value, thereby reducing the feature dimensionality. This helps reduce the length of the feature vector, simplify the feature representation, reduce computational cost, and mitigate the possibility of overfitting. Mean pooling can help retain the most important feature information in the image patch feature map. By performing mean pooling on each feature matrix, the average feature of each channel can be extracted, retaining information that contributes to the overall feature representation. Global mean pooling can reduce the number of model parameters. Compared to fully connected layers or other operations, global mean pooling does not require additional parameters and simply performs an averaging operation on the feature matrices. Therefore, it can reduce the number of model parameters and improve model efficiency. Mean pooling improves the model's invariance to translation to a certain extent. By performing mean pooling on the entire feature map, the features can be made insensitive to the specific location of objects in the image, thereby improving the model's generalization ability. Global mean pooling can accelerate model training and inference. Because mean pooling is simple and efficient, it can reduce computational effort, speed up model training, and even improve model speed during inference. Therefore, by performing global mean pooling on each feature matrix of a product image patch feature map, we can reduce feature dimensionality, retain key features, reduce the number of parameters, improve translation invariance, and accelerate training and inference, resulting in a more simplified, efficient, and effective feature vector representation of the product image patch.

[0048] In this embodiment of the present application, the packaging design image fusion module 130 is configured to fuse and modulate the global-local feature vector of the packaging design image block and the feature vector of the product image block to obtain an optimized feature vector for product packaging design analysis. It should be understood that the global-local feature vector of the packaging design image block and the feature vector of the product image block each capture feature information at different levels. Fusion of these two feature vectors can integrate global and local information, providing a more comprehensive and rich feature representation, helping to better describe the characteristics of the product packaging design. There is a certain correlation between the packaging design image and the product image, and fusion of their feature vectors can better capture this correlation. By integrating these two types of feature information, the connection between the packaging design and the product can be more accurately analyzed, providing stronger support for subsequent analysis and decision-making. Combining global-local features with product features can improve feature accuracy and representational capabilities. This fusion can compensate for the limitations of each feature vector, thereby enhancing the feature's discriminability and expressiveness, making the final feature vector more representative. By fusing feature vectors from different levels and sources, the risk of model overfitting can be reduced. Fusion can help the model better generalize to new data and improve its robustness. Fusion of different feature vectors can enhance model performance. By combining global-local features and product features, the feature vector can be made more discriminative, thereby improving the performance of the model in packaging design analysis tasks.

[0049] Specifically, the packaging design image fusion module includes: a packaging design image fusion unit, which is used to perform weighted fusion of the global-local feature vector of the packaging design image block and the feature vector of the product image block to obtain a product packaging design analysis feature vector; and a packaging design image optimization unit, which performs eigenregression-guided feature kernel space mapping modulation on the product packaging design analysis feature vector to obtain an optimized feature vector.

[0050] In particular, when segmenting packaging design images and extracting features through CNN models, while CNNs excel at identifying local patterns and basic structures, these fundamental features may not directly represent complex higher-order relationships. For example, a product's packaging design may consist of multiple elements (such as color combinations, pattern layouts, and text typesetting), and the interactions between these elements are often not simply linear superpositions, but rather influence each other through complex nonlinear mechanisms. Traditional CNN-based methods struggle to fully capture these deep relationships because they primarily rely on convolutional and pooling layers to extract spatial information, mechanisms that are limited in their ability to capture subtle and complex interactions across regions or elements. Furthermore, when fusing feature vectors at different levels (such as global semantic feature vectors with local correlation feature vectors, or packaging design features with product image features), simply concatenating or adding these vectors may not be sufficient to represent the complex interactions between them. This approach assumes that the feature vectors are independent or linearly correlated. However, in reality, different visual and design elements may have complex dependencies and interactions that cannot be fully captured by simple linear operations. Therefore, further, the product packaging design analysis feature vector is subjected to eigenregression-guided feature kernel space mapping modulation to obtain an optimized feature vector.

[0051] More specifically, the packaging design image optimization unit is used to: first, construct a pixel-level information association matrix of the product packaging design analysis feature vector, which is expressed as follows:

[0052]

[0053] Where V represents the product packaging design analysis feature vector, v i and v j denote the eigenvalues ​​of the i-th and j-th positions of the product packaging design analysis feature vector, d(v i ,v j ) represents the calculation of Euclidean distance, D i,j Represents the eigenvalue of the (i, j) position of the pixel-by-pixel granularity information correlation matrix.

[0054] That is, by analyzing the correlation strength between each unit within the product packaging design analysis feature vector pixel by pixel, the implicit visual logic of the packaging design, such as the global style consistency and local element coordination, is transformed into a computable topological relationship network, thereby revealing the distribution pattern of redundant information in the feature space and the synergistic mechanism of key visual elements. Specifically, by quantifying the similarity of the response patterns of each pixel-level unit within the product packaging design analysis feature vector, it is possible to accurately locate visual interference areas such as improper color matching of graphics and chaotic information hierarchy in the packaging design. At the same time, it captures redundant patterns or invalid decorations that appear repeatedly across regions. The generated pixel-level information correlation matrix provides structured relationship constraints for subsequent feature fusion.

[0055] Secondly, the kernel domain feature extraction is performed on the pixel-level information association matrix based on the convolution layer to obtain the kernel domain nonlinear excitation matrix for product packaging design analysis, which is expressed as:

[0056] M=Conv(D)

[0057] Wherein, D represents the pixel-by-pixel granularity information association matrix, Conv represents the convolutional layer, and M represents the product packaging design analysis kernel domain nonlinear excitation matrix.

[0058] That is, by introducing the convolution layer, the discrete pixel-level information association matrix is ​​upgraded to a higher-order association topological space, thereby capturing the nonlinear visual semantic structures such as chain association transmission (such as multi-level color block linkage) and local association clusters (such as the symbiotic relationship between text and background) implicit in the packaging design features. Specifically, the convolution kernel reconstructs the deep contextual relationship in the pixel-level information association matrix in a data-driven manner, and maps the pixel-level association intensity into semantically interpretable association topological features (for example, identifying the nonlinear collaborative constraint relationship between the "primary color block-auxiliary pattern-text information" in the packaging design), thereby generating a nonlinear excitation matrix in the product packaging design analysis kernel domain, so that the subsequent modulation stage can accurately adjust the layout logic and visual weight distribution of packaging elements based on these learned association topological rules.

[0059] Then, the pixel-level information association matrix is ​​subjected to spectral feature analysis to obtain a sequence of characteristic eigencomponent coding vectors of product packaging design analysis features, which can be expressed as follows:

[0060]

[0061] Where T represents the transpose of the vector, Λ represents the diagonal matrix, λ1 and λ m They represent the first and mth eigenvalues ​​of the diagonal matrix respectively, U represents the sequence of the encoding vector of the characteristic eigencomponent of product packaging design analysis, x1, x2, x mRepresent the first, i-th and m-th product packaging design analysis feature eigencomponent encoding vectors respectively.

[0062] That is, the complex pixel-level information association matrix is ​​orthogonally decomposed into a sequence of product packaging design analysis feature intrinsic component encoding vectors that characterize the visual laws of packaging design. For example, the low-frequency correlation basis vectors that dominate the overall coordination of the packaging (such as the uniformity pattern of large-area color blocks) and the high-frequency correlation basis vectors that depict the contrast of details (such as the sharpness pattern of text edges) are extracted, thereby establishing a decoupled feature expression from macro-style to micro-elements. Specifically, the sequence of product packaging design analysis feature intrinsic component encoding vectors generated by spectral decomposition can explicitly separate the function-oriented core patterns in packaging design (such as the contrast correlation basis related to information transmission efficiency) and redundant interference patterns (such as the distracting random texture correlation basis), providing an interpretable weight adjustment basis for subsequent feature modulation, and ultimately improving the interpretability and robustness of the classifier in style consistency assessment and information hierarchy discrimination.

[0063] Next, each product packaging design analysis feature intrinsic component encoding vector in the sequence of the product packaging design analysis feature intrinsic component encoding vector is input into the self-attention driven feature importance adjustment module to obtain a sequence of enhanced product packaging design analysis feature intrinsic component encoding vectors, which can be expressed as:

[0064] Y=Transformer{[x1,x2,…,x m ]}=[y1,y2,…,y m ]

[0065] Among them, Transformer represents the self-attention model, Y represents the sequence of the encoding vector of the enhanced product packaging design analysis feature intrinsic component, y1, y2, y m They represent the first, i-th and m-th enhanced product packaging design analysis feature eigencomponent encoding vectors respectively.

[0066] That is, a cross-modal dependency graph is established between the eigencomponent encoding vectors of the product packaging design analysis features, and the decision logic is simulated through the self-attention mechanism. Thus, under the premise of preserving the orthogonal structure of the spectral decomposition, the semantic weight distribution of each basis vector is dynamically reconstructed according to the overall feature context of the current packaging design. Specifically, by calculating the mutual correlation between basis vectors (such as the need for synergistic reinforcement of the color block uniformity pattern and the text contrast pattern in a specific design), the self-attention mechanism can adaptively enhance the high-contribution patterns that meet the "clean color block" standard (such as suppressing distracting high-frequency texture association basis vectors), while identifying and amplifying the visual redundancy elimination rules implicit in the cross-basis vector combination (such as automatically reducing the weight when detecting conflicts between color block-pattern association basis vectors). This allows the enhanced sequence of enhanced product packaging design analysis feature eigencomponent encoding vectors to retain the physical interpretability of the original spectral decomposition and embed a dynamic feature selection mechanism for commercial packaging optimization goals.

[0067] Then, each enhanced product packaging design analysis feature intrinsic component coding vector in the sequence of the enhanced product packaging design analysis feature intrinsic component coding vector is mapped to the product packaging design analysis kernel domain nonlinear excitation matrix to obtain a sequence of product packaging design analysis feature intrinsic component kernel mask coding vectors, which is expressed as follows:

[0068]

[0069] in, represents matrix multiplication, S represents the characteristic scale of the nonlinear excitation matrix in the kernel domain of product packaging design analysis, and y i represents the i-th enhanced product packaging design analysis feature intrinsic component encoding vector, L represents the length of the enhanced product packaging design analysis feature intrinsic component encoding vector, z i Represents the kernel mask encoding vector of the eigencomponent of the i-th product packaging design analysis feature.

[0070] Specifically, the saliency-weighted enhanced product packaging design analysis feature intrinsic component encoding vectors are combined with the product packaging design analysis kernel domain nonlinear excitation matrix learned by the convolution kernel for cross-domain feature interaction, thereby injecting data-driven local complex association rules into the global structural features while retaining the interpretability of the orthogonal basis vectors. This generates a sequence of product packaging design analysis feature intrinsic component kernel mask encoding vectors that incorporates both the physical meaning of the orthogonal basis patterns (such as global color block coordination) and the integration of local association semantics (such as the symbiotic strength of color blocks and patterns), providing the subsequent fusion module with a dual-domain feature representation that is both structurally robust and semantically sensitive.

[0071] Finally, the sequence of the product packaging design analysis feature intrinsic component kernel mask encoding vectors is fused to obtain the optimized product packaging design analysis feature vector, which is expressed as follows:

[0072] V=Concat{z1,z2,…,z m}

[0073] Among them, Concat represents the cascade function, z1, z2, z m They represent the first, i-th and m-th product packaging design analysis feature eigencomponent kernel mask encoding vectors respectively, and V' represents the optimized product packaging design analysis feature vector.

[0074] That is, through the cascade fusion method, the feature dimensions that are strongly correlated with the commercial packaging optimization objectives in the kernel mask encoding vector of the intrinsic components of each product packaging design analysis feature are selectively enhanced. The generated optimized product packaging design analysis feature vector simultaneously embeds the interpretable structural laws of the packaging design and the complex association constraints driven by data, so that the classifier can synchronously evaluate the global style unity, local element coordination and information communication efficiency of the design scheme based on the optimized feature vector.

[0075] In this embodiment of the present application, the packaging design image analysis module 140 is configured to pass the optimized product packaging design analysis feature vector through a classifier to obtain a classification result. The classification result indicates whether the product packaging design complies with standards. It should be understood that, using image recognition technology, artificial intelligence can help identify whether a product packaging design complies with relevant laws, regulations, and standards. Whether a product packaging design complies with relevant laws, regulations, and standards is extremely important, as non-compliant packaging designs may lead to legal issues or mislead consumers. By inputting the optimized product packaging design feature vector into a classifier for classification, automated evaluation of the product packaging design can be achieved. This automated evaluation process improves efficiency and saves time and labor costs. The classifier can determine whether a product packaging design complies with standards based on pre-set standards or rules. By comparing the design feature vector with these standards, the quality and degree of compliance of the design can be quantitatively assessed. The classification result can quickly determine whether the product packaging design complies with standards, providing timely feedback to designers and relevant personnel. This helps to identify design issues early and make adjustments, thereby improving design quality. The classification result can provide support and reference for decision-making. Based on the classification result, relevant parties can decide whether design adjustments are necessary to ensure that the product packaging design complies with standards and meets requirements. By continuously evaluating product packaging designs in categories, a monitoring mechanism can be established to promptly identify design deviations or problems and take corrective measures. This helps to continuously improve the quality and compliance of product packaging designs.

[0076] Accordingly, in one embodiment of the present application, the packaging design image analysis module is configured to: use the classifier to process the optimized product packaging design analysis feature vector using the following formula to obtain the classification result;

[0077] Wherein, the formula is: softmax{(W n ,B n ):…:(W1,B1)|X}, where W1 to W n is the weight matrix, B1 to B n is the bias vector, X is the feature vector for optimizing product packaging design analysis, softmax represents the softmax function, and O represents the classification result.

[0078] In summary, the packaging design optimization system and method based on big data analysis in the embodiments of the present application utilize product images and packaging design images to analyze and optimize packaging design, and enhance the attractiveness and communication effect of product packaging by adjusting colors, optimizing patterns, improving information communication, etc., so as to meet consumers' high requirements for visual experience and provide more attractive and information-communicating packaging design solutions for beauty products and other fields.

[0079] As described above, the packaging design optimization system 100 based on big data analysis according to the embodiment of the present application can be implemented in various terminal devices, such as a server of the packaging design optimization system based on big data analysis. In one example, the packaging design optimization system 100 based on big data analysis can be integrated into the terminal device as a software module and / or hardware module. For example, the packaging design optimization system 100 based on big data analysis can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the packaging design optimization system 100 based on big data analysis can also be one of the many hardware modules of the terminal device.

[0080] Alternatively, in another example, the packaging design optimization system 100 based on big data analysis and the terminal device may also be separate devices, and the packaging design optimization system 100 based on big data analysis may be connected to the terminal device via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0081] Figure 5 Flowchart of the packaging design optimization method based on big data analysis according to an embodiment of the present application. Figure 5As shown, the packaging design optimization method based on big data analysis according to the embodiment of the present application includes the steps of: S110, acquiring a product image and a packaging design image; S120, extracting a packaging design image block global-local feature vector and a product image block feature vector from the packaging design image and the packaging design image, respectively; S130, fusing and modulating the packaging design image block global-local feature vector and the product image block feature vector to obtain an optimized product packaging design analysis feature vector; S140, passing the optimized product packaging design analysis feature vector through a classifier to obtain a classification result, and the classification result is used to indicate whether the product packaging design meets the standard.

[0082] Here, those skilled in the art will understand that the specific operations of each step in the above packaging design optimization method based on big data analysis have been referred to above. Figures 1 to 4 The description of the packaging design optimization system based on big data analysis has been introduced in detail, and therefore, its repeated description will be omitted.

[0083] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.

[0084] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0085] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0086] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0087] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0088] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0089] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A packaging design optimization system based on big data analysis, characterized in that: include: Packaging design image acquisition module, used to acquire product images and packaging design images; a packaging design image processing module for extracting a packaging design image block global-local feature vector and a product image block feature vector from the packaging design image and the packaging design image, respectively; A packaging design image fusion module is used to fuse and modulate the global-local feature vector of the packaging design image block and the feature vector of the product image block to obtain an optimized product packaging design analysis feature vector; The packaging design image analysis module is used to pass the optimized product packaging design analysis feature vector through a classifier to obtain a classification result, and the classification result is used to indicate whether the product packaging design meets the standards.

2. The packaging design optimization system based on big data analysis according to claim 1 is characterized in that: The packaging design image processing module includes: a packaging image processing unit, configured to perform image block processing on the packaging design image and then perform feature extraction to obtain a global-local feature vector of the packaging design image block; The product image processing unit is configured to perform convolution coding on the product image to obtain the product image block feature vector.

3. The packaging design optimization system based on big data analysis according to claim 2 is characterized in that: The packaging image processing unit includes: An image blocking subunit, configured to perform image blocking processing on the packaging design image to obtain a packaging design image block sequence; a first convolutional encoding subunit, configured to pass the package design image block sequence through a first convolutional neural network model as a filter to obtain a plurality of package design image block feature vectors; a packaging image feature extraction subunit, configured to extract a global semantic feature vector of the packaging design image block and a local correlation feature vector of the packaging design image block from the plurality of packaging design image block feature vectors; The global-local fusion subunit is used to fuse the global semantic feature vector of the packaging design image block and the local association feature vector of the packaging design image block to obtain a global-local feature vector of the packaging design image block.

4. The packaging design optimization system based on big data analysis according to claim 3 is characterized in that: The image blocking subunit is configured to perform uniform image blocking processing on the packaging design image to obtain the packaging design image block sequence.

5. The packaging design optimization system based on big data analysis according to claim 4 is characterized in that: The first convolutional encoding subunit includes: The package design image block convolution subunit is used to perform convolution processing on the input data to obtain a convolution feature map; The packaging design image block pooling subunit is used to pool each feature matrix of the convolution feature map along the channel dimension to obtain a pooled feature map; a packaging design image block activation subunit, configured to perform nonlinear activation on the pooled feature map to obtain an activation feature map; Among them, the output of the last layer of the first convolutional neural network model is the feature vectors of the multiple packaging design image blocks, and the input of the first layer of the first convolutional neural network model is each packaging design image block in the packaging design image block sequence.

6. The packaging design optimization system based on big data analysis according to claim 5 is characterized in that: The packaging image feature extraction subunit includes: a packaging design feature cascade secondary subunit, configured to cascade the plurality of packaging design image block feature vectors to obtain a global semantic feature vector of the packaging design image block; The second-level sub-unit of the two-dimensional arrangement of packaging design features is used to arrange the multiple packaging design image block feature vectors in two dimensions into a packaging design image block feature matrix and then pass it through an image feature extractor based on a second convolutional neural network model to obtain a local correlation feature vector of the packaging design image block.

7. The packaging design optimization system based on big data analysis according to claim 6 is characterized in that: The product image processing unit includes: A product image multi-feature extraction subunit, configured to pass the product image through a convolutional neural network model with multi-feature extraction capabilities to obtain a plurality of product image block feature maps; A product image feature fusion subunit, configured to fuse the multiple product image block feature maps to obtain a product image block feature map; The product image feature dimensionality reduction subunit is used to perform global mean pooling on each feature matrix along the channel dimension of the product image block feature map to obtain the product image block feature vector.

8. The packaging design optimization system based on big data analysis according to claim 7 is characterized in that: The packaging design image fusion module includes: a packaging design image fusion unit, configured to perform weighted fusion on the global-local feature vector of the packaging design image block and the feature vector of the product image block to obtain a product packaging design analysis feature vector; The packaging design image optimization unit performs eigenregression-guided feature kernel space mapping modulation on the product packaging design analysis feature vector to obtain an optimized product packaging design analysis feature vector.

9. The packaging design optimization system based on big data analysis according to claim 8 is characterized in that: The packaging design image optimization unit is used to: Constructing a pixel-level information association matrix of the product packaging design analysis feature vector; Performing kernel domain feature extraction on the pixel-level information association matrix based on a convolutional layer to obtain a kernel domain nonlinear excitation matrix for product packaging design analysis; Performing spectral feature analysis on the pixel-level information association matrix to obtain a sequence of product packaging design analysis feature eigencomponent encoding vectors; Inputting each product packaging design analysis feature intrinsic component encoding vector in the sequence of product packaging design analysis feature intrinsic component encoding vectors into a self-attention driven feature importance adjustment module to obtain a sequence of enhanced product packaging design analysis feature intrinsic component encoding vectors; Mapping each enhanced product packaging design analysis feature intrinsic component coding vector in the sequence of enhanced product packaging design analysis feature intrinsic component coding vectors to the product packaging design analysis kernel domain nonlinear excitation matrix to obtain a sequence of product packaging design analysis feature intrinsic component kernel mask coding vectors; The sequence of the product packaging design analysis feature eigencomponent kernel mask encoding vectors is fused to obtain the optimized product packaging design analysis feature vector.

10. A packaging design optimization method based on big data analysis, characterized in that: include: Get product images and packaging design images; extracting a package design image block global-local feature vector and a product image block feature vector from the package design image and the package design image, respectively; fusing and modulating the global-local feature vector of the packaging design image block and the feature vector of the product image block to obtain an optimized product packaging design analysis feature vector; The optimized product packaging design analysis feature vector is passed through a classifier to obtain a classification result, and the classification result is used to indicate whether the product packaging design meets the standards.

Images