Intelligent product aesthetics evaluation method based on aesthetic preference

Abstract

The invention discloses an aesthetic preference-based intelligent product aesthetics evaluation method. The method comprises the steps of constructing an intelligent product aesthetics evaluation system with first-level and second-level evaluation indexes; secondly, collecting scoring data of first-level indexes and consumer preferences, and constructing intelligent product consumer aesthetic preference linear regression; collecting consumer aesthetic form scores, establishing an intelligent product aesthetic form evaluation data set, and training a consumer general aesthetic form evaluation and prediction model; training is adjusted on the basis of the model to obtain a personalized aesthetic form evaluation and prediction model for a specific consumer group; and evaluating the prediction score, the artificially input aesthetic experience and the aesthetic culture score based on the personalized aesthetic form to obtain a final consumer aesthetic preference score. According to the method, semi-automatic intelligent evaluation of the intelligent product is realized, the screening and optimization efficiency of the design scheme is improved, and meanwhile, the preference degree of the target consumer group for the intelligent product design scheme can be accurately analyzed.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent product design, and in particular relates to an intelligent product aesthetics evaluation method based on aesthetic preference. Background Art

[0002] Product aesthetic evaluation is an important issue in aesthetics. Research on product form aesthetic evaluation can help designers understand the factors that affect the formation of product aesthetic feelings and the internal connections and laws of the aesthetic evaluation system. Common methods include using the analytic hierarchy process to establish the weight relationship between aesthetic indicators. With the development of machine learning, it is possible to simulate the user's visual experience and automatically evaluate the aesthetics of product images, thereby providing designers with timely aesthetic feedback and guidance.

[0003] Automated aesthetic assessment methods can be divided into two categories: General Image Aesthetic Assessment (GIAA) and Personalized Image Aesthetic Assessment (PIAA). The purpose of General Image Aesthetic Assessment (GIAA) is to infer the aesthetic experience of most people, while PIAA is designed for a user's aesthetic rating of an image. Since people's aesthetic evaluation of products is subjective, it is crucial to simulate the aesthetic criteria of individual users to estimate the aesthetic preferences of images for personalized groups. This is very valuable in many user-centric applications such as personalized product recommendations, personalized product titles, and personalized product enhancements. Summary of the invention

[0004] In view of the above, an object of the present invention is to provide an intelligent product aesthetics evaluation method based on aesthetic preferences, which can provide targeted design guidance to achieve rapid and automated evaluation of product design solutions.

[0005] The technical solution adopted by the present invention is:

[0006] Constructing an aesthetic evaluation index for an intelligent product image, wherein the aesthetic evaluation index includes three primary and nineteen secondary intelligent product evaluation indexes;

[0007] Collect data on consumers’ preferences for three primary indicators, namely aesthetic form, aesthetic experience, and aesthetic culture, and construct a linear regression of consumers’ aesthetic preferences for product images;

[0008] Collect consumer preference information and image aesthetic form labeling scores, establish a smart product consumer aesthetic form evaluation data set, and establish a consumer general aesthetic form evaluation prediction model;

[0009] Adjustments are made on the basis of the general aesthetic form evaluation prediction model to obtain an aesthetic form scoring prediction model for personalized consumer labels; the aesthetic form scoring prediction model is trained, and after the training is completed, the personalized aesthetic form prediction scores of consumers in the personalized group for the target image are obtained based on the model.

[0010] Based on the predicted score of the personalized aesthetic form evaluation of the target image, the user manually inputs the score of aesthetic experience and aesthetic culture, and the final aesthetic preference score is calculated according to the multivariate linear regression equation of the three first-level indicators.

[0011] The present invention also provides an intelligent product aesthetic evaluation device based on aesthetic preference, comprising: a processor, and a memory communicatively connected to the processor;

[0012] The memory stores computer-executable instructions;

[0013] The processor executes the computer-executable instructions stored in the memory to implement the above method.

[0014] The present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to implement the above method when executed by a processor.

[0015] The beneficial effects of the present invention are as follows: the present invention realizes semi-automatic intelligent evaluation of intelligent products by constructing an intelligent product aesthetic evaluation method based on aesthetic preference, accelerates the screening and optimization efficiency of design schemes, and can accurately analyze the preference of target consumer groups for intelligent product design schemes, thereby designing products that are more in line with market trends and consumer preferences, thereby expanding the company's market share. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0017] Figure 1 is a flow chart of a method for aesthetic evaluation of an intelligent product based on aesthetic preference provided in an embodiment;

[0018] Figure 2 It is a technical framework diagram of an aesthetic preference-based intelligent product aesthetic evaluation method provided in an embodiment. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] The inventive concept of the present invention is:

[0021] In view of the problem that the existing technology for the aesthetic evaluation method of smart products is not deep enough and the evaluation dimension is single, the aesthetic evaluation system described in the embodiment of the present invention includes primary and secondary evaluation indicators; secondly, collect data on the three primary indicators of aesthetic form, aesthetic experience, and aesthetic culture, and the degree of consumer preference, and construct a linear regression of consumer aesthetic preferences for product images; collect consumer preference information and product aesthetic form scores, establish a consumer evaluation data set for the aesthetic form of smart products, and establish a consumer general aesthetic form score prediction model; make adjustments based on the general aesthetic form evaluation prediction model, and train a specific group of consumers to predict the personalized aesthetic form evaluation score for the target image. Based on the personalized aesthetic form evaluation prediction score of the target image, the user manually enters the score of the aesthetic experience and aesthetic culture, and the final aesthetic preference score is calculated according to the multivariate linear regression equation of the three primary indicators.

[0022] Adjustments are made based on the general aesthetic form prediction model to train and obtain personalized aesthetic form prediction scores for target images for specific groups of consumers.

[0023] To achieve the above object, the present invention comprises the following steps:

[0024] An intelligent product image aesthetic evaluation index is constructed, wherein the aesthetic evaluation index includes 3 primary and 19 secondary intelligent product evaluation indexes.

[0025] Establish smart product image data including three primary indicators of aesthetic form, aesthetic experience, and aesthetic culture and the degree of consumer aesthetic preference, and construct a linear regression of consumer aesthetic preference for smart product images.

[0026] Collect consumer preference information and product aesthetic form scores, establish a consumer evaluation dataset of intelligent product aesthetic form, and establish a consumer general aesthetic form score prediction model.

[0027] Adjustments are made based on the general aesthetic form evaluation prediction model to train specific groups of consumers to predict personalized aesthetic form evaluation scores for target images.

[0028] Based on the predicted score of the personalized aesthetic form evaluation of the target image, the user manually inputs the score of aesthetic experience and aesthetic culture, and the final aesthetic preference score is calculated according to the multivariate linear regression equation of the three first-level indicators.

[0029] In order to construct aesthetic evaluation indicators for smart product images, crawler software was used to crawl the literature on the Sciencedirector&Web of science core database, and the search keywords were papers published on the theme of smart product aesthetics and smart product attributes. The subject areas are journals in the fields of social sciences, computer science, and engineering machinery research. The data used in the study were downloaded from the WOS core database and Science director. The publication time was from 2000 to 2023, and the language was limited to English. The document types included research papers, conference papers, and review papers. A total of 102 basic documents were obtained, and 505 keywords were extracted from the 102 documents. The 505 keywords were merged and classified, and the obtained indicator keywords were compared, combined, and configured according to three attribute clusters: aesthetic form, aesthetic experience, and aesthetic culture. On this basis, a three-layer framework of smart product attributes was established. The first layer is three clusters. The second layer includes 19 attributes.

[0030] The consumer aesthetic preference linear regression model is characterized in that the consumer aesthetic preference linear regression model is constructed for the smart product image, and a linear regression model between aesthetic form, aesthetic experience, aesthetic culture and consumer aesthetic preference is established based on user scores.

[0031] The expert aesthetic form evaluation prediction model is characterized by collecting consumer preference information, establishing a smart product consumer preference dataset, and designing a general aesthetic extraction module to obtain the aesthetic score distribution and multiple objective attributes of most users on the aesthetic form of product images.

[0032] The consumer personalized aesthetic form evaluation model is characterized in that the aesthetic form score prediction model based on personalized consumer tags is constructed for the smart product image, including using a multi-attribute interactive reasoning network. By interacting multiple subjective attributes of the user with multiple objective attributes of the image, the obtained multi-attribute interactive features are integrated with the aesthetic form score distribution, thereby predicting the consumer's personalized aesthetic form score.

[0033] Figure 1 Schematic diagram of the process flow of the intelligent product aesthetic evaluation method provided in the embodiment of the present invention. Figure 1 As shown, the embodiment provides a method for aesthetic evaluation of a smart product based on aesthetic preference, comprising the following steps:

[0034] S1, building aesthetic evaluation indicators for smart products:

[0035] The literature on the Science director & Web of science core database was crawled using crawler software, and the search keywords were papers published on the theme of smart product aesthetics and smart product attributes. The subject areas are journals in the fields of social science, computer science, and engineering machinery research. The data used in the study were downloaded from the WOS core database and Sciencedirector. The publication time was from 2000 to 2023, the language was limited to English, and the document types included research papers, conference papers, and review papers. A total of 102 basic documents were obtained, and 505 keywords were extracted from the 102 documents. The 505 keywords were merged and classified, and the obtained indicator keywords were compared, combined, and configured according to three attribute clusters: aesthetic form, aesthetic experience, and aesthetic culture. On this basis, a three-layer framework of smart product attributes was established. The first layer is three clusters. The second layer includes 19 attributes. It includes three primary indicators of aesthetic form, aesthetic experience, and aesthetic culture and 19 secondary indicators of product attributes.

[0036]

[0037]

[0038] S2, constructs the regression relationship between the first-level aesthetic evaluation index and consumer aesthetic preference for smart products. According to the aesthetic preference of consumers, the first-level product aesthetic index of smart product evaluation is scored by consumers using the method of scoring.

[0039] Based on the scoring scores, the regression relationship between the three primary indicators of aesthetic form, aesthetic experience, and aesthetic culture and consumers' aesthetic preferences is established, and the multivariate linear regression equation is determined. The regression equation is as follows:

[0040] α hps =0.4897α f +0.0547α e +0.0091α c +0.9306 (1)

[0041] where α hps represents the aesthetic preference score of consumers, α f , α e , α c They represent aesthetic form, aesthetic experience and aesthetic culture respectively.

[0042] S2, in order to establish a sample database of the aesthetic form of consumer smart products, the data content includes attribute annotation information and personal social information. The format of the data is similar to the information annotated by the annotator for a picture, including the aesthetic form score given by the annotator, as well as the annotator's age, gender, education level, personality traits, self-monitoring level, marital status, and place of origin.

[0043] S3, in order to obtain the universal aesthetics of the image, the pre-trained image classification network is used to first obtain the features of the image, and multi-task learning based on convolutional neural network (CNN) is performed to extract universal aesthetic attributes and image features.

[0044] The proposed CNN inherits the typical ResNet, removes the fully connected layer, uses the pre-trained parameters as the initial weights of the CNN, and adds a global average pooling layer (GAP) and two fully connected layers (FC). Such a network can map images to multiple aesthetic attributes and score distributions, such as Figure 2 shown.

[0045] Specifically, for an input image x, the general aesthetic extraction module can be formulated as:

[0046]

[0047] Where θ represents the CNN parameters, f θ , FC θa and FC θd Representing and predicting multiple aesthetic attributes and score distribution The corresponding FC layer parameters.

[0048] S4, in order to obtain consumers' personalized aesthetic preferences and make the model more applicable to a specific user, the model is fine-tuned for the target user. The loss function for fine-tuning MLP and FC can be expressed as:

[0049]

[0050] where Nb and Ns represent the number of training users and the corresponding rated images, respectively.

[0051] The objective attributes and score distribution of the image are obtained using the general aesthetic extraction module, which can be defined as:

[0052]

[0053] in and Represents the aesthetic score of the i-th image.

[0054] Then, using the user's subjective attributes and the predicted objective attributes Interact and integrate interactive features and score distribution To obtain a personalized aesthetic score, the calculation formula is:

[0055]

[0056] in represents the predicted aesthetic score. Generally speaking, a specific user can only provide a small number of annotated samples for model fine-tuning.

[0057] S5, based on the predicted score of personalized aesthetic form evaluation of the target image, the user manually inputs the score of aesthetic experience and aesthetic culture, and the final aesthetic preference score is calculated according to the multivariate linear regression equation mentioned in S2.

[0058] The specific implementation methods described above provide a detailed description of the technical solutions and beneficial effects of the present invention. It should be understood that the above is only the most preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, supplements and equivalent substitutions made within the scope of the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for evaluating the aesthetics of intelligent products based on aesthetic preference, characterized in that: The following steps are involved: Constructing an aesthetic evaluation index for an intelligent product image, wherein the aesthetic evaluation index includes three primary and nineteen secondary intelligent product evaluation indexes; Collect data on consumers’ preferences for three primary indicators, namely aesthetic form, aesthetic experience, and aesthetic culture, and construct a linear regression of consumers’ aesthetic preferences for product images; Collect consumer preference information and image aesthetic form labeling scores, establish a smart product consumer aesthetic form evaluation data set, and establish a consumer general aesthetic form evaluation prediction model; Adjustments are made on the basis of the general aesthetic form evaluation prediction model to obtain an aesthetic form scoring prediction model for personalized consumer labels; the aesthetic form scoring prediction model is trained, and after the training is completed, personalized aesthetic form prediction scores of consumers in the personalized group for the target image are obtained based on the model; Based on the predicted score of the personalized aesthetic form evaluation of the target image, the user manually inputs the score of aesthetic experience and aesthetic culture, and the final aesthetic preference score is calculated according to the multivariate linear regression equation of the three first-level indicators.

2. The method for aesthetic evaluation of intelligent products based on aesthetic preference according to claim 1, characterized in that: The specific steps of constructing the intelligent product image aesthetic evaluation index are as follows: merging and classifying multiple keywords collected from the literature; The obtained keywords are compared, combined and configured according to three attribute clusters: aesthetic form, aesthetic experience and aesthetic culture; A three-layer framework of smart product attributes is established, with the first layer consisting of three clusters and the second layer including nineteen attributes.

3. The method for aesthetic evaluation of intelligent products based on aesthetic preference according to claim 1, characterized in that: The consumer aesthetic preference linear regression for constructing smart product images is specifically: Based on user scores, a linear regression model is established between aesthetic form, aesthetic experience, aesthetic culture and consumers' aesthetic preference.

4. The method for evaluating the aesthetics of intelligent products based on aesthetic preferences according to claim 1, characterized in that: The adjustments are made based on the model of general aesthetic form evaluation prediction, specifically: An aesthetic form evaluation prediction model based on personalized consumer labels is constructed, including the use of a multi-attribute interactive reasoning network. By interacting multiple subjective attributes of the user with multiple objective attributes of the image, the obtained multi-attribute interaction features are integrated with the aesthetic form score distribution, thereby predicting the consumer's personalized aesthetic form evaluation score.

5. An intelligent product aesthetic evaluation device based on aesthetic preference, characterized in that: include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method according to any one of claims 1 to 4.

6. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 4 when executed by a processor.

Images