In order to design a cultural and creative product that matched the target image, this paper proposed to use EEG, interactive genetic algorithm (IGA), and back propagation neural network (BPNN) to analyze the users’ image preferences. Firstly, the pictures of cultural elements were grouped according to the pleasantness value and emotional state by PAD emotion scale, and the brain waves induced by the pictures of cultural elements with different pleasure degree were recorded by electroencephalograph. Then, the preference of cultural elements was obtained according to the theory of frontal alpha asymmetry. Secondly, the semantic difference method was used to carry out questionnaire survey to users, and the factor analysis method was used to statistically analyze the survey results to extract the perceptual image semantics of users for cultural and creative products. Thirdly, an interactive evolutionary design system based on IGA and BPNN was constructed. According to the cultural elements preferred by users, the designer designed the initial set of morphological characteristics, and the fitness value was determined according to the degree of user preference for the image semantics. Meanwhile, in order to reduce the fatigue caused by users’ interaction evaluation, BPNN was introduced to simulate artificial evaluation. Finally, the proposed method was verified by the practice of flavoring bottle design. User preference requirement could be used as feedback information to help designers understand users’ design emotional need and generate design schemes that satisfied the users’ perceptual image.
With the advent of experience economy, it is a key link of innovative product design to obtain user preferences quickly and accurately [
In the process of cultural and creative product design, many scholars have studied the extraction of cultural genes and design elements. For example, Gou et al. [
When a user is observing a product, the external visual stimulation will induce the changes of EEG. Analyzing EEG signals can accurately and objectively measure users’ perception, preference and emotion, and obtain users’ psychological needs. At present, EEG technology has been widely applied in the field of industrial design, such as commercial advertising design [
In the traditional cultural and creative product shape design, designers use personal experience and subjective speculation to obtain users’ emotional need. Because of the lack of scientific and objective evaluation mechanism, users’ image preference cannot be reflected in product design accurately and quickly. It is a research hotspot to integrate decision makers’ advantage into evolutionary design methods. Gong et al. [
In order to reduce the users’ fatigue during the interaction evaluation in IGA, this paper introduced neural network to assist IGA. Neural network is a nonlinear algorithm, which is often used to establish the relationship between complex input and output variables, and is successfully applied to the field of product shape design by perceptual image. Using the method of fuzzy neural network, Hsiao and Tsai [
This paper took the flavoring bottle design as an example to explore the innovative shape design scheme. Based on the cognition of users and designers, the product scheme was optimized by combining IGA and BPNN, to reflect the users’ emotional need in product design accurately.
As shown in Figure
A summary of the proposed method.
According to the different frequencies, EEG activity is divided into
PAD emotion scale can accurately evaluate the emotional state from three dimensions of pleasure, arousal, and dominance, and it has good structural validity. Therefore, this paper used PAD emotion scale to measure users’ emotion for different cultural elements [
PAD emotion scale consists of twelve pairs of adjectives representing different emotional state (see Figure
PAD emotion scale.
As shown in Figure
The samples of cultural elements from “Shu culture”. (a) Jinsha. (b) Sanxingdu. (c) Three kingdoms personages. (d) Panda. (e) Sichuan opera mask. (f) National costume.
According to the corresponding dimension of adjectives, the emotional values of the three dimensions were calculated by formulas (
The calculated data of PAD emotion scale.
Picture number | Pleasure |
Arousal |
Dominance |
Category of emotional state |
---|---|---|---|---|
1 | 2.750 | 2.375 | 0.255 | Type 1 |
2 | −2.175 | 2.825 | 1.750 | Type 8 |
3 | 1.750 | 1.750 | 0.015 | Type 1 |
4 | 1.825 | −1.250 | 0.250 | Type 5 |
5 | −2.375 | 3.015 | 1.875 | Type 8 |
6 | 3.000 | 2.750 | 2.125 | Type 1 |
According to the value of pleasure
Picture sample groups.
Group | The sample number | The value of pleasure range |
---|---|---|
Like | 1, 6 | [2, 4] |
General | 3, 4 | (−2, −2) |
Dislike | 2, 5 | [−4, −2] |
Subject
Twenty students in the department of industrial design were invited as subjects. They were required to have a good rest before the experiment, without taking medicine and drinking. A small gift was given as a reward after the experiment. Experimental environment and equipment
The experiment was carried out in the human-machine laboratory, the indoor temperature was suitable, and there was no noise interference. The experiment used the EEG equipment from German Brain Products, including amplifier, electrode cap, conductive paste and other hardware, stimulus presentation software E-Prime, data acquisition software Brain Vison Recorder, and analysis and processing software Brain Vision Analyzer 2.1. EEG recording
The cerebral cortex is mainly divided into frontal lobe, parietal lobe, occipital lobe, and temporal lobe. Each brain region contains a large number of neurons, taking on different tasks. The frontal lobe is mainly related to thinking, emotion, planning, and needs. Therefore, the frontal lobe was selected as the study object in this experiment. The EEG data of channels such as Experimental procedures
Position distribution of electrode.
The six picture samples were processed by the Photoshop software with uniform size and white background and the cultural elements expressed by black lines to exclude the influence of irrelevant factors such as color. The experimental process was presented by E-prime software. As shown in Figure Experimental steps
Experimental flow chart.
First, prepare for the work. Before the formal experiment, in order to eliminate the nervousness, the subjects should be informed of the experimental purpose, experimental process, and the matters needing attention in the experimental process. Secondly, the electrode cap was worn on the subject’s head, and it was placed in the way from front to back. The placement of the electrode is shown in Figure
The EEG power spectrum of the collected data was analyzed by Brain Vision Analyzer 2.1. The EEG signals evoked by the picture presentation 5000 ms were extracted. After digital filtering, segmentation, and operation, EEG data were quantitatively analyzed. Then, the average power of
The average power of
The average power of
The average power of
To sum up, under the stimulation of the favorite picture samples, the average power of
The power ratio of
Picture number |
|
|
|
Attribute |
---|---|---|---|---|
1 | 3.621 | 4.293 | 0.84 | Like |
2 | 4.028 | 3.527 | 1.14 | Dislike |
3 | 3.148 | 3.18 | 0.99 | General |
4 | 3.491 | 3.678 | 0.95 | General |
5 | 3.648 | 2.962 | 1.23 | Dislike |
6 | 3.778 | 5.135 | 0.74 | Like |
According to Kansei engineering, the following steps were used to obtain the image semantics of users on cultural and creative products:
The first was to select image vocabulary. The emotional preference for cultural and creative products could be abstracted as adjectives, and the selection of representative adjectives would affect the accuracy and authenticity of experimental results. The perceptual image vocabularies were used to describe the consumers’ cognitive and visual feelings about cultural and creative products. These vocabularies constituted the semantic set
The second was to use pictures to arouse the emotional image feelings of different dimensions of vocabularies and to strengthen the users’ perception of image scale.
Finally, the products and the perceptual image adjectives were evaluated interactively. The scale was designed according to the fuzzy statement “the perceptual cognitive preference
There were 10 to 20 image adjectives initially screened out. If there were too many image vocabularies, it was not conducive to the study and interpretation of user image, and it would increase the cognitive burden of user image evaluation. Therefore, factor analysis was used to extract a few comprehensive image vocabulary to reduce the cognitive dimension. Firstly, the relationship between variables was tested by Kaiser-Meyer-Olkin (KMO) test. The KMO value was closer to 1, the correlation between variables was stronger, and the original variables were suitable for factor analysis. Generally, if the KMO measure was greater than 0.7, and this analysis method could be adopted. Then, Bartlett test of sphericity was used to test the hypothesis that the correlation coefficient matrix was a unit matrix. If the original hypothesis could not be rejected, it could be considered that there was no significant difference between the correlation coefficient matrix and the unit matrix, and then the original variables were not suitable for factor analysis.
Through Internet, books, and other ways, fifteen adjectives were collected to express the image feelings of cultural and creative products (see Table
Image vocabulary.
1 | Unadorned |
2 | Fashionable |
3 | Succinct |
4 | Modern |
5 | Interesting |
6 | Elegant |
7 | Novel |
8 | Delicate |
9 | Fluent |
10 | Practical |
11 | Classical |
12 | Attractive |
13 | Unique |
14 | Personalized |
15 | Advanced |
As mentioned in Section
The experimental results were analyzed by SPSS statistics software 19.0. First by calculating the correlation coefficient matrix, anti-image correlation matrix, Bartlett’s test of sphericity, and Kaiser-Meyer-Olkin (KMO) test, the relationship between variables was tested. The statistics observed value was 341.662 in Bartlett’s test of sphericity; since the corresponding probability of
Four perceptual image semantic factors have been extracted by the principal component analysis method. The cumulative variance contribution rate is 79.348% (see Table
Total variance explained.
Component | Initial eigenvalues | Extraction sums of squared loadings | Rotation sums of squared loadings | ||||||
---|---|---|---|---|---|---|---|---|---|
Total | % of variance | Cumulative % | Total | % of variance | Cumulative % | Total | % of variance | Cumulative % | |
1 | 6.215 | 41.434 | 41.434 | 6.215 | 41.434 | 41.434 | 3.678 | 24.520 | 24.520 |
2 | 2.348 | 15.651 | 57.085 | 2.348 | 15.651 | 57.085 | 2.872 | 19.144 | 43.663 |
3 | 2.016 | 13.437 | 70.522 | 2.016 | 13.437 | 70.522 | 2.692 | 17.948 | 61.611 |
4 | 1.324 | 8.826 | 79.348 | 1.324 | 8.826 | 79.348 | 2.660 | 17.737 |
|
5 | 0.619 | 4.129 | 83.477 | ||||||
6 | 0.525 | 3.503 | 86.980 | ||||||
7 | 0.457 | 3.045 | 90.025 | ||||||
8 | 0.373 | 2.490 | 92.514 | ||||||
9 | 0.311 | 2.070 | 94.585 | ||||||
10 | 0.263 | 1.754 | 96.339 | ||||||
11 | 0.218 | 1.453 | 97.791 | ||||||
12 | 0.126 | 0.839 | 98.630 | ||||||
13 | 0.106 | 0.706 | 99.336 | ||||||
14 | 0.051 | 0.342 | 99.678 | ||||||
15 | 0.048 | 0.322 | 100.000 |
In Table
Rotated component matrix.
Component | ||||
---|---|---|---|---|
1 | 2 | 3 | 4 | |
Classical | − |
−0.129 | −0.167 | −0.098 |
Fashionable |
|
−0.205 | 0.551 | −0.012 |
Succinct |
|
−0.304 | 0.211 | 0.377 |
Modern |
|
−0.140 | 0.456 | 0.019 |
Fluent |
|
−0.173 | 0.300 | 0.205 |
Novel | −0.114 |
|
−0.167 | 0.021 |
Interesting | 0.121 |
|
−0.373 | −0.110 |
Personalized | −0.430 |
|
0.106 | −0.136 |
Unique | −0.648 |
|
0.143 | 0.028 |
Advanced | 0.219 | −0.135 |
|
0.199 |
Delicate | 0.294 | −0.008 |
|
−0.079 |
Elegant | 0.503 | −0.402 |
|
0.107 |
Practical | 0.054 | −0.223 | 0.009 |
|
Unadorned | 0.135 | −0.150 | 0.001 |
|
Attractive | 0.132 | 0.264 | 0.118 |
|
Based on the above analysis, the vocabularies with high load factor value were selected from each group to define users’ image demand for cultural and creative products: distinctive style, novel and interesting, exquisite texture, practical, and unadorned. In this way, the dimension of fifteen variables was reduced to four factors, and most information of the original variables could be reflected.
In the construction of evolutionary design framework of cultural and creative product, the shape of flavoring bottle was taken as the research object. And the shape elements of flavoring bottle were decomposed into bottle cap, bottle body, and decorative pattern.
The origin of cultural image was national costume that selected by the EEG experiment in Section
The characteristics of Yi nationality’s costume.
Headdress |
|
|
|
|
|
|
|
|
|
|
||||
Both men’s and women’s upper outer garment with buttons on the right |
|
|
|
|
|
|
|
||
|
||||
Men with trousers |
|
|
|
|
|
||||
Women with pleated skirts |
|
|
|
|
|
||||
Dressed in “Cha er wa” |
|
|
|
|
The most representative morphological characteristic of Yi nationality’s costume is the headdress for both men and women. They wear upper outer garment with buttons on the right, men wearing trousers and women wearing pleated skirt. Moreover, the Yi people always like to wear the “Cha er wa.” Most of the costume patterns come from the worship of the Yi people, which can be roughly divided into plants, animals, natural phenomena, social life, and geometric patterns. We invited designers to extract the characteristics of Yi nationality’s costumes from Table
The morphological characteristics of flavoring bottle.
Bottle cap |
|
|
|
|
|
|
|
|
|
||||||||
Bottle body |
|
|
|
|
|
|
|
|
|
||||||||
Decorative pattern |
|
|
|
|
|
|
|
|
Any complex product form was composed of three basic elements, point, line, and surface, which formed the morphological characteristics of flavoring bottle. According to the morphological method, the flavoring bottle was divided into three types of morphological element units according to its constituent elements, and multiple choices were provided in each type (eight choices were taken as examples in this paper). Combining all the choices, the optimal combination scheme could be selected among the numerous overall schemes. A morphological design model
Using the optimization ability of genetic algorithm, the morphological characteristics of flavoring bottle were coded to cross, mutate, and select, so as to get the flavoring bottle shape preferred by users. Through the interactive evaluation by users for perceptual image vocabulary, the fitness value of the individual was obtained. The product shape evolution was conducted as follows.
There are many forms of encoding. This paper adopted the most widely used binary encoding method proposed by Holland. In the encoding process, binary encoding was carried out for the morphological characteristics in each morphological element unit
The initial population of flavoring bottle shape.
|
|
|
|
||
|
|
|
010100100101 | 011101010110 | 001110000110 |
|
||
|
|
|
|
||
|
|
|
001100010001 | 010101010110 | 010001000111 |
Since it was difficult to describe the image perception by constructing function in traditional genetic algorithm, the users’ interactive evaluation of image vocabulary was used as the individual fitness value. Each user had different perceptual cognition of individual evaluation. In order to reduce the error caused by users’ undirected and uncertain fitness value evaluation, a certain weight should be given to different perceptual image vocabulary. The variance contribution of the factor reflected its explanatory ability to the total variance of the original variable. The higher the value was, the higher the importance of the corresponding factor was. Therefore, the variance contribution rate obtained from factor analysis in Section
During the interactive evaluation, the individuals generated by each generation needed to score and the user was easily fatigued. So, BPNN was introduced to simulate the users’ interactive evaluation, to solve the fatigue problem of IGA and improve the evolutionary efficiency.
BPNN belongs to forward network, which has the advantages of simple network structure, strong nonlinear approximation ability, fast convergence speed, and global convergence. This matched well with the problem to be solved in this paper, that is, the black box model composed of product morphological characteristics elements and user perceptual image. There was no accurate method to express the functional relationship between them. Therefore, it was a good choice to describe the relationship between product morphological characteristics elements and user perceptual image using neural network algorithm. This paper used feedforward BPNN as simulation evaluation model.
The neural network structure of morphological image evaluation model consisted of three layers: input layer, hidden layer, and output layer (see Figure
The neural network structure of morphological image evaluation model of flavoring bottle.
The most commonly used method to determine hidden nodes was the “trial and error method,” that was, setting fewer hidden nodes first, gradually increasing through experiments until the appropriate number was found. Some studies also agreed that when the number of neurons in the hidden layer was half of the total number of the nodes in input layer and output layer, the MSE value was also small. Therefore, this paper set the hidden nodes as
In general, the transfer function of hidden layer in BPNN was S-type function, and the output layer was linear function. In this paper, Tan-Sigmoid function was selected as the activation function of hidden layer. The range of this function was (−1, 1), and its function form was as follows:
The linear function Purelin was selected as the activation function of output layer, and its function form was as follows:
As shown in Figure
The flow of evolutionary algorithm.
Start with initial parameters setting. The initial population is generated randomly, and the number of each generation is set as popsize = 6, the search algebra is 50, the crossing rate is 0.7, and the mutation rate is 0.1.
The randomly generated design scheme of flavoring bottle shape is evaluated, and the system begins to train the neural network. If the decision maker feels tired, go to step 4; otherwise, go to step 3.
Determine whether meet the terminal condition of the algorithm. If so, output the optimal solution; otherwise, optimize, and go to step 2.
Determine whether the error meets the accuracy requirement. If it meets the requirement, then simulate the user evaluation through neural network, go to step 3; otherwise; go to step 2.
The terminal conditions of the algorithm are as follows:
When the user evaluation meets the following formula more than three generations in a row, the interactive genetic process is stopped, and the system outputs the optimal solution of flavoring bottle shape design.
When the interactive genetic evaluation reaches the preset termination algebra, that is, after 50 generations, the algorithm terminates, and the system outputs the optimal solution of flavoring bottle shape design.
The algorithm framework flow of this system is shown in Figure
Based on the design method proposed above, the shape design system of cultural and creative product was constructed, and a case study of flavoring bottle shape design was carried out. In the process of system construction, the neural network toolbox of Matlab was used to train, update, and apply the evaluation model.
Combined with the characteristics of the problem to be solved in this paper, the function type and parameters have been determined.
The trainlm () was set as the training function. In Matlab toolbox, trainlm () is a forward network function trained by Levenberg–Marguardt rules, and it is an optimized training algorithm combining the advantages of functions trainbp () and trainbpx ().
The learngdm () was set as the learning function of neural network, and the initial value of the learning rate was set as 0.05. The learngdm () is a gradient descent learning function with additional momentum factors in Matlab, which improves the speed of learning and the reliability of the algorithm by introducing the method of changing weight of momentum factors.
S-type function (Tan-Sigmoid) and linear function (Purelin) were selected as the activation functions of hidden layer and output layer, and their functions were expressed as tansig () and purelin () in Matlab.
The mean square error (MSE) was set as the performance evaluation function. The function expression was as follows:
In this paper, by calling neural network model in Matlab, a shape design system for cultural and creative product was constructed. The system design interface is shown in Figure
Evolutionary design system.
For the generated individuals (flavoring bottle schemes), decision maker needed to score each scheme according to five-point scale method, and the system automatically calculated the fitness value of each scheme. As shown in Figure
Five users were invited to test the system, and 4.5 points was set as the target fitness value. As shown in Figure
The mean error precision value.
In the interactive evaluation process, after 14 generations of artificial evaluation, if the user felt fatigued, automatic evaluation could be selected. If the user did not feel tired, they could continue to evaluate the bottle shape until a satisfactory solution was obtained or the maximum number of evaluations was reached.
It could be concluded from the above tests that users could select automatic evaluation after interactive evaluation for 14 generations, and the neural network simulated the remaining evaluations (from the 15th generation to the 40th generation), that is, 14 ∗ 6 = 84 fitness values were evaluated manually, and 26 ∗ 6 = 156 fitness values were evaluated by the neural network. Compared with IGA, IGA integrated with neural network could reduce user fatigue and improve solution quality and speed.
A male and a female target user obtained satisfactory solution using this evolutionary system (see Figure
The schemes after system evolution.
The flavoring bottle design sketch.
Based on the morphological foundation obtained by the system evolution, the detailed design was carried out. The scheme in Figure
The text design.
The 3D printing model.
The method proposed in this paper can help designers to design cultural and creative products that conform to the users’ target image. However, there are still some shortcomings in this study. The feasibility of the method was verified by taking the flavoring bottle design as an example. In the future, more cultural products need to be refined, and models of all kinds of household products should be established for designers to reference in creative and cultural product design. The relationship between the characteristics in frequency domain and the degree of pleasure under the stimulus of cultural elements with different forms was explored, which only analyzed the characteristics in frequency domain, not involved in the time domain feature, amplitude, and composition and did not consider the effect of color factor and culture factor. The next step of research is to dig deeper into this aspect. As an exploratory study, this paper used a relatively basic interactive genetic algorithm and neural network algorithm to verify the feasibility of the system. In the future, we need to compare the evolutionary effect of different algorithms in the cultural and creative product design and further optimize the algorithm to improve system efficiency and product design quality. The disassembly of flavoring bottle shape was relatively simple, and its morphological characteristics were not very rich. In the future research, we can establish a gene bank of morphological characteristics of cultural and creative products, which is conducive to the evolution of better products. The current system generated two-dimensional wireframe pattern. In the follow-up work, the design scheme can be three-dimensional by adding other programs and tools.
The result of EEG experiment showed that the frontal Through the user perceptual image investigation, we have realized the research on the perceptual positioning of cultural and creative products. To quantitatively analyze the degree of user preference for the image semantics and master the image feelings of the users for the cultural and creative products, which is conducive to further product shape design and optimization. Based on the cultural elements obtained from the EEG experiment, the morphological characteristics were coded to crossed, mutated, and selected, and the individual fitness value was obtained through the users’ interactive evaluation for perceptual image vocabulary, so as to conduct product shape evolution. In order to increase the evolutionary efficiency and improve the fatigue error caused by IGA, BPNN was introduced to simulate users’ interaction evaluation. Finally, the evolutionary design system was constructed, and the proposed method was verified by the example of flavoring bottle design.
The data used to support the findings of this study are included within the article.
The authors declare that they have no conflicts of interest.
This work was supported by the Sub-project of National Key Research and Development program under grant no. 2018YFC0310201-08 and Nanchong Social Sciences Research 13th Five-Year Plan under grant no. NC2017C050.