Electroencephalogram aesthetic evaluation processing method and device for cross-subject scene, medium and terminal
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHENZHEN UNIV
- Filing Date
- 2022-09-06
- Publication Date
- 2026-05-22
AI Technical Summary
The current aesthetic evaluation process suffers from excessive human interference, resulting in inaccurate and unobjective evaluation results.
By selecting a training set of images with real rating labels from a pre-set aesthetic dataset, collecting raw EEG data of the target object, performing preprocessing and feature calculation, constructing an initial aesthetic evaluation model, and obtaining a mature aesthetic evaluation model through deep learning training, and finally performing aesthetic evaluation and classification based on this model.
It enhances the objectivity of aesthetic assessment, reduces human interference, improves the accuracy of assessment, and enables aesthetic assessment across subject scenarios, applicable to multiple individuals, and can provide corresponding aesthetic category suggestions for target images.
Smart Images
Figure CN115690528B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of aesthetic evaluation, and more particularly to a method, apparatus, medium, and terminal for aesthetic evaluation and processing of electroencephalogram (EEG) signals across different subjects and scenarios. Background Technology
[0002] The human body's perception of aesthetics is also an emotional process; human perception of aesthetics has a strong emotional component. Take artworks, which fall under the category of aesthetics, as an example. During their creation, the creator is driven by certain emotional motivations and hopes that the viewer can empathize with the artist. The aesthetic evaluation of an image is closely related to whether it is visually appealing, while the emotional evaluation it evokes is related to whether the target audience experiences pleasure when viewing it. Therefore, the aesthetic evaluation of an image is an extremely subjective matter. On the other hand, the emotions evoked by viewing emotionally charged images can be divided into positive and negative emotions. Humans can experience positive emotions when viewing aesthetically pleasing images, which is beneficial to their physical and mental health. Therefore, aesthetics and emotion are interconnected. .
[0003] The human brain plays an important role in the generation and activity of emotions. With the increasing maturity of EEG emotion recognition research, EEG signals collected from the brain are a non-invasive and reliable signal that can detect information related to changes in emotional state. Moreover, the emotion recognition method based on EEG signals has good objectivity and high temporal resolution. Aesthetics can stimulate emotions, and emotional activities can be captured by corresponding feedback from human EEG signals.
[0004] Most current aesthetic evaluation processes are conducted manually, which leads to excessive interference from human factors and results in inaccurate aesthetic evaluations. Confirm . Summary of the Invention
[0005] In view of the shortcomings of the prior art, the purpose of this application is to provide a method, device, medium and terminal for aesthetic evaluation of EEG signals across subject scenarios, which aims to solve the problem that there is too much interference from human factors in the aesthetic evaluation process and the aesthetic evaluation results are not accurate enough.
[0006] To address the aforementioned technical problems, the first aspect of this application provides a method for aesthetic evaluation and processing of EEG signals across different subject scenarios, the method comprising:
[0007] Several aesthetic images with real rating labels were selected from the pre-set aesthetic dataset as the image training set;
[0008] Raw EEG data of the target subjects as they viewed the aesthetic images were collected in chronological order.
[0009] The raw EEG data is preprocessed and feature calculated to obtain EEG training data.
[0010] The initial aesthetic evaluation model was trained based on the EEG training data and the image training set to obtain a mature aesthetic evaluation model.
[0011] The target image is aesthetically evaluated based on the mature aesthetic evaluation model. The EEG data of the target image is processed and input into the mature aesthetic evaluation model for calculation to obtain the aesthetic score of the target image. Based on the aesthetic score, the target image is given the corresponding aesthetic category suggestion.
[0012] As a further improvement to the technical solution, the step of selecting several aesthetic images with real rating labels from a preset aesthetic dataset as an image training set includes:
[0013] The aesthetic image dataset that simultaneously has emotion rating labels and aesthetic rating labels is used as the preset aesthetic dataset;
[0014] Aesthetic images are selected from the preset aesthetic dataset as an image training set, wherein the emotion rating labels of the aesthetic images in the image training set follow a Gaussian distribution.
[0015] As a further improved technical solution, the step of collecting raw EEG data of the target object viewing the aesthetic image in chronological order includes:
[0016] Raw EEG data of the target object were collected in chronological order as it viewed a preset number of aesthetic images within a preset time period, wherein the viewing time for each aesthetic image was the same.
[0017] As a further improved technical solution, the step of preprocessing and feature calculation of the raw EEG data to obtain EEG training data includes:
[0018] The raw EEG data was filtered sequentially using a Butterworth zero-phase hysteresis bandpass filter and a Butterworth zero-phase hysteresis notch filter to obtain the processed EEG data.
[0019] Two features, differential entropy and power spectral density, are extracted from the processed EEG data to obtain EEG training data.
[0020] As a further improved technical solution, the step of training the initial aesthetic evaluation model based on the EEG training data and the image training set to obtain a mature aesthetic evaluation model includes:
[0021] Construct an initial aesthetic evaluation model, wherein the initial aesthetic evaluation model includes a common feature extractor, a domain feature extractor, and a domain classifier;
[0022] One of the EEG data in the EEG training data is taken as the target domain data, and the EEG data other than the target domain data is taken as the source domain data. The target domain data and the source domain data are input into a common feature extractor for extraction to obtain domain-invariant features.
[0023] The domain-invariant features of each source domain data are paired with the domain-invariant features of the target domain data, and then input into the domain feature extractor to obtain the corresponding domain features.
[0024] The domain classifier is trained based on the domain features and the true rating labels to obtain a mature aesthetic evaluation model.
[0025] As a further improved technical solution, the step of training the domain classifier based on the domain features and the true rating labels to obtain a mature aesthetic evaluation model includes:
[0026] The domain features are input into the domain classifier for calculation to obtain the predicted value;
[0027] The predicted value is compared with the true value in the true rating label. If the predicted value is less than or greater than the true value, the classifier continues to be trained. If the predicted value is equal to the true value, a mature aesthetic evaluation model is output.
[0028] As a further improved technical solution, the aesthetic evaluation of the target image based on the mature aesthetic evaluation model involves processing the EEG data while viewing the target image and inputting it into the mature aesthetic evaluation model for calculation to obtain an aesthetic score for the target image. Based on the aesthetic score, corresponding aesthetic category suggestions are given to the target image, including:
[0029] Based on the mature aesthetic evaluation model, the target image is aesthetically evaluated. EEG data is collected when viewing the target image. The EEG data is preprocessed and feature calculated sequentially to obtain EEG feature data.
[0030] The EEG feature data is input into the mature aesthetic evaluation model for calculation to obtain the aesthetic score of the target image, and the target image is given corresponding aesthetic category suggestions based on the aesthetic score.
[0031] A second aspect of this application provides a cross-subject scene electroencephalogram (EEG) signal aesthetic evaluation device, comprising:
[0032] The filtering module is used to select a number of aesthetic images with real rating labels from the preset aesthetic dataset as the image training set;
[0033] The acquisition module is used to collect raw EEG data of the target object while viewing the aesthetic images in chronological order.
[0034] The preprocessing module is used to preprocess and perform feature calculations on the raw EEG data to obtain EEG training data;
[0035] The training module is used to train the initial aesthetic evaluation model based on the EEG training data and the image training set to obtain a mature aesthetic evaluation model.
[0036] The evaluation module is used to evaluate the target image based on the mature aesthetic evaluation model. It processes the EEG data when viewing the target image and inputs it into the mature aesthetic evaluation model for calculation to obtain the aesthetic score of the target image. Based on the aesthetic score, it provides the target image with corresponding aesthetic category suggestions.
[0037] A third aspect of this application provides a computer-readable storage medium storing one or more programs that can be executed by one or more processors to implement the steps in the cross-subject scene electroencephalogram signal aesthetic evaluation processing method as described above.
[0038] A fourth aspect of this application provides a terminal device, which includes: a processor, a memory, and a communication bus; the memory stores a computer-readable program that can be executed by the processor;
[0039] The communication bus enables communication between the processor and the memory;
[0040] When the processor executes the computer-readable program, it implements the steps in the cross-subject scene EEG signal aesthetic evaluation processing method as described above.
[0041] Beneficial Effects: Compared with existing technologies, the cross-subject scene EEG signal aesthetic evaluation processing method of the present invention includes: selecting several aesthetic images with real rating labels from a preset aesthetic dataset as an image training set; collecting raw EEG data of the target object when viewing the aesthetic images in chronological order; preprocessing and feature calculation of the raw EEG data to obtain EEG training data; training an initial aesthetic evaluation model based on the EEG training data and the image training set to obtain a mature aesthetic evaluation model; performing aesthetic evaluation on the target image based on the mature aesthetic evaluation model, processing the EEG data when viewing the target image and inputting it into the mature aesthetic evaluation model for calculation to obtain the aesthetic score of the target image, and giving the target image corresponding aesthetic category suggestions based on the aesthetic score; after adopting the above method, the present invention captures information related to aesthetic perception activities through EEG signals, enhances the objectivity of the aesthetic evaluation process, reduces human interference, and improves the accuracy of aesthetic evaluation. Through deep learning technology, the computational connection between EEG signals and aesthetic perception activities is obtained, and this computational connection is not only applicable to a single individual, but can be transformed into a group commonality. Attached Figure Description
[0042] Figure 1 This is a flowchart of the aesthetic evaluation and processing method for EEG signals across different subject scenarios according to the present invention.
[0043] Figure 2 This is a structural schematic diagram of the terminal device provided by the present invention.
[0044] Figure 3 This is a structural block diagram of the device provided by the present invention.
[0045] Figure 4 This is a flowchart of the aesthetic evaluation and processing method for EEG signals across different subject scenarios according to the present invention.
[0046] Figure 5 This is a framework diagram of the aesthetic evaluation model for the cross-subject scene aesthetic evaluation processing method of EEG signals of the present invention.
[0047] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation
[0048] To facilitate understanding of this application, a more complete description will be provided below with reference to the accompanying drawings. Preferred embodiments of this application are shown in the drawings. However, this application can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the disclosure of this application.
[0049] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein in the specification of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application.
[0050] The inventors discovered through research that the existing technology has the following problems:
[0051] (1) The human brain plays an important role in the generation and activity of emotions. With the increasing maturity of EEG emotion recognition research, the EEG signals collected from the brain are a non-invasive and reliable signal that can detect information related to changes in emotional state. Moreover, the EEG signal emotion recognition method has good objectivity and high temporal resolution. Aesthetics can stimulate emotions, and emotional activities can capture corresponding feedback in human EEG signals.
[0052] The current aesthetic evaluation process suffers from excessive human interference, is too subjective, and lacks objectivity, resulting in unsatisfactory final evaluation results.
[0053] (2) Information related to aesthetic perception exists in human brain activity, but current research is mostly explained from the perspective of neuroscience, and no research has yet explored the relationship between the two from a computational perspective. Early research on aesthetic preferences was mainly from a psychological perspective.
[0054] Different individuals' EEG data follow different marginal distributions, which is a challenge in using EEG signals for aesthetic evaluation in cross-subject scenarios. Cross-subject research conditions are also a hot topic in EEG emotion recognition research. How to improve the generalization ability of EEG emotion classification models on samples from different individuals is one of the most challenging frontier directions in the field of EEG emotion recognition.
[0055] like Figure 1 As shown in the embodiment of this application, a method for aesthetic evaluation and processing of EEG signals across subject scenarios includes the following steps:
[0056] S1, Select several aesthetic images with real rating labels from the preset aesthetic dataset as the image training set;
[0057] Specifically, the preset aesthetic dataset is an aesthetic image dataset with emotion and aesthetic annotations. This dataset was obtained through rigorous and extensive statistical experiments. The scenes in the aesthetic images are all natural landscapes. Several aesthetic images were selected from the preset aesthetic dataset according to the Gaussian distribution of emotion scores as the image training set.
[0058] The step of selecting a number of aesthetic images with real rating labels from a preset aesthetic dataset as an image training set includes the following steps:
[0059] S101, The aesthetic image dataset that simultaneously has emotion rating labels and aesthetic rating labels is used as the preset aesthetic dataset;
[0060] S102, Select aesthetic images from the preset aesthetic dataset as an image training set, wherein the emotion rating labels of the aesthetic images in the image training set follow a Gaussian distribution.
[0061] The pre-defined aesthetic dataset is an aesthetic image dataset with both emotion and aesthetic annotations. This dataset was obtained through rigorous and extensive statistical experiments. The aesthetic images consist of natural landscapes, and there are 20,994 images in total. Each image was rated by 20 different people on an aesthetic and emotion scale from 1 to 7. The average score was then used as the aesthetic and emotion score for the corresponding image. Both the aesthetic and emotion scores are based on a score of 4. Whether analyzed from the perspective of aesthetic or emotion scores, the overall scores show a Gaussian distribution, which conforms to the general laws of statistics. The aesthetic and emotion scores are used as the aesthetic rating labels and emotion rating labels for the aesthetic images.
[0062] Ten sets of aesthetic images were selected from a pre-defined aesthetic dataset. Each set contained 300 aesthetic images, and the emotion scores of these 300 aesthetic images followed a Gaussian distribution, with aesthetic scores ranging from 1 to 7. These ten sets of aesthetic images were used as the image training set.
[0063] S2, Collect raw EEG data of the target object when viewing the aesthetic image in chronological order;
[0064] Specifically, the target subject is instructed to focus on the perceptual activity of aesthetic stimuli, and raw EEG data of the target subject as they view the aesthetic images are collected in chronological order.
[0065] The raw EEG data collected from the target object while viewing the aesthetic images in chronological order includes:
[0066] Raw EEG data of the target object were collected in chronological order as it viewed a preset number of aesthetic images within a preset time period, wherein the viewing time for each aesthetic image was the same.
[0067] The target subjects were humans, with a total of 10 participants. Subjects were required to have sufficient sleep and rest in the days leading up to the data collection to maintain a good mental state and avoid affecting the quality of the collected EEG signals. The total collection time for each subject was 30 minutes, with each individual collection session lasting 10 minutes. Therefore, each subject required 3 collection sessions, with each session spaced no more than 12 hours apart, to ensure that the EEG signals of the subjects conformed to a nearly uniform marginal distribution. The sampling frequency of the EEG equipment was set to 500 Hz. The EEG equipment and data processing platform used in this data collection were provided by IMOTION.
[0068] During the data collection, the target subjects were instructed to wear the EEG device correctly and then view randomly presented aesthetic images on a computer screen. Viewers were only required to consider whether the image was beautiful and whether it made them happy or unhappy. Each image was displayed for 6 seconds, and each target subject viewed a total of 300 aesthetic images. In a single 10-minute experiment, 100 aesthetic images were randomly presented. These 100 images were randomly divided into two groups of 50 each. Before the images in each group were displayed, a completely white image was shown for 30 seconds. This allowed the target subjects to adjust their mood, clear their minds, and focus their attention on the experiment. This 30-second window also served as a time for the target subjects to adjust to a comfortable state for observing the subsequent aesthetic images. The entire data collection process was conducted under conditions of no other light interference. During the playback of the aesthetic images, due to the sensitivity of EEG signals, the target subjects were required to remain as still as possible.
[0069] S3, preprocess and perform feature calculations on the raw EEG data to obtain EEG training data;
[0070] The collected raw EEG data were preprocessed and feature calculated sequentially to extract precise EEG training data for model training.
[0071] The step of preprocessing and feature calculation of the raw EEG data to obtain EEG training data includes the following steps:
[0072] S301, Butterworth zero-phase hysteresis bandpass filter and Butterworth zero-phase hysteresis notch filter are used sequentially to filter the raw EEG data to obtain the processed EEG data;
[0073] S302, Extract the differential entropy and power spectral density features from the processed EEG data to obtain EEG training data.
[0074] Specifically, after collecting the raw EEG data, it is necessary to perform certain preprocessing. Due to the sensitivity of EEG, the raw data contains a lot of noise, so preprocessing is required. During the preprocessing process, Butterworth zero-phase hysteresis bandpass filter and Butterworth zero-phase hysteresis notch filter are used to filter the raw EEG data. This is mainly to filter out signals that interfere with the EEG signal, such as eyelid twitching or changes in sitting posture during the acquisition process. In addition, an artifact threshold needs to be set to reject data points whose absolute value exceeds the preset voltage threshold, so as to ensure the quality of the acquired EEG signal. The preprocessed data is then obtained.
[0075] Next, relevant features are extracted from the processed EEG data, primarily the differential entropy and power spectral density of the EEG signal. Feature extraction is performed according to five frequency bands: delta, theta, alpha, beta, and gamma. The differential entropy and power spectral density formulas are used to calculate the features. During feature calculation, the selection of the time window for the EEG waveform also needs to be appropriate. For a single target object viewing an aesthetic image, the acquired EEG duration is 6 seconds. Therefore, moving time windows of 3 seconds, 4 seconds, 5 seconds, and 6 seconds are selected respectively to calculate the differential entropy and power spectral density features, ultimately obtaining the EEG training data.
[0076] The formula for calculating differential entropy is as follows:
[0077]
[0078] In the formula for calculating differential entropy, The probability density function representing EEG information. This indicates the range of values for EEG information.
[0079] The formula for calculating power spectral density is shown below:
[0080]
[0081] In the formula for calculating the power spectral density characteristics, This indicates the collected electroencephalogram (EEG) signals. yes The continuous Fourier transform.
[0082] S4, the initial aesthetic evaluation model is trained based on the EEG training data and the image training set to obtain a mature aesthetic evaluation model;
[0083] The EEG training data is used as input to the initial aesthetic evaluation model. The predicted value is obtained after calculation by the initial aesthetic evaluation model. The emotion rating label and aesthetic rating label of the image training set are used as the real rating label. The predicted value is compared with the real value in the real rating label. If the predicted value is less than or greater than the real value, the classifier continues to be trained. If the predicted value is equal to the real value, the mature aesthetic evaluation model is output.
[0084] The step of training the initial aesthetic evaluation model based on the EEG training data and the image training set to obtain a mature aesthetic evaluation model includes the following steps:
[0085] S401, Construct an initial aesthetic evaluation model, wherein the initial aesthetic evaluation model includes a common feature extractor, a domain feature extractor, and a domain classifier;
[0086] S402, take one of the EEG data in the EEG training data as the target domain data, take the EEG data other than the target domain data as the source domain data, and input the target domain data and the source domain data into the common feature extractor for extraction to obtain domain-invariant features.
[0087] S403, perform pairwise matching of the domain-invariant features of each source domain data with the domain-invariant features of the target domain data, and input the matching data into the domain feature extractor to obtain the corresponding domain features;
[0088] S404, The domain classifier is trained based on the domain features and the true rating labels to obtain a mature aesthetic evaluation model.
[0089] The process of training the domain classifier based on the domain features and the true rating labels to obtain a mature aesthetic evaluation model includes the following steps:
[0090] The domain features are input into the domain classifier for calculation to obtain the predicted value;
[0091] The predicted value is compared with the true value in the true rating label. If the predicted value is less than or greater than the true value, the classifier continues to be trained. If the predicted value is equal to the true value, a mature aesthetic evaluation model is output.
[0092] Specifically, the image training set used in this embodiment is scored from 1 to 7 for aesthetics. During the training process, aesthetic images belonging to the score interval [1,3) are defined as not beautiful, aesthetic images belonging to the score interval [3,5) are defined as average, and aesthetic images belonging to the score interval [5,7] are defined as beautiful. Therefore, the aesthetic evaluation task performed in this embodiment is essentially a three-classification task from the perspective of deep learning. Thus, the initial aesthetic evaluation model mainly includes three parts: a common feature extractor, a domain feature extractor, and a domain classifier.
[0093] During the training process of the model, the EEG data of one experimental subject in the EEG training data is used as the target domain data, and the EEG data of other experimental subjects is used as the source domain data. The target domain data and the source domain data are input into a common feature extractor for extraction to obtain domain-invariant features.
[0094] The domain-invariant features of each source domain data are paired with the domain-invariant features of the target domain data, and then fed into the corresponding domain feature extractor to obtain the domain features of the target and source domains obtained from the same domain feature extractor. The formula calculates the distance between the two in deep space, reducing This is to make the target domain and the source domain closer in the feature space, thus helping the target domain make better predictions. The formula is as follows:
[0095]
[0096] in Represents a mapping function. Represents the regenerating nucleus Hilbert space. Represents an instance matrix consisting of data from each source domain. The eigenvectors in the model. Represents an instance matrix composed of target domain data. The feature vector in This represents the data for each source domain. This represents the target domain data. Indicates the quantity of data in the source domain. Indicates the number of data in the target domain.
[0097] The obtained domain features are input into the corresponding multivariate classifier. For the training of each classifier, the cross-entropy defined by the following equation is used to evaluate the classifier's loss. This classifier is used The training is performed using data from multiple source domains, as shown in the following equation:
[0098]
[0099] in This represents the eigenvectors in an instance matrix composed of source domain data. This represents the data for each source domain. This represents the predicted label matrix for each source domain data. This represents a matrix consisting of real rating labels for each source domain. Represents the mathematical expectation. This represents cross-entropy.
[0100] If this The average of the classification results is the prediction result, which can lead to excessive variance, especially when the samples in the target domain lie on the decision boundary. To reduce variance, a measure of difference loss is introduced, making... When the predictions of the classifiers converge, the difference loss is defined by the following equation:
[0101]
[0102] in Represents an instance matrix composed of source domain data. The feature vector in This represents the data for each source domain. This represents the predicted label matrix of the target domain data. This represents a matrix composed of the true rating labels of the target domain. It represents the mathematical expectation.
[0103] reduce To better extract the domain features of each pair of source and target domains, reducing the classifier's loss will result in better predictions of the source domain. Reducing the difference loss is to obtain more convergent prediction results. The training of the overall model is based on the final overall loss, defined by the following equation:
[0104]
[0105] in and These are the hyperparameters that need to be set during training. This refers to classification loss. It means , This refers to difference loss.
[0106] S5. Based on the mature aesthetic evaluation model, the target image is aesthetically evaluated. The EEG data when viewing the target image is processed and input into the mature aesthetic evaluation model for calculation to obtain the aesthetic score of the target image. Based on the aesthetic score, the target image is given corresponding aesthetic category suggestions.
[0107] The step of conducting an aesthetic evaluation of the target image based on the mature aesthetic evaluation model, which involves processing the EEG data while viewing the target image and inputting it into the mature aesthetic evaluation model for calculation to obtain an aesthetic score for the target image, and then providing corresponding aesthetic category suggestions for the target image based on the aesthetic score, includes the following steps:
[0108] S501, Based on the mature aesthetic evaluation model, perform aesthetic evaluation on the target image, collect EEG data when viewing the target image, and perform preprocessing and feature calculation on the EEG data in sequence to obtain EEG feature data;
[0109] S502, the EEG feature data is input into the mature aesthetic evaluation model for calculation to obtain the aesthetic score of the target image, and the target image is given corresponding aesthetic category suggestions based on the aesthetic score.
[0110] Specifically, when conducting aesthetic evaluation of a target image, it is necessary to collect the user's electroencephalogram (EEG) data while viewing the target image. The EEG data is then preprocessed and feature-calculated in step S3 to obtain EEG feature data. This EEG feature data is then input into the mature aesthetic evaluation model for calculation. This model is applicable to cross-subject experimental scenarios, meaning that the EEG feature data of multiple experimental subjects can be input simultaneously. The aesthetic category of the aesthetic image viewed at that time is obtained from the EEG data of the user viewing the aesthetic image, namely, three aesthetic categories: not beautiful, average, and beautiful.
[0111] Based on the above-mentioned cross-subject scene EEG signal aesthetic evaluation processing method, this embodiment provides a cross-subject scene EEG signal aesthetic evaluation device, including:
[0112] Filtering module 1 is used to select a number of aesthetic images with real rating labels from the preset aesthetic dataset as the image training set;
[0113] Acquisition module 2 is used to collect raw EEG data of the target object while viewing the aesthetic images in chronological order;
[0114] Preprocessing module 3 is used to preprocess and perform feature calculations on the raw EEG data to obtain EEG training data;
[0115] Training module 4 is used to train the initial aesthetic evaluation model based on the EEG training data and the image training set to obtain a mature aesthetic evaluation model;
[0116] Evaluation module 5 is used to evaluate the target image based on the mature aesthetic evaluation model. It processes the EEG data when viewing the target image and inputs it into the mature aesthetic evaluation model for calculation to obtain the aesthetic score of the target image. Based on the aesthetic score, it provides the target image with corresponding aesthetic category suggestions.
[0117] Furthermore, it is worth noting that the working process of the EEG signal aesthetic evaluation device based on cross-subject scene provided in this embodiment is the same as the working process of the EEG signal aesthetic evaluation processing method based on cross-subject scene described above. For details, please refer to the working process of the EEG signal aesthetic evaluation processing method based on cross-subject scene, which will not be repeated here.
[0118] Based on the above-described method for aesthetic evaluation and processing of EEG signals across different subjects, this embodiment provides a computer-readable storage medium storing one or more programs that can be executed by one or more processors to implement the steps in the method for aesthetic evaluation and processing of EEG signals across different subjects as described in the above embodiment.
[0119] like Figure 2 As shown, based on the above-described cross-subject scenario EEG signal aesthetic evaluation processing method, this application also provides a terminal device, which includes at least one processor 20; a display screen 21; and a memory 22, and may also include a communications interface 23 and a bus 24. The processor 20, display screen 21, memory 22, and communications interface 23 can communicate with each other via the bus 24. The display screen 21 is configured to display a preset user guidance interface in the initial setup mode. The communications interface 23 can transmit information. The processor 20 can call logical instructions in the memory 22 to execute the methods described in the above embodiments.
[0120] Furthermore, the logical instructions in the aforementioned memory 22 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium.
[0121] The memory 22, as a computer-readable storage medium, can be configured to store software programs, computer-executable programs, such as program instructions or modules corresponding to the methods in the embodiments of this disclosure. The processor 20 executes functional applications and data processing by running the software programs, instructions, or modules stored in the memory 22, thereby implementing the methods in the above embodiments.
[0122] The memory 22 may include a program storage area and a data storage area. The program storage area may store the operating system and application programs required for at least one function; the data storage area may store data created based on the use of the terminal device. Furthermore, the memory 22 may include high-speed random access memory (RAM) and non-volatile memory. Examples include various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks, as well as transient storage media.
[0123] Compared to existing technologies, the present invention's cross-subject scene EEG signal aesthetic evaluation processing method includes: selecting several aesthetic images with real rating labels from a preset aesthetic dataset as an image training set; collecting raw EEG data of the target object while viewing the aesthetic images in chronological order; preprocessing and feature calculation of the raw EEG data to obtain EEG training data; training an initial aesthetic evaluation model based on the EEG training data and the image training set to obtain a mature aesthetic evaluation model; and performing aesthetic evaluation on the target image based on the mature aesthetic evaluation model, by processing the EEG data while viewing the target image and inputting it into the mature aesthetic evaluation model. The method calculates and obtains an aesthetic score for the target image, and provides corresponding aesthetic category suggestions based on the aesthetic score. This invention uses EEG signals to capture information related to aesthetic perception, enhancing the objectivity of the aesthetic evaluation process, reducing human interference, and improving the accuracy of aesthetic evaluation. Through deep learning technology, it obtains a computational connection between EEG signals and aesthetic perception. This computational connection is not limited to a single individual; it can be transformed into a commonality among groups. Appreciating aesthetics can sometimes have a soothing effect on the soul. This method can achieve the effect of soothing the soul by observing EEG signals and using appropriate aesthetic content.
[0124] It should be understood that the application of the present invention is not limited to the examples above. Those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.
Claims
1. A method for aesthetic evaluation and processing of EEG signals across different subject scenarios, characterized in that, include: Several aesthetic images with real rating labels were selected from the pre-set aesthetic dataset as the image training set; Raw EEG data of the target subjects as they viewed the aesthetic images were collected in chronological order. The raw EEG data is preprocessed and feature calculated to obtain EEG training data. The initial aesthetic evaluation model was trained based on the EEG training data and the image training set to obtain a mature aesthetic evaluation model. The target image is evaluated based on the mature aesthetic evaluation model. The EEG data of the target image is processed and input into the mature aesthetic evaluation model for calculation to obtain the aesthetic score of the target image. Based on the aesthetic score, the target image is given the corresponding aesthetic category suggestion. The process of training the initial aesthetic evaluation model based on the EEG training data and the image training set to obtain a mature aesthetic evaluation model includes: Construct an initial aesthetic evaluation model, wherein the initial aesthetic evaluation model includes a common feature extractor, a domain feature extractor, and a domain classifier; One of the EEG data in the EEG training data is taken as the target domain data, and the EEG data other than the target domain data is taken as the source domain data. The target domain data and the source domain data are input into a common feature extractor for extraction to obtain domain-invariant features. The domain-invariant features of each source domain data are paired with the domain-invariant features of the target domain data and input into the domain feature extractor to obtain the corresponding domain features. The distance between the domain-invariant features of the source domain data and the domain-invariant features of the target domain data in the deep space is calculated using the MMD loss formula to reduce the MMD loss and better extract the domain features of each pair of source and target domains. The domain classifier is trained based on the domain features and the true rating labels to obtain a mature aesthetic evaluation model.
2. The method for aesthetic evaluation and processing of EEG signals across different subjects in a scene according to claim 1, characterized in that, The step of selecting a number of aesthetic images with real rating labels from a preset aesthetic dataset as the image training set includes: The aesthetic image dataset that simultaneously has emotion rating labels and aesthetic rating labels is used as the preset aesthetic dataset; Aesthetic images are selected from the preset aesthetic dataset as an image training set, wherein the emotion rating labels of the aesthetic images in the image training set follow a Gaussian distribution.
3. The method for aesthetic evaluation and processing of EEG signals across different subjects in a scene, as described in claim 1, is characterized in that... The raw EEG data collected sequentially from the target subject while viewing the aesthetic images includes: Raw EEG data of the target object were collected in chronological order as it viewed a preset number of aesthetic images within a preset time period, wherein the viewing time for each aesthetic image was the same.
4. The method for aesthetic evaluation and processing of EEG signals across different subjects in a scene according to claim 1, characterized in that, The step of preprocessing and feature calculation of the raw EEG data to obtain EEG training data includes: The raw EEG data was filtered sequentially using a Butterworth zero-phase hysteresis bandpass filter and a Butterworth zero-phase hysteresis notch filter to obtain the processed EEG data. Two features, differential entropy and power spectral density, are extracted from the processed EEG data to obtain EEG training data.
5. The method for aesthetic evaluation and processing of EEG signals across different subjects in a scene according to claim 1, characterized in that, The process of training the domain classifier based on the domain features and the true rating labels to obtain a mature aesthetic evaluation model includes: The domain features are input into the domain classifier for calculation to obtain the predicted value; The predicted value is compared with the true value in the true rating label. If the predicted value is less than or greater than the true value, the classifier continues to be trained. If the predicted value is equal to the true value, a mature aesthetic evaluation model is output.
6. The method for aesthetic evaluation and processing of EEG signals across different subjects in a scene according to claim 1, characterized in that, The aesthetic evaluation of the target image based on the mature aesthetic evaluation model involves processing the EEG data while viewing the target image and inputting it into the mature aesthetic evaluation model for calculation to obtain an aesthetic score for the target image. Based on the aesthetic score, corresponding aesthetic category suggestions are given to the target image, including: Based on the mature aesthetic evaluation model, the target image is aesthetically evaluated. EEG data is collected when viewing the target image. The EEG data is preprocessed and feature calculated sequentially to obtain EEG feature data. The EEG feature data is input into the mature aesthetic evaluation model for calculation to obtain the aesthetic score of the target image, and the target image is given corresponding aesthetic category suggestions based on the aesthetic score.
7. A device for aesthetic evaluation of EEG signals across different subjects in a scene, used to implement the aesthetic evaluation and processing method for EEG signals across different subjects in any one of claims 1-6, characterized in that, include: The filtering module is used to select a number of aesthetic images with real rating labels from the preset aesthetic dataset as the image training set; The acquisition module is used to collect raw EEG data of the target object while viewing the aesthetic images in chronological order. The preprocessing module is used to preprocess and perform feature calculations on the raw EEG data to obtain EEG training data; The training module is used to train the initial aesthetic evaluation model based on the EEG training data and the image training set to obtain a mature aesthetic evaluation model. The evaluation module is used to evaluate the target image based on the mature aesthetic evaluation model. It processes the EEG data when viewing the target image and inputs it into the mature aesthetic evaluation model for calculation to obtain the aesthetic score of the target image. Based on the aesthetic score, it provides the target image with corresponding aesthetic category suggestions.
8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores one or more programs, which can be executed by one or more processors to implement the steps in the cross-subject scene electroencephalogram signal aesthetic evaluation processing method as described in any one of claims 1-6.
9. A terminal device, characterized in that, include: Processor, memory, and communication bus; the memory stores a computer-readable program that can be executed by the processor; The communication bus enables communication between the processor and the memory; When the processor executes the computer-readable program, it implements the steps in the cross-subject scene EEG signal aesthetic evaluation processing method as described in any one of claims 1-6.