A method for constructing a virtual scenario for mental health care using generative adversarial networks (GANs)

Abstract

The present invention relates to mental health care, and in particular to a method for constructing a mental health care virtual scene using a generative adversarial network (GAN). According to the analysis of mental health care needs, the type of virtual scene simulation and environmental parameters are determined, including the lighting, color and sound required for anxiety relief or depression symptom management; then an emotion perception module is developed to analyze the user's voice, expression and physiological signals through deep learning technology to identify the user's emotional state, and dynamically adjust the virtual environment through a feedback mechanism according to the emotional state; an adaptive GAN model is constructed, in which a generator network adaptively generates a scene that meets the user's emotional needs according to the input of the emotion perception module, while a discriminator network verifies the authenticity of the generated scene and evaluates the effectiveness of the scene in improving the user's emotions; a real-time interactive feedback system is implemented to allow the scene to change in real time when the user interacts with it, and the scene generation algorithm is optimized through machine learning to enhance the user experience and ensure personalization.

Description

Technical Field

[0001] The present invention relates to the field of mental health care technology, and in particular to a method for constructing a mental health care virtual scene using a generative adversarial network (GAN). Background Art

[0002] Although the current methods for constructing virtual scenarios for mental health care have demonstrated their potential in treatment practice, they still have many shortcomings and drawbacks. First, existing methods often lack sufficient flexibility and adaptability in personalized treatment. Virtual treatment environments are usually preset or designed based on user needs in general situations. This "one-size-fits-all" approach ignores the huge differences between users, such as individual psychological states, emotional response patterns, and their specific needs for treatment. This lack of personalization may lead to poor treatment effects because standardized virtual environments may not accurately match the treatment needs of specific users. For example, for users with high resistance or specific psychological histories, standardized virtual scenarios may be too stimulating or insufficient to induce sufficient treatment participation. Second, technical limitations are also an important drawback of existing methods. Although virtual reality (VR) technology provides an immersive experience in vision and hearing, it still has limitations in simulating real-world interactions. For example, tactile feedback is often missing, which limits the ability to simulate real-world situations, especially in treatment scenarios that require simulated physical touch. In addition, existing VR devices may cause users to feel uncomfortable during use due to their lack of ergonomic design, such as dizziness, nausea, or visual fatigue, all of which may affect the user's treatment experience and treatment outcomes. In addition, the issue of technology acceptance cannot be ignored. Although many patients may benefit from virtual treatment, for some user groups, especially the elderly or users with poor technical skills, complex technical operations may become a barrier to participation in treatment. These users may be confused or frustrated by the technical threshold, which may affect their treatment participation and treatment outcomes. Summary of the invention

[0003] The purpose of the present invention is to provide a method for constructing a virtual scene of mental health care using a generative adversarial network (GAN), thereby solving some of the drawbacks and shortcomings pointed out in the background technology.

[0004] The present invention solves the above-mentioned technical problems by adopting the following technical solutions: a method for constructing a mental health care virtual scene using a generative adversarial network (GAN), comprising: firstly, determining the type of virtual scene simulation and environmental parameters according to mental health care needs analysis, including lighting, color and sound required for anxiety relief or depression symptom management;

[0005] Then, we developed an emotion perception module that uses deep learning technology to analyze the user's voice, facial expressions, and physiological signals to identify the user's emotional state, and dynamically adjust the virtual environment based on the emotional state through a feedback mechanism;

[0006] Then, an adaptive GAN model is constructed, in which the generator network adaptively generates scenes that meet the user's emotional needs based on the input of the emotion perception module, while the discriminator network verifies the authenticity of the generated scenes and evaluates the effectiveness of the scenes in improving the user's emotions;

[0007] Finally, a real-time interactive feedback system is implemented, allowing the scene to change in real time when the user interacts with it, and the scene generation algorithm is optimized through machine learning to improve the user experience and ensure personalization.

[0008] Furthermore, the process of determining the type and environmental parameters of the virtual scene simulation includes:

[0009] S1. First, by integrating multimodal data integration technology, the advanced function F1 is used to calculate the user's comprehensive psychological state S. The formula is:

[0010] F1(S)=∫(α·log(1+e HR )+β·tanh(RT)+γ·σ(AP))dτ

[0011] Where HR is heart rate, RT is reaction time, AP is response mode, α, β, γ are adjustment parameters, and τ represents time variable;

[0012] S2. Secondly, an intelligent parameter recommendation system based on user data is used to apply the machine learning model formula F2 to determine the personalized scene environment parameter P, which is:

[0013]

[0014] where p n is the scene parameter, λ n is the attenuation factor, ω is the weight coefficient, S n Indicates the psychological state contribution corresponding to each environmental parameter;

[0015] S3, finally, the dynamic environment adaptation technology uses real-time data streams to dynamically adjust the virtual scene, applying the calculus formula F3, the formula is:

[0016]

[0017] Where V i is a virtual environment variable, ρ i ,μ is the adjustment coefficient, C i and H i Represent current and historical sentiment data respectively.

[0018] Furthermore, the emotion perception module is composed of the following parts:

[0019] P1. First, develop a multimodal emotion recognition system, using function F1 to integrate speech, expression and physiological signal data. The formula is defined as:

[0020]

[0021] Where X, Y, and Z represent the data inputs obtained from speech, expression, and physiological sensors, respectively, and a, b, and c are adjustment parameters;

[0022] P2, then implement the real-time feedback mechanism of emotional state, and dynamically adjust the virtual environment settings through the calculus formula F2, the formula is:

[0023]

[0024] Where E represents the emotional state index obtained after F1 function analysis, k and ω are adjustment coefficients;

[0025] P3, finally, the adaptive learning and personalized adjustment mechanism is adopted to apply the adaptive function F3, which is described as:

[0026]

[0027] Where V represents the user's feedback data on the virtual environment, and g and λ are the parameters of adaptive learning.

[0028] Furthermore, the method for constructing an adaptive GAN model includes:

[0029] First, we build a deep convolutional generative network (DCGAN) and integrate the emotion perception module to provide emotion feature vectors generated by multimodal data including speech, expression, and physiological signals. We adaptively adjust the generated virtual scene content, including color tone, texture, and light, based on the real-time emotion analysis results.

[0030] Then, an adaptive discriminator network is developed to not only verify the authenticity of the generated scenes, but also evaluate the actual impact of the scenes on users’ emotions. The emotion regulation effect of the scenes is evaluated by analyzing the users’ behavioral and physiological data in a specific virtual environment.

[0031] Finally, a dynamic feedback optimization mechanism is implemented, using reinforcement learning or online learning strategies to adjust the parameters of the generator and discriminator based on the user's real-time feedback on the scene, so as to optimize the GAN model.

[0032] Furthermore, the construction of a deep convolutional generative network DCGAN includes:

[0033] First, a multimodal data processing module is designed, and a deep neural network DNN is used to extract features from speech, facial expressions and physiological signals. Through high-level functions:

[0034]

[0035] Where X, Y, and Z represent the data obtained from voice, expression, and physiological sensors, respectively, and a, b, c, α, and β are adjustment parameters;

[0036] Then adjust the deep convolutional generation network DCGAN so that the generator receives the emotion feature vector as input and applies the calculus formula:

[0037]

[0038] Control the generated virtual scene content, including tone, texture and light, where V represents the emotion feature vector, λ i ,ω i Represents the weights and frequency coefficients of different features;

[0039] Finally, a real-time emotional feedback system is implemented to adjust the generated content according to the user's behavior and physiological reactions in the virtual environment, using an adaptive learning formula:

[0040]

[0041] Where θ is the parameter of the generator, and γ, κ, ν are adjustment coefficients.

[0042] Furthermore, the adaptive discriminator network development structure includes:

[0043] S1. First, develop an adaptive discriminator network to verify the authenticity of the generated scene, using advanced functions:

[0044]

[0045] Analyze the user's behavioral data x and physiological data y in the virtual environment, where α i and β j is the adjustment coefficient, which is used to evaluate the actual impact of the scene on the user's emotions;

[0046] S2. Then integrate user behavior responses and physiological parameters, including heart rate, galvanic skin response, facial expression and eye movement data, and apply data fusion algorithm:

[0047]

[0048] Where z represents the data collected from various sensor devices, γ k and δ k For the adjustment coefficient, provide emotional response data set;

[0049] S3. Finally, the real-time evaluation and feedback mechanism of emotion regulation effect is used, and the calculus formula is adopted:

[0050]

[0051] Where θ represents the parameters of the discriminator, κ, ξ, ω are adjustment coefficients, which enable the discriminator to adjust its evaluation criteria and the output of the generator based on the user's real-time data, thereby constructing emotional responsiveness generation and real-time adaptive adjustment of the virtual environment.

[0052] Furthermore, the implementation of the dynamic feedback optimization mechanism includes the following steps:

[0053] First, we integrate reinforcement learning into the GAN model by defining the reward function:

[0054]

[0055] Where G represents the scene output by the generator, x represents the user's behavior and physiological data, and θ G and θ D denote the parameters of the generator and the discriminator respectively, and Ω denotes the space of all emotional states; the reward function guides the training process of the generator and the discriminator based on the user's satisfaction with the generated scene, so that the GAN model obtains a higher reward when generating content that is liked by the user;

[0056] Then, a real-time data feedback loop is implemented, and the online learning formula is applied to update the model parameters in real time and adjust the weights of the generator and discriminator, where the integration is performed within the user interaction time window T:

[0057]

[0058] where η is the learning rate, γ is the time decay factor, R(t) is the actual reward, and is the prediction reward.

[0059] Furthermore, the real-time interactive feedback system comprises the following three steps:

[0060] S1. First, we realize real-time interactive data capture, using advanced functions:

[0061]

[0062] where S = {s1(t),s2(t),…,s n (t)} represents a series of user behavior and physiological data collected from sensors, α i It is an adjustment parameter used to evaluate the user's emotional state in real time and adjust the virtual scene;

[0063] S2. Then develop a dynamic scene adjustment algorithm and apply the calculus formula:

[0064]

[0065] where X = {x1(t), x2(t), …, x m (t)} represents the dynamically adjusted scene parameters, including lighting and sound, γ j and β j is the adjustment factor;

[0066] S3. Finally, deploy the machine learning optimization strategy using the function:

[0067]

[0068] Where θ represents the parameters of the scene generation algorithm, and φ(t) represents the weight function obtained from user feedback, which is used to optimize the generation algorithm in real time and improve user experience.

[0069] Beneficial effects of the present invention:

[0070] 1. Personalized treatment improvement: By monitoring the user's behavior and physiological responses in real time, GAN is able to dynamically adjust the virtual environment to ensure that each user receives personalized treatment for their specific emotional and psychological state. This personalized approach can improve the relevance and effectiveness of treatment because it more precisely matches the user's specific needs.

[0071] 2. Real-time feedback and adaptive adjustment: The integration of reinforcement learning and online learning strategies enables the system to adjust its generated content in real time based on the user's immediate feedback. This not only improves the user's experience, but also makes the treatment process more flexible and can respond to the fluctuations and changes of the user's emotions in real time.

[0072] 3. Enhanced emotion recognition and response: Through advanced functions and machine learning algorithms, the system can accurately identify the user's emotional state and adjust the virtual environment accordingly. This accurate emotion recognition and response mechanism helps provide more effective emotion regulation strategies, which are more effective in treating emotional disorders such as anxiety and depression.

[0073] 4. Improve user engagement and satisfaction: Dynamic and responsive virtual environments can better engage users in the treatment process and increase their motivation. Increased engagement is often directly related to higher treatment compliance and better treatment outcomes. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 This is a flow chart of the method for constructing a virtual scene of mental health care using a generative adversarial network (GAN) according to the present invention.

[0075] Figure 2Flowchart of the method for constructing an adaptive GAN model in the present invention. DETAILED DESCRIPTION

[0076] The specific implementation modes of the present invention will be described in detail below in conjunction with the accompanying drawings.

[0077] The method of constructing a virtual scene for mental health care using a generative adversarial network (GAN) first conducts a mental health care needs analysis to determine the specific type of virtual scene simulation and its related environmental parameters, and to identify and respond to specific treatment needs, such as anxiety relief or depressive symptom management, which includes selecting appropriate lighting intensity, color schemes, and sound settings; for example, to relieve anxiety, the scene includes soft lighting, soothing blue-green tones, and natural sounds, such as the sound of flowing water or gentle wind, while brighter lighting and more motivational background music can be used for depressive symptom management; the selection of these parameters is based on research on the impact of emotions in psychological theory and clinical practice, and aims to induce emotional changes in users by adjusting the virtual environment, thereby achieving a therapeutic effect.

[0078] Then, an emotion perception module is developed, which uses deep learning technology to analyze the user's voice, facial expressions and physiological signals to identify the user's emotional state; it involves extracting emotion-related features from the user's voice tonality, speaking speed, facial expression changes and physiological parameters such as heart rate and skin electrical response, and processing them using a trained deep neural network model to identify the user's emotional state, such as happiness, sadness, anger or anxiety; the identified emotional state is then used to dynamically adjust the virtual environment through a feedback mechanism. For example, if the user is detected to be anxious, the system adjusts the environmental settings, such as changing the color temperature, adjusting the background music and sound intensity, or introducing visual relaxation elements (such as natural landscapes) to help the user relax and improve his mental state.

[0079] Then, an adaptive GAN model is constructed, including a generator network and a discriminator network. The generator network is designed to adaptively generate scenes that meet the user's emotional needs based on the input of the emotion perception module. For example, if the emotion perception module identifies the user as anxious, the generator generates a quiet forest scene or other environment that helps to relax. The generator receives specific emotional indicators from the emotion perception module, such as emotional state and intensity, and uses these data to adjust the visual and auditory elements of the scene to influence the user's emotional state in the best way. At the same time, the task of the discriminator network is to verify the authenticity of the generated scene and ensure the naturalness and realism of the scene. In addition, the discriminator also evaluates the effectiveness of the scene in improving the user's emotions, and judges the therapeutic effect of the scene by analyzing the emotional change data after the user interacts with the scene. In this way, the discriminator not only ensures the quality of the scene, but also helps optimize the performance of the generator through a continuous feedback loop to ensure that each generated scene can effectively respond to the user's emotional needs.

[0080] Finally, a real-time interactive feedback system is implemented, so that the virtual scene can change in real time according to the user's feedback when the user interacts with it, thereby providing a more personalized experience; machine learning technology is used to continuously optimize the scene generation algorithm, and the parameters of the generator are adjusted according to the user's behavior and feedback data to ensure that the scene content can respond to the user's emotions and changes in needs in real time. For example, if the user shows a negative reaction to a scene element, the system will learn this feedback and adjust the element or the entire scene to reduce the user's discomfort; similarly, if a certain scene setting can significantly improve the user's comfort or therapeutic effect, the system will also strengthen this type of scene element.

[0081] Embodiment 1:

[0082] In this embodiment, multimodal data integration technology is integrated to calculate the user's comprehensive psychological state S. User A participates in a virtual therapy session aimed at managing his anxiety symptoms. In this session, the following parameters are continuously monitored and recorded: heart rate (HR), reaction time (RT), and response mode (AP). The calculation formula used is:

[0083] F1(S)=∫(α·log(1+e HR )+β·tanh(RT)+γ·σ(AP))dτ

[0084] Among them, the coefficients α, β, and γ are adjustment parameters used to adjust the contribution of different physiological and behavioral data to the assessment of psychological state. In a typical treatment session, the average heart rate (HR) is set to 85bpm, the reaction time (RT) is set to 250 milliseconds, and the standardized score of the response mode (AP) is set to 0.7. The adjustment parameters α = 0.5, β = 0.3, and γ = 0.2 are set. These values ​​are usually based on empirical data or preliminary clinical studies to optimize the sensitivity and specificity of the model.

[0085] Substituting these parameters and actual data into the formula, the psychological state S of a specific time window (such as one hour) can be calculated. The specific calculation steps are as follows:

[0086] 1. Calculate the contribution of heart rate: α·log(1+e 85 )≈0.5·4.45=2.225

[0087] 2. Calculate the contribution of reaction time: β·tanh(0.25)≈0.3·0.244=0.0732

[0088] 3. Calculate the contribution of the response mode: γ·σ(0.7)≈0.2·0.668=0.1336

[0089] Adding these values ​​together yields an estimate of S = S≈2.225+0.0732+0.1336=2.432, and this composite score is then used to dynamically adjust the virtual therapy environment for user A. For example, if S exceeds a certain threshold (e.g., 2.5), the system adjusts the environment settings, adjusts the colors in the virtual scene to a more soothing blue-green hue, reduces stimulating elements in the scene, or adds soft background music to help reduce user A's anxiety level.

[0090] Secondly, an intelligent parameter recommendation system based on user data is used to optimize his virtual treatment environment. It is assumed that the comprehensive psychological state S of user A has been calculated through the previous steps, and then a personalized virtual environment is customized based on this state. The machine learning model formula F2 used is as follows:

[0091]

[0092] There are three main environmental parameters p1, p2, and p3, which represent the scene's light intensity, color tone, and background music volume. n Both have a decay factor λ n These attenuation factors are used to adjust the contribution of different parameters to the psychological state S n The specific numerical settings are as follows:

[0093] λ1=0.5(light intensity)

[0094] λ2=0.3(hue)

[0095] λ3=0.2(background music)

[0096] ω is the weight coefficient, which is set to 0.8. This is based on the empirical value range in the optimization process. Usually, this coefficient is between 0.5 and 1.0.

[0097] Considering that the value of user A's comprehensive psychological state S calculated before is about 2.432, the optimal value of each parameter can be calculated by F2. Substituting the specific S value into the F2 formula, we get:

[0098]

[0099] Furthermore, the partial derivatives of each parameter can be calculated to determine the most appropriate adjustment. Through continuous learning and feedback from the machine learning model, these parameters can be dynamically adjusted:

[0100] If the partial derivatives indicate that increasing the light intensity will reduce S, the system will appropriately enhance the lighting of the virtual environment.

[0101] If changing the hue is more appropriate to maintain or improve the psychological state, the system will adjust the hue.

[0102] If the background music adjustment has the greatest impact on S, the volume will be adjusted.

[0103] By monitoring User A's reactions in real time and recalculating these parameters, the system is able to ensure that the virtual environment is always adapted to his current psychological needs, thereby maximizing the therapeutic effect.

[0104] Finally, the dynamic environment adaptation technology is implemented, using real-time data streams to dynamically adjust the virtual scene to respond to the immediate psychological needs of user A. The core of the application is the calculus formula F3, which is in the following form:

[0105]

[0106] In this formula, V i Represents various variables of the virtual environment, such as light intensity, sound level, and scene color; ρ i and μ are adjustment coefficients, which control the response intensity of environmental variables to current and historical emotional data respectively; C i and H i Represents current and historical sentiment data respectively. There are three main environment variables to adjust, corresponding to three different sets of coefficients. You can set:

[0107] For light intensity ρ1=0.8,μ1=0.5

[0108] For sound level ρ2 = 0.6, μ2 = 0.3

[0109] For scene color ρ3=0.4,μ3=0.2

[0110] The coefficient value range is usually determined based on preliminary clinical research and user feedback testing, and is generally between 0.2 and 1.0.

[0111] Substitute these parameters into the F3 formula to perform an actual calculation example. Assume that at a specific evaluation moment, the current emotional data C shows that user A feels a lot of pressure, and the data obtained by the system from the physiological monitoring device shows that the heart rate increases (indicating is a positive value), while historical data H (such as the average heart rate in the past hour) shows that it has been at a high level. Based on these real-time data, the system calculates the environmental parameters that need to be adjusted, such as reducing the light intensity or lowering the volume of background music.

[0112] For example, if the system calculates that the light intensity needs to be adjusted, it does so by reducing V1 while dynamically adjusting based on the result of F3, as calculated below:

[0113]

[0114] if If is negative (indicating a need to reduce light intensity), the expression will further strengthen the reduction in light intensity, thereby helping to reduce the stress level of user A. In this way, the system continuously uses real-time data to optimize the virtual environment, ensuring that it can effectively adapt to and alleviate user A's psychological state and achieve personalized treatment effects.

[0115] Embodiment 2:

[0116] In this embodiment, a multimodal emotion recognition system is first developed to more accurately identify and respond to the user's emotional state by integrating voice, expression and physiological signal data from different sensors and using advanced deep learning technology. To this end, a composite function F1 is used to integrate these data, and the calculation results of the function are used to guide the subsequent virtual scene generation.

[0117] Specifically, the function F1 is defined as:

[0118]

[0119] Among them, X, Y, and Z represent the data obtained from voice, expression, and physiological sensors, respectively. For example, X can be the numerical representation of the pitch and intensity of the voice, Y can be the expression intensity value analyzed by the facial expression recognition system, and Z can be the measurement value of heart rate or other physiological parameters.

[0120] The settings are based on previous experimental data and related research, and the range of values ​​of the adjustment parameters is selected as follows:

[0121] a: The weight that affects the voice data, ranging from 0.1 to 0.5.

[0122] b: The weight that affects the expression data, the range is set from 0.05 to 0.3.

[0123] c: The weight that affects the physiological signal data, which ranges from 0.2 to 0.6.

[0124] Assuming that at a specific evaluation time, the following data input is obtained: X = 0.8, Y = 0.6, Z = 85 (heart rate). Selecting parameters a = 0.3, b = 0.2, c = 0.4, F1 can be calculated:

[0125]

[0126] Calculate 0.3 0.8 2 We get 0.192. We get about 0.155, and 0.4·ln(86) gets about 3.384. Setting dt to a simplified unit of time (e.g. 1 second), the overall expression is approximately:

[0127] F1=exp(-(0.192+0.155+3.384))=exp(-3.731)≈0.024

[0128] This calculation result F1 reflects the intensity of the user's emotional state at that moment. A lower F1 value indicates that the user experiences higher emotional stress. This result will guide the generator of the virtual scene to adjust the scene elements to better suit the user's emotional needs, such as reducing stress elements in the environment and adding relaxing elements such as soft background music or quiet natural scenery, thereby achieving the purpose of mental health care.

[0129] Next, we will implement a real-time feedback mechanism for the emotional state, using the calculus formula F2 to dynamically adjust the virtual environment. Formula F2 is defined as follows:

[0130]

[0131] Among them, E represents the emotional state index obtained by analyzing the function F1 discussed above, and k and ω are adjustment coefficients, which control the intensity of the influence of the emotional index on the scene adjustment and the frequency of periodic changes, respectively.

[0132] Based on the previous simulation and experimental data, it can be set:

[0133] The value of k ranges from 0.1 to 1.0, which allows its sensitivity to environmental adjustments to be adjusted according to different treatment needs.

[0134] The value range of ω is between 0.5π and 2π, which allows the environment to respond to periodic changes from slow to fast.

[0135] At a certain moment, after calculation by F1, the value of E is 0.024, which is the result of the previous calculation. Set k = 0.5 and ω = π (1 cycle is a change frequency of 2 seconds). According to formula F2, the dynamic adjustment amount of the emotional state at a certain moment t (for example, t = 1 second) can be calculated:

[0136]

[0137] Since sin(π·1)=0, no matter Regardless of the value of (0.5·0.024), the result of F2 will be 0. This means that at t=1 second, the environment does not need to be adjusted. This calculation can help the system identify when to adjust and when to keep the current state, thereby avoiding unnecessary or excessive adjustments and ensuring a smooth transition of the environment and a comfortable experience for users.

[0138] In a real-time system, this calculation will be performed continuously, and each change in t will result in a different sin(ωt) value, which in turn affects the dynamic adjustment of the environment.

[0139] Finally, the adaptive learning and personalized adjustment mechanism is adopted, which is implemented through the adaptive function F3 to further optimize the virtual environment to provide a more refined treatment response. The definition of function F3 is:

[0140]

[0141] Where V represents the user's feedback data on the virtual environment, g and λ are adjustment parameters used to control the adaptability of learning and the sensitivity of feedback. For practical applications, the value range of parameters g and λ is set as follows:

[0142] g: Controls the magnitude of the feedback influence, ranging from 0.1 to 1.0.

[0143] λ: Adjusts the attenuation speed of feedback, with a value range of 0.01 to 0.1.

[0144] In the treatment scenario of user A, the user behavior monitoring system collects his behavior data in a specific virtual environment. These data may include his movement speed, gaze point, interaction frequency, etc., so as to calculate the comprehensive feedback value V. Set V = 0.75, select g = 0.5 and λ = 0.05 for example calculation:

[0145] First, calculate gV=0.5×0.75=0.375.

[0146] Next, calculate Since gV is a linear function of V, its second-order derivative is 0.

[0147] Finally, substituting these results into the adaptive function, we can calculate:

[0148] F3(0.75)=0·e -0.05×0.75 =0

[0149] Since the second-order derivative is 0, it indicates that under the currently selected parameters, changes in V will not cause changes in environmental adjustments. This result indicates that based on the current user data and parameter settings, the virtual environment is sufficiently adapted to the user's needs, or the values ​​of g and λ need to be adjusted to enhance the sensitivity of the model.

[0150] In addition, by adjusting the values ​​of g and λ, different adaptive learning strategies can be explored to achieve a more personalized treatment environment, thereby continuously improving treatment outcomes and user satisfaction.

[0151] Embodiment 3:

[0152] In this embodiment, a deep convolutional generative network (DCGAN) is constructed to integrate multimodal data, generate emotional feature vectors, and adaptively adjust the tone, texture and light of the virtual scene accordingly. The core of this method lies in the application of the deep learning framework, especially the design and implementation of the multimodal data processing module.

[0153] First, the deep neural network (DNN) extracts features from speech, facial expressions, and physiological signals. To explain the feasibility of this step in detail, the function F1 is used, and the specific formula is:

[0154]

[0155] In this formula:

[0156] X(t), Y(t) and Z(t) represent the data acquired from speech, expression and physiological sensors at time t, respectively.

[0157] a, b, c, α, and β are adjustment parameters used to adjust the weight and influence of each data source in the comprehensive emotion feature vector.

[0158] To ensure the continuity and depth of the example, set:

[0159] a=0.3,α=0.1 adjust the influence of speech data;

[0160] b = 0.5, β = 2π to adjust the periodic influence of expression data;

[0161] c=0.4 is used to adjust the influence of physiological signal data, especially heart rate or skin conductance.

[0162] Set within a specific evaluation time T = 1 second, the data obtained are as follows:

[0163] The voice data X(t) has an amplitude value of 0.8;

[0164] The expression data Y(t) has an emotion recognition score of 0.6;

[0165] The physiological signal Z(t) is a heart rate value of 70.

[0166] Substituting into the formula and calculating F1, we get:

[0167]

[0168] To simplify the calculation, set cos(2π·0.6) to approximately 1. is approximately equal to 25, and the above formula is roughly:

[0169]

[0170] This value F1=10.74 reflects the user's comprehensive emotional state at that point in time. Based on this result, the DCGAN generator will adaptively adjust the tone, texture, and light of the virtual scene to respond to the user's emotional changes in the best way, thereby providing a personalized treatment environment.

[0171] The deep convolutional generative network (DCGAN) is then adjusted so that its generator can adaptively adjust the generated virtual scene content, including tone, texture, and light, according to the emotional feature vector V derived by the multimodal data processing module. To this end, the calculus formula F2 is introduced to calculate and apply the adjustment of the emotional feature vector to the virtual scene content:

[0172]

[0173] Among them, V represents the emotional feature vector, λ i and ω i are the weights and frequency coefficients of different features. Set the value range of these parameters to optimize the scene generation process:

[0174] λ i Controls the influence of each emotion feature, ranging from 0.1 to 1.0.

[0175] ω i Controls the temporal variation frequency of the emotional feature, with the value range set to 0.5π to 2π.

[0176] Consider three main emotional features: heart rate change V1, facial expression intensity V2, and voice emotion intensity V3. These emotional features are evaluated once in the evaluation cycle, and the resulting emotional feature vector is V = {0.8, 0.6, 0.9}.

[0177] Setting parameters:

[0178] λ1=0.5,ω1=π

[0179] λ2=0.7,ω2=1.5π

[0180] λ3=0.3,ω3=2π

[0181] Substituting these parameters into the F2 formula, we can calculate the adjustment impact corresponding to each emotional feature:

[0182]

[0183] Once simplified, these integrals can be calculated using numerical integration methods and the results applied to dynamically adjust the scene's hue, texture, and lighting. For example, if the heart rate-induced adjustment values ​​indicate a need to enhance the soothing elements of the scene, the scene generator increases the weight of blue-green hues, reduces the complexity of the texture, or adjusts the lighting to create a more relaxing environment.

[0184] Finally, a real-time emotional feedback system is implemented to adjust the generated content according to the user's behavior and physiological response in the virtual environment. The specific expression using the adaptive learning formula F3 is:

[0185]

[0186] Among them, θ represents the parameters in the generator responsible for adjusting the virtual scene elements, while γ, κ, and ν are adjustment coefficients that control the learning rate, decay factor, and vibration frequency, respectively.

[0187] Set the following parameter ranges:

[0188] γ controls the overall learning rate and is set to a value between 0.05 and 0.2. This range allows the system to respond to changes in user emotions at a moderate speed without over-adjusting and causing instability in the scene.

[0189] κ, as a decay coefficient, affects the speed of learning and its sensitivity to long-term and short-term emotions, and its value range is set to 0.01 to 0.1. A smaller κ value can keep the memory for a longer time, while a larger value makes the system quickly forget the old emotional data.

[0190] ν defines the vibration frequency of parameter adjustment, and the value range is set to π to 3π. The system can adjust the scene settings at a certain rhythm to adapt to the reaction cycle of different users.

[0191] Assume that at a specific time t, based on user feedback, the generator parameter θ needs to be adjusted. Set γ = 0.1, κ = 0.05, ν = 2π, and set the initial value of θ(t) to 0.8. Apply the F3 formula to calculate the rate of parameter adjustment:

[0192]

[0193] When t = 1 second, substitute the calculation:

[0194]

[0195] This means that at t = 1 second, there is no need to adjust θ, because the periodic nature of sin(2πt) causes the adjustment to be zero. This dynamic adjustment mechanism based on a time-varying function allows the system to make necessary adjustments at the appropriate time and remain stable at other times, thereby ensuring the coherence of the virtual environment and the consistency of the user experience.

[0196] Embodiment 4:

[0197] In this embodiment, an adaptive discriminator network is first developed to verify the authenticity of the generated scenarios and evaluate the actual impact of these scenarios on the user's emotions. To this end, a function F1 is introduced, which combines the user's behavioral data x and physiological data y in the virtual environment to quantify and evaluate the impact of the virtual environment on the user's emotional state.

[0198] Specifically, the higher-level function F1 used by the adaptive discriminator network is defined as:

[0199]

[0200] In this formula:

[0201] x i Represents data obtained from user behavior monitoring, such as movement speed, interaction frequency, etc.

[0202] y j Represents data obtained from user physiological monitoring, such as heart rate, skin galvanic response, etc.

[0203] α i and β j is the adjustment coefficient, which is used to adjust the weight of behavioral data and physiological data in the assessment of emotional state.

[0204] Set 1 and β j The value range of is:

[0205] α i The value range of is 0.1 to 1.0, which allows the discriminator to flexibly adjust its sensitivity to behavioral data.

[0206] β j The value range of is 0.05 to 0.5, which allows the discriminator to maintain some flexibility in responding to physiological data.

[0207] Assume that in a certain test session, user x's data includes a moving speed of 0.8 m / s and an interaction frequency of 20 times / minute, while y's data includes a heart rate of 85 bpm. Select α1 = 0.5, β1 = 0.3, and you can calculate F1:

[0208] F1(0.8,85)=∫exp(-0.5·0.8 2 )·cos(0.3·85)dt

[0209] Here, exp(-0.5 0.8 2 ) is used to calculate the psychological impact factor, while cos(0.3·85) is used to calculate the physiological impact factor. If this integration is set to be performed in a short period of time (such as within 1 minute), the numerical integration method can be used to obtain the specific evaluation value of the emotion regulation effect.

[0210] Then the user behavioral responses and physiological parameters are integrated and the data fusion algorithm F2 is applied to provide a comprehensive emotional response dataset.

[0211] The expression of data fusion algorithm F2 is defined as:

[0212]

[0213] in:

[0214] z represents data collected from a variety of sensor devices, including heart rate, galvanic skin response, facial expression, and eye movement data.

[0215] γ k and δ k is the adjustment coefficient, which is used to adjust the influence weight and nonlinear response of each data in the final emotion evaluation.

[0216] Set the value range of the adjustment coefficient:

[0217] γ k Controls the contribution of each data type to the final sentiment evaluation, with the value range set to 0.1 to 0.5, which can ensure the balance of different data sources in sentiment evaluation.

[0218] δ k Used to adjust the nonlinear response of the data. The value range is set to 1 to 3, so that the system can show a more dynamic response under different emotional states.

[0219] In a session of user A in the virtual environment, the sensor data is as follows:

[0220] Heart rate z1 = 85 bpm,

[0221] Skin galvanic response z2 = 0.4 microsiemens,

[0222] Facial expression intensity z3 = 0.6,

[0223] Eye tracking speed z4 = 15 movements per second.

[0224] The coefficients are selected as γ1=0.3, δ1=2; γ2=0.2, δ2=3; γ3=0.4, δ3=1; γ4=0.1, δ4=2, which can be substituted into F2 to calculate the emotional response data set:

[0225] F2(z)=0.3·(log(1+85)) 2 +0.2·(log(1+0.4)) 3 +0.4·(log(1+0.6)) 1 +0.1

[0226] (log(1+15)) 2

[0227] Simplifying the calculation, we can roughly get:

[0228] log(86)≈4.45

[0229] log(1.4)≈0.34

[0230] log(1.6)≈0.47

[0231] log(16)≈2.77

[0232] Bring it into calculation:

[0233] F2(z)≈0.3·4.45 2 +0.2 0.34 3 +0.4 0.47+0.1 2.77 2 ≈6.01+0.007+0.188+0.77

[0234] =6.975

[0235] This calculation result provides a comprehensive score that reflects the overall emotional state of user A in the virtual environment, which will guide subsequent adjustments to the virtual environment to better suit his emotional needs.

[0236] Finally, using the real-time evaluation and feedback mechanism of emotion regulation effect, the calculus formula F3 is used to dynamically adjust the parameters of the discriminator and generator to ensure that the virtual environment can reflect the user's emotional state in real time and make necessary adjustments. Formula F3 is defined as follows:

[0237]

[0238] in:

[0239] θ(t) represents the parameters of the discriminator, which directly affects the generation of the virtual environment.

[0240] κ, ξ, and ω are the regulation coefficients that control the sensitivity and responsiveness of the emotion regulation mechanism.

[0241] Set the value range of the adjustment coefficient:

[0242] κ controls the decay rate and is set in the range of 0.01 to 0.1, which allows the system to maintain a certain sensitivity in responding to user emotions.

[0243] ξ affects the amplitude of the periodic adjustment and is set in the range of 0.5 to 1.5 so that the adjustment amplitude is neither too large nor too small.

[0244] ω controls the vibration frequency and its value range is set to π to 3π so that periodic adjustment can be made within a certain frequency.

[0245] Assume that a user is interacting in a virtual environment, and the discriminator parameters θ(0) are initially set to 1.0, κ = 0.05, ξ = 1.0, and ω = 2π. Assume that the adjustment at t = 1 second can apply F3:

[0246]

[0247] Since sin(2π·1)=0, we need to calculate the derivative of the first part. Assuming that the change of θ(t) is small, it can be approximated as linear in a short time, so The value of depends mainly on the rate of change of θ(t). If θ(t) increases to 1.01:

[0248]

[0249] Through calculation, it can be found that this slight change leads to a non-zero but small adjustment value of F3. This shows that the system can fine-tune the discriminator parameters to reflect the real-time changes of the user's emotions, thereby dynamically adjusting the virtual environment.

[0250] Embodiment 5:

[0251] In this embodiment, a dynamic feedback optimization mechanism is implemented to optimize the construction of virtual scenarios for mental health care, and the parameters of the generator and the discriminator are adjusted according to the real-time feedback of the users to improve the personalization and satisfaction of the scenarios.

[0252] Ensemble of reinforcement learning strategies

[0253] First, reinforcement learning is integrated into the GAN model to dynamically adjust the parameters of the generator and discriminator. This process is achieved by defining a reward function R, which takes into account the user's satisfaction with the generated scene and guides the model training accordingly. The reward function R is defined as follows:

[0254]

[0255] This function consists of two parts:

[0256] 1. Item 1 Measures the difference between the generator output G and the user's expectation x (consisting of the user's behavior and physiological data). This term encourages the generator output to be as close to the user's expectation as possible.

[0257] 2. The second term sin(∫ Ω θ D xdω) evaluates the discriminator parameters θ DThe correspondence with the user data x promotes the discriminator to effectively reflect the user's emotions while verifying the authenticity of the scene.

[0258] Parameter settings

[0259] Tuning parameter: θ G and θ D They represent the parameters of the generator and discriminator, respectively, which are dynamically adjusted through the training process to maximize the reward function R.

[0260] Data representation: x represents the behavioral and physiological data collected from the user, such as heart rate, facial expressions, etc., which are used to calculate the satisfaction indicator in the reward function.

[0261] Assume that the user data x collected at a certain time point includes a heart rate of 80 bpm and a smile intensity of 0.6. Assume that the initial parameters θ G = 1.0 and θ D = 0.5. Calculating the reward function involves the following steps:

[0262] 1. Calculation The value of , set in a given emotional state space Ω, this difference is 0.8 (this is a simplified numerical example).

[0263] 2. Substitute into the reward function:

[0264] Parameter value range

[0265] θ G and θ D The adjustment depends on the output of the reward function. Their range of variation is dynamically adjusted according to the feedback data during training, generally in the range of [0,2].

[0266] In this way, the generator and discriminator can learn how to better respond to the user's emotional state and preferences, gradually improving the realism and satisfaction of the generated scenes.

[0267] Then, a real-time data feedback loop is implemented to update the model parameters in real time using online learning formulas. The weights of the generator and discriminator are dynamically adjusted during the process to ensure that the virtual scene can continuously adapt to the user's emotional and behavioral changes.

[0268] The online learning formula is designed to update the parameters of the GAN model according to real-time data within the user interaction time window T. The formula is expressed as:

[0269]

[0270] in:

[0271] η is the learning rate, which determines the step size of parameter update.

[0272] γ is the time decay factor that controls how quickly the influence of old data on the learning process decreases.

[0273] R(t) is the actual reward obtained from user interaction.

[0274] is the reward predicted by the model.

[0275] Parameter value range

[0276] The learning rate η is usually set between 0.001 and 0.1 to ensure that the learning process is both stable and sensitive enough.

[0277] The time decay factor γ is usually in the range of 0.01 to 0.1 to appropriately extend the impact of historical data on current learning.

[0278] The user’s actual feedback reward R(t) and predicted reward The following was recorded during the 10-second interaction window:

[0279] At t = 0 seconds, R(t) = 0.8,

[0280] At t = 5 seconds, R(t) = 0.9,

[0281] At t = 10 seconds, R(t) = 0.95,

[0282] Choosing η = 0.05 and γ = 0.05, we can estimate Δθ using the numerical integration method:

[0283] calculate Values ​​at t=0, 5, 10 seconds: 0.01, 0.0025, 0.0025

[0284] Apply the time decay factor and integral calculation:

[0285] In this way, it is possible to calculate how the parameters θ should be adjusted throughout the time window T to reduce the difference between the actual reward and the predicted reward and optimize the generated virtual environment.

[0286] Embodiment 6:

[0287] In this embodiment, a real-time interactive feedback system is first implemented to achieve real-time interactive data capture, in which a series of user behavior and physiological data collected from sensors are processed using a high-level function F1. These data are used to evaluate the user's emotional state in real time and adjust the virtual scene accordingly to improve the treatment effect and user experience.

[0288] The higher function F1 is defined as:

[0289]

[0290] In this formula:

[0291] S={s1(t),s2(t),…,s n (t)} represents a series of user behavior and physiological data collected from sensors, such as heart rate, galvanic skin response, facial expression, eye tracking, etc.

[0292] α i is a tuning parameter used to regulate the contribution of different types of data to the assessment of emotional states.

[0293] To ensure the scientificity and effectiveness of the implementation plan, the adjustment parameter α is set i The value range of is:

[0294] α i The value range of is set to 0.1 to 1.0, allowing different physiological and behavioral data to make moderate contributions in the assessment of emotional state.

[0295] The three key indicators of the user monitoring are set: heart rate s1(t), skin galvanic response s2(t), and eye movement speed s3(t). The monitoring period T is set to 10 seconds, and the parameters α1 = 0.5, α2 = 0.3, and α3 = 0.2.

[0296] The acceleration (second derivative) of these data during the period t=0 sec to t=10 sec is as follows (simplified estimates):

[0297]

[0298] Apply the formula for calculation:

[0299]

[0300] This calculation result will be used to evaluate the user's emotional state. If the value shows that the user's emotional state is relatively stable or improved, the virtual scene is adjusted appropriately; if not, the system needs to continue to adjust to achieve better treatment effects.

[0301] Then, a dynamic scene adjustment algorithm was developed, using the calculus formula F2 to achieve dynamic adjustment of scene parameters such as lighting and sound in the virtual environment to adapt to the user's real-time emotional changes and feedback.

[0302] The calculus formula F2 used by the dynamic scene adjustment algorithm is defined as follows:

[0303]

[0304] In this formula:

[0305] X={x1(t),x2(t),…,x m (t)} represents the dynamically adjusted scene parameters.

[0306] x j (t) is a specific scene parameter, such as light intensity or sound level.

[0307] γ j and β j It is the adjustment coefficient that controls the adjustment sensitivity and response characteristics of parameters in different scenarios.

[0308] Set the value range of the adjustment coefficient:

[0309] γ j The value range of is set to 0.1 to 1.0, so that the adjustment of parameters in different scenes is flexible enough.

[0310] β j The value range of is set to 0.01 to 0.1, which controls the curve shape of the response and ensures a moderate nonlinear response.

[0311] In a specific treatment scenario, two key scenario parameters are monitored and adjusted: light x1(t) and background music volume x2(t). The monitoring period t is set from 0 to 10 seconds, and the parameters γ1 = 0.5, β1 = 0.05; γ2 = 0.3, β2 = 0.03.

[0312] At a certain point in time, the light intensity x1(t) = 200 lux, and the background music volume x2(t) = 50 dB. Applying F2 to calculate the derivatives of light and sound adjustments is as follows:

[0313]

[0314] By simplified numerical methods, the values ​​of these derivatives can be estimated. In the big x j When the value of (t) approaches saturation, the derivative will be very small, indicating that the adjustment change is small under these parameter settings, reflecting that the system is relatively stable for the current settings.

[0315] Finally, a machine learning optimization strategy is deployed to optimize the mental health care virtual scene constructed by the generative adversarial network (GAN) using function F3. This strategy aims to respond to and optimize the user experience in real time by dynamically adjusting the parameters of the scene generation algorithm. This approach involves adjusting the generation parameters using a weight function derived from user feedback, ensuring that the virtual scene can continuously adapt and respond to the user's emotional and behavioral feedback.

[0316] The specific definition of the optimization function F3 is:

[0317]

[0318] In this formula:

[0319] θ represents the parameters of the scene generation algorithm, which directly affect the lighting, color tone, texture and other aspects of the virtual environment.

[0320] φ(t) is a weight function obtained from user feedback and adjusted according to the user's response at different time points.

[0321] P(θ|t) represents the probability distribution of the generated scene corresponding to the parameter θ at a given time t.

[0322] Set the correlation between the weight function φ(t) and the scene parameter θ:

[0323] The weight function φ(t) can be set according to factors such as the frequency of user interaction and the magnitude of emotional changes. For example, the weight can be increased when the user's emotions fluctuate greatly to promote the system's response to these changes.

[0324] It is assumed that within a period of time T = 10 seconds, the user's response to specific scene elements (such as specific light intensity and music volume) is observed and the relevant feedback data is recorded. Based on this data, φ(t) is set as follows:

[0325] In the time period when users show high satisfaction, the value of φ(t) is high, such as φ(t) = 0.8;

[0326] During the period of low satisfaction, the value of φ(t) is low, such as φ(t) = 0.2.

[0327] Set the initial value and variation range of the scene parameter θ as follows:

[0328] Initially θ=1.0, and it is adjusted to θ=1.05 or higher according to feedback needs.

[0329] The optimal θ update is determined by calculating F3:

[0330]

[0331] The system dynamically adjusts the generated parameters θ to ensure real-time adaptability and personalized adjustment of the virtual environment. Through such calculations, the system can carefully reflect the changes in user needs, adjust the generated virtual scenes in a timely manner, and improve user experience satisfaction and treatment effects.

[0332] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A method for constructing a virtual scene of mental health care using a generative adversarial network (GAN), characterized in that The following steps are involved: First, based on the mental health care needs analysis, the type of virtual scene simulation and environmental parameters, including the lighting, color, and sound required for anxiety relief or depression symptom management, were determined; Then, we developed an emotion perception module that uses deep learning technology to analyze the user's voice, facial expressions, and physiological signals to identify the user's emotional state, and dynamically adjust the virtual environment based on the emotional state through a feedback mechanism; Then, an adaptive GAN model is constructed, in which the generator network adaptively generates scenes that meet the user's emotional needs based on the input of the emotion perception module, while the discriminator network verifies the authenticity of the generated scenes and evaluates the effectiveness of the scenes in improving the user's emotions; Finally, a real-time interactive feedback system is implemented, allowing the scene to change in real time as the user interacts with it, and the scene generation algorithm is optimized through machine learning to improve the user experience and ensure personalization; The process of determining the type and environmental parameters of the virtual scene simulation includes: S1. First, by integrating multimodal data integration technology, the advanced function F1 is used to calculate the user's comprehensive psychological state S. The formula is: F1(S)=∫(α·log(1+e HR )+β·tanh(RT)+γ·σ(AP))dτ Where HR is heart rate, RT is reaction time, AP is response mode, α, β, γ are adjustment parameters, and τ represents time variable; S2. Secondly, an intelligent parameter recommendation system based on user data is used to apply the machine learning model formula F2 to determine the personalized scene environment parameter P, which is: where p n is the scene parameter, λ n is the attenuation factor, ω is the weight coefficient, S n Indicates the psychological state contribution corresponding to each environmental parameter; S3, finally, the dynamic environment adaptation technology uses real-time data streams to dynamically adjust the virtual scene, applying the calculus formula F3, the formula is: Where V i is a virtual environment variable, ρ i ,μ is the adjustment coefficient, C i and H i Represent current and historical sentiment data respectively; The emotion perception module consists of the following parts: P1. First, develop a multimodal emotion recognition system, using function F1 to integrate speech, expression and physiological signal data. The formula is defined as: Where X, Y, and Z represent the data inputs obtained from speech, expression, and physiological sensors, respectively, and a, b, and c are adjustment parameters; P2, then implement the real-time feedback mechanism of emotional state, and dynamically adjust the virtual environment settings through the calculus formula F2, the formula is: Where E represents the emotional state index obtained after F1 function analysis, k and ω are adjustment coefficients; P3, finally, the adaptive learning and personalized adjustment mechanism is adopted to apply the adaptive function F3, which is described as: Where V represents the user's feedback data on the virtual environment, g and λ are the parameters of adaptive learning; The method for constructing an adaptive GAN model includes: First, we build a deep convolutional generative network (DCGAN) and integrate the emotion perception module to provide emotion feature vectors generated by multimodal data including speech, expression, and physiological signals. We adaptively adjust the generated virtual scene content, including color tone, texture, and light, based on the real-time emotion analysis results. Then, an adaptive discriminator network is developed to not only verify the authenticity of the generated scenes, but also evaluate the actual impact of the scenes on users’ emotions. The emotion regulation effect of the scenes is evaluated by analyzing the users’ behavioral and physiological data in a specific virtual environment. Finally, a dynamic feedback optimization mechanism is implemented, using reinforcement learning or online learning strategies to adjust the parameters of the generator and discriminator based on the user's real-time feedback on the scene, so as to optimize the GAN model.

2. The method for constructing a virtual mental health care scene using a generative adversarial network (GAN) according to claim 1, characterized in that The construction of the deep convolutional generation network DCGAN includes: First, a multimodal data processing module is designed, and a deep neural network DNN is used to extract features from speech, facial expressions and physiological signals. Through high-level functions: Where X, Y, and Z represent the data obtained from voice, expression, and physiological sensors, respectively, and a, b, c, α, and β are adjustment parameters; Then adjust the deep convolutional generation network DCGAN so that the generator receives the emotion feature vector as input and applies the calculus formula: Control the generated virtual scene content, including tone, texture and light, where V represents the emotion feature vector, λ i ,ω i Represents the weights and frequency coefficients of different features; Finally, a real-time emotional feedback system is implemented to adjust the generated content according to the user's behavior and physiological reactions in the virtual environment, using an adaptive learning formula: Where θ is the parameter of the generator, and γ, κ, ν are adjustment coefficients.

3. The method for constructing a virtual mental health care scene using a generative adversarial network (GAN) according to claim 1, characterized in that The adaptive discriminator network development structure includes: S1. First, develop an adaptive discriminator network to verify the authenticity of the generated scene, using advanced functions: Analyze the user's behavioral data x and physiological data y in the virtual environment, where α i and β j is the adjustment coefficient, which is used to evaluate the actual impact of the scene on the user's emotions; S2. Then integrate user behavior responses and physiological parameters, including heart rate, galvanic skin response, facial expression and eye movement data, and apply data fusion algorithm: Where z represents the data collected from various sensor devices, γ k and δ k For the adjustment coefficient, provide emotional response data set; S3. Finally, the real-time evaluation and feedback mechanism of emotion regulation effect is used, and the calculus formula is adopted: Where θ represents the parameters of the discriminator, κ, ξ, ω are adjustment coefficients, which enable the discriminator to adjust its evaluation criteria and the output of the generator based on the user's real-time data, thereby constructing emotional responsiveness generation and real-time adaptive adjustment of the virtual environment.

4. The method for constructing a virtual mental health care scene using a generative adversarial network (GAN) according to claim 1, characterized in that The implementation of the dynamic feedback optimization mechanism comprises the following steps: First, we integrate reinforcement learning into the GAN model by defining the reward function: Where G represents the scene output by the generator, x represents the user's behavior and physiological data, and θ G and θ D denote the parameters of the generator and the discriminator respectively, and Ω denotes the space of all emotional states; the reward function guides the training process of the generator and the discriminator based on the user's satisfaction with the generated scene, so that the GAN model obtains a higher reward when generating content that is liked by the user; Then, a real-time data feedback loop is implemented, and the online learning formula is applied to update the model parameters in real time and adjust the weights of the generator and discriminator, where the integration is performed within the user interaction time window T: where η is the learning rate, γ is the time decay factor, R(t) is the actual reward, and is the prediction reward.

5. The method for constructing a virtual mental health care scene using a generative adversarial network (GAN) according to claim 1, characterized in that The real-time interactive feedback system comprises the following three steps: S1. First, we realize real-time interactive data capture, using advanced functions: where S = {s1(t),s2(t),…,s n (t)} represents a series of user behavior and physiological data collected from sensors, α i It is an adjustment parameter used to evaluate the user's emotional state in real time and adjust the virtual scene; S2. Then develop a dynamic scene adjustment algorithm and apply the calculus formula: where X = {x1(t), x2(t), …, x m (t)} represents the dynamically adjusted scene parameters, including lighting and sound, γ j and β j is the adjustment factor; S3. Finally, deploy the machine learning optimization strategy and use the function: Where θ represents the parameters of the scene generation algorithm, and φ(t) represents the weight function obtained from user feedback, which is used to optimize the generation algorithm in real time and improve user experience.

Images