A hand-held stress relieving pinching device for student emotional management and psychological counseling

Abstract

The application discloses a hand-held pressure relieving pinch tool for student emotion management and psychological counseling, relates to the technical field of human-computer interaction and psychophysiological monitoring, and comprises a pinch tool body, which is internally integrated with a pressure sensing module, a physiological signal acquisition module and a tactile feedback module. The pressure sensing module is used for collecting pinch pressure values and time variation curves of the pinch tool body applied by a user in real time. The physiological signal acquisition module is used for collecting physiological signals of the user. The tactile feedback module is used for providing tactile feedback to the user. The processing module is arranged in communication connection with the modules and is used for executing synchronous recording and correlation analysis of multi-mode data streams. The system can simultaneously acquire mechanical action data and autonomic nervous physiological response data of the user without interfering with natural pinch behavior of the user, so that the contradiction that a traditional pressure relieving pinch tool cannot perceive a real emotional state of the user and a simple psychological counseling software lacks embodied behavior input is solved.

Description

Technical Field

[0001] This invention relates to the field of human-computer interaction and psychophysiological monitoring technology, specifically to a handheld stress-relief device for student emotion management and psychological counseling. Background Technology

[0002] Currently, products used for emotion management and psychological counseling mainly fall into two categories. One category consists of physical stress-reduction tools, represented by hand-held kneading devices, stress balls, and squeeze toys. Users release tension or shift attention through repeated squeezing and rubbing. These products are simple in structure and easy to use, but their function is limited to providing tactile or force feedback; they cannot perceive the user's true emotional state, nor can their stress-reduction effect be quantitatively assessed. The other category comprises digital psychological counseling solutions, primarily mobile applications, online psychological counseling platforms, and emotion monitoring systems based on biosignals. These systems identify user emotions through questionnaires, heart rate variability analysis, and skin conductance monitoring, and provide relaxation guidance, cognitive behavioral therapy, and other content.

[0003] However, traditional stress-relief kneading devices are merely tools for venting emotions. They lack any form of correlation analysis between the kneading action and the user's autonomic nervous system responses (such as skin conductance and heart rate changes). This prevents the system from utilizing the emotional information inherent in the kneading behavior itself, and from adjusting feedback strategies based on the user's real-time physiological state. Furthermore, while simple psychological counseling software can identify emotions, its identification criteria are usually independent of the user's physical movements, ignoring the temporal consistency between behavioral characteristics such as kneading frequency and intensity changes and physiological signals. This not only limits the accuracy of emotion recognition but also makes it easy for users to interfere with the recognition results by intentionally controlling their breathing or emotional expression, meaning there is insufficient resistance to faking. Summary of the Invention

[0004] The purpose of this invention is to provide a handheld pressure-reducing device for student emotion management and psychological counseling, addressing the limitations of existing technologies that cannot adjust feedback strategies based on the user's real-time physiological state. Furthermore, while simple psychological counseling software can identify emotions, its identification criteria are usually independent of the user's body movements, ignoring the temporal consistency between behavioral characteristics such as pressure frequency and intensity changes and physiological signals. This not only limits the accuracy of emotion recognition but also makes it easy for users to interfere with the recognition results by intentionally controlling their breathing or emotional expression, indicating insufficient resistance to spoofing.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a handheld stress-relief pinch tool for student emotion management and psychological counseling, comprising a pinch tool body, wherein the body integrates,

[0006] The pressure sensing module is used to collect the pinching pressure value P(t) applied by the user to the kneading body and its time change curve in real time, where t represents time;

[0007] The physiological signal acquisition module is used to acquire the user's skin conductance activity signal G(t) and heart rate H(t) in real time.

[0008] The signal tactile feedback module is used to output variable vibration and damping force feedback F to the user. fb (t);

[0009] The processing module is communicatively connected to the pressure sensing module, the physiological signal acquisition module, and the tactile feedback module.

[0010] The processing module performs the following operations:

[0011] The pinch pressure value P(t), skin conductance signal G(t), and heart rate H(t) are received and recorded synchronously to form a time-correlated multimode data stream. ;

[0012] Based on the multimodal data stream Extract at least one emotion-related feature. ,in The number of features;

[0013] The emotion-related features Input a pre-trained emotion state discrimination model and output the current user's emotion category. and intensity level λ∈[0,1], where K is the total number of emotion categories;

[0014] According to the emotion category And the intensity level λ, dynamically generating feedback signals. This drives the haptic feedback module to execute the corresponding haptic output mode, wherein... This is a preset feedback mapping function;

[0015] The processing module also performs the following operations: classifying the emotion category Intensity level λ and corresponding multimode data stream The emotion profile is stored in the user's emotion profile and used to update the emotion state discrimination model.

[0016] Furthermore, the emotion-related features include:

[0017] Pressure fluctuation frequency domain characteristics ,in [x] represents the Fourier transform, and T is the sampling time window;

[0018] Characteristics of skin conductance response latency ,in This refers to the initial moment of the pinching action. The preset skin conductance response threshold;

[0019] Heart rate variability and phase difference characteristics of pinch rhythm:

[0020]

[0021] in The high-frequency component is obtained after the heart rate signal is high-pass filtered. Let be the Hilbert transform of the pressure signal P(t), and arg denote the complex argument.

[0022] Furthermore, the emotion state discrimination model is a temporal classification model based on a long short-term memory network, and its input is a multimodal feature sequence of length L. The output is:

[0023]

[0024] in Let be the hidden state vector of the LSTM network at time t. To output the weight matrix, For bias terms, It is a normalized exponential function.

[0025] Furthermore, the tactile output mode is determined by a feedback control signal. The decision specifically includes:

[0026] Same frequency mode: When and hour, Where α is the feedback gain coefficient, For high intensity threshold, A collection of venting emotions;

[0027] Inverse guidance mode: when and hour, ,in For the guiding gain coefficient, This is a breathing guidance cycle. This is a collection of anxiety-related emotions.

[0028] Gradual relaxation mode: when hour, ,in The relaxation amplitude coefficient, To relax the guiding frequency, is the relaxation decay time constant.

[0029] Furthermore, the processing module also performs the following operations: calculating the cross-modal consistency index. :

[0030]

[0031] in To achieve average kneading pressure, The average skin conductance signal;

[0032] when When the processing module determines that the user is faking emotions, it marks the interaction data for model correction. This is the preset consistency threshold.

[0033] Furthermore, the user sentiment profile includes an incrementally updated user feature vector. :

[0034]

[0035] in This is the user feature vector after the nth interaction. The emotion-related features extracted during the nth interaction. The learning rate;

[0036] The update of the emotion state discrimination model is based on the user feature vector. The following loss function is used:

[0037]

[0038] in Let i be the true sentiment label of the i-th sample. Here, N represents the probability predicted by the model, and N is the sample size. The regularization coefficient is . This is the group reference feature vector.

[0039] Furthermore, the processing module also performs the following operations: calculating the time delay between the pinching action and the skin conductance response. :

[0040]

[0041] in, This is the current analysis time. For time delay variables, W is the maximum allowable delay; W is the width of half the window. It is a Gaussian weighted function. The standard deviation is the window value; the processing module will use the time delay. As the emotion-related feature One component.

[0042] Furthermore, the processing module also performs the following operations: calculating the correlation coefficient between the peak-to-trough frequency of pressure and the low-frequency and high-frequency power ratio of heart rate variability:

[0043]

[0044] in, The peak-valley frequency of the pressure signal P(t) is calculated by detecting the time interval between local maxima and minima of the pressure signal.

[0045] The ratio of low-frequency power to high-frequency power of heart rate H(t), where low-frequency power... High-frequency power ;

[0046] This represents the covariance operation. This represents the standard deviation calculation;

[0047] The processing module uses the correlation coefficient as the emotion association feature. One component.

[0048] Furthermore, the processing module also performs the following operations: calculating the sample entropy of the pinching intensity sequence. :

[0049]

[0050] Where P is the sequence of kneading pressure values. N is the sequence length;

[0051] m is the embedding dimension, representing the length of the subsequence used for comparison;

[0052] r is the similarity tolerance, which takes the value of 0.1 to 0.25 times the standard deviation of the pressure signal;

[0053] A is the number of similar subsequence pairs of length m+1;

[0054] B is the number of similar subsequence pairs of length m;

[0055] The processing module will process the sample entropy. As the emotion-related feature One component is used to characterize the irregularity of the user's pinching behavior.

[0056] Furthermore, the processing module also performs the following operations: executes an online adaptive update mechanism to update the emotion state discrimination model. :

[0057]

[0058]

[0059] in, The model parameters are at time t;

[0060] The learning rate decays over time. The initial learning rate;

[0061] loss function Gradients with respect to model parameters;

[0062] The emotion-related features extracted at time t;

[0063] The emotion category label for time t;

[0064] The decay time constant;

[0065] The loss function Using cross-entropy loss:

[0066]

[0067] in, For indicator functions; when the emotion category C equals the first... kind The value is 1 if it is true, and 0 otherwise. For the model to predict belonging to the first The probability of a class.

[0068] Compared with the prior art, the present invention integrates a pressure sensing module, a physiological signal acquisition module and a tactile feedback module into the hand-held kneading device body, and sets up a processing module that communicates with the above modules to perform synchronous recording and correlation analysis of multi-mode data streams. This allows the system to acquire the user's mechanical action data and autonomic nervous physiological response data simultaneously without interfering with the user's natural kneading behavior. This solves the contradiction that traditional stress-relief kneading devices cannot sense the user's true emotional state and that simple psychological counseling software lacks embodied behavioral input.

[0069] By extracting emotion-related features, including frequency domain features of pressure fluctuations, skin conductance response latency, and the phase difference between heart rate variability and pinching rhythm, through the processing module, and inputting these features into a pre-trained emotion state discrimination model, the system can identify emotions based on the temporal consistency between the user's pinching behavior and physiological response, significantly improving the accuracy of emotion judgment and its anti-faking ability. Furthermore, by dynamically generating feedback control signals based on emotion category and intensity level, the tactile feedback module outputs the same frequency mode, the opposite phase guidance mode, or the progressive relaxation mode. This allows the tactile feedback to form a personalized closed-loop regulation with the user's current emotional state. Specifically, when the user is in a cathartic mood, their pinching behavior is strengthened to promote catharsis; when the user is in an anxious mood, their breathing rhythm is guided with an opposite phase rhythm; and when the emotion intensity is low, a sinusoidal relaxation wave is used to induce deep relaxation. This achieves immediate and adaptive regulation from emotion perception to tactile intervention.

[0070] By calculating the cross-modal consistency index and judging emotional masquerading behavior through the processing module, the system can identify users' intentional or unintentional concealment responses. This data is then labeled and used for model correction, effectively avoiding training data contamination and improving the model's stability in real-world scenarios. By constructing incrementally updated user feature vectors and updating the emotional state discrimination model based on a regularized loss function, the system can achieve personalized adaptive learning while protecting user privacy. That is, as the number of uses increases, the model parameters gradually converge to the emotional expression patterns of specific users, thus providing increasingly accurate emotion recognition and intervention effects.

[0071] By processing modules to calculate the Gaussian weighted cross-correlation time delay between the pinching action and the skin conductance response, the correlation coefficient between the pressure peak and valley frequencies and the low-frequency to high-frequency power ratio of heart rate variability, and the sample entropy of the pinching intensity sequence, the system can capture subtle differences in the user's emotional state from multiple complementary time domain, frequency domain, and nonlinear dynamic dimensions, forming a rich feature space and providing highly discriminative input information for the emotion discrimination model. Attached Figure Description

[0072] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0073] Figure 1 This is a schematic diagram of the overall structure of the kneading tool body provided in an embodiment of the present invention;

[0074] Figure 2 This is a schematic diagram of the overall process provided for an embodiment of the present invention. Detailed Implementation

[0075] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0076] As attached Figure 1 To be continued Figure 2 As shown:

[0077] Example:

[0078] This invention provides a handheld stress-relief tool for student emotion management and psychological counseling, comprising a tool body, which integrates...

[0079] The pressure sensing module is used to collect the pinching pressure value P(t) applied by the user to the kneading body and its time change curve in real time, where t represents time;

[0080] The physiological signal acquisition module is used to acquire the user's skin conductance activity signal G(t) and heart rate H(t) in real time.

[0081] The signal tactile feedback module is used to output variable vibration and damping force feedback F to the user. fb (t);

[0082] The processing module is communicatively connected to the pressure sensing module, the physiological signal acquisition module, and the tactile feedback module.

[0083] The processing module performs the following operations:

[0084] Receives and synchronously records the pinch pressure value P(t), skin conductance signal G(t), and heart rate H(t), forming a time-correlated multimode data stream. ;

[0085] Based on multimodal data stream Extract at least one emotion-related feature. ,in The number of features;

[0086] Emotion-related features Input a pre-trained emotion state discrimination model and output the current user's emotion category. and intensity level λ∈[0,1], where K is the total number of emotion categories;

[0087] According to emotion category And the intensity level λ, dynamically generating feedback signals. The haptic feedback module is driven to execute the corresponding haptic output mode, where This is a preset feedback mapping function;

[0088] The processing module also performs the following operations: categorizes emotions. Intensity level λ and corresponding multimode data stream The emotion profile is stored in the user's emotional profile and used to update the emotion state discrimination model.

[0089] Furthermore, emotion-related features include:

[0090] Pressure fluctuation frequency domain characteristics ,in [x] represents the Fourier transform, and T is the sampling time window;

[0091] Characteristics of skin conductance response latency ,in This refers to the initial moment of the pinching action. The preset skin conductance response threshold;

[0092] Heart rate variability and phase difference characteristics of pinching rhythm:

[0093]

[0094] in The high-frequency component is obtained after the heart rate signal is high-pass filtered. Let be the Hilbert transform of the pressure signal P(t), and arg denote the complex argument.

[0095] The emotion state discrimination model is a temporal classification model based on a long short-term memory network, and its input is a multimodal feature sequence of length L. The output is:

[0096]

[0097] in Let be the hidden state vector of the LSTM network at time t. To output the weight matrix, For bias terms, It is a normalized exponential function.

[0098] In addition, the tactile output mode is controlled by feedback signals. The decision specifically includes:

[0099] Same frequency mode: When and hour, Where α is the feedback gain coefficient, For high intensity threshold, A collection of venting emotions;

[0100] Inverse guidance mode: when and hour, ,in For the guiding gain coefficient, This is a breathing guidance cycle. This is a collection of anxiety-related emotions.

[0101] Gradual relaxation mode: when hour, ,in The relaxation amplitude coefficient, To relax the guiding frequency, is the relaxation decay time constant.

[0102] Preferably, the processing module further performs the following operations: calculating the cross-modal consistency index. :

[0103]

[0104] in To achieve average kneading pressure, The average skin conductance signal;

[0105] when At that time, the processing module determines that the user is faking emotions and marks the interaction data for model correction. This is the preset consistency threshold.

[0106] It should be noted that the user sentiment profile includes an incrementally updated user feature vector. :

[0107]

[0108] in This is the user feature vector after the nth interaction. The emotion-related features extracted during the nth interaction. The learning rate;

[0109] The update of the emotion state discrimination model is based on user feature vectors. The following loss function is used:

[0110]

[0111] in Let i be the true sentiment label of the i-th sample. Here, N represents the probability predicted by the model, and N is the sample size. The regularization coefficient is . This is the group reference feature vector.

[0112] Furthermore, the processing module also performs the following operations: calculating the time delay between the pinching action and the skin conductance response. :

[0113]

[0114] in, This is the current analysis time. For time delay variables, W is the maximum allowable delay; W is the width of half the window. It is a Gaussian weighted function. The standard deviation of the window; the processing module will delay the time. As a feature associated with emotion One component.

[0115] Specifically, the processing module also performs the following operations: calculating the correlation coefficient between the peak and trough frequencies of pressure and the low-frequency and high-frequency power ratio of heart rate variability:

[0116]

[0117] in, The peak-valley frequency of the pressure signal P(t) is calculated by detecting the time interval between local maxima and minima of the pressure signal.

[0118] The ratio of low-frequency power to high-frequency power of heart rate H(t), where low-frequency power... High-frequency power ;

[0119] This represents the covariance operation. This represents the standard deviation calculation;

[0120] The processing module uses the correlation coefficient as a feature related to sentiment. One component.

[0121] Furthermore, the processing module also performs the following operations: calculating the sample entropy of the pinching intensity sequence. :

[0122]

[0123] Where P is the sequence of kneading pressure values. N is the sequence length;

[0124] m is the embedding dimension, representing the length of the subsequence used for comparison;

[0125] r is the similarity tolerance, which takes the value of 0.1 to 0.25 times the standard deviation of the pressure signal;

[0126] A is the number of similar subsequence pairs of length m+1;

[0127] B is the number of similar subsequence pairs of length m;

[0128] The processing module will process the sample entropy. As a feature associated with emotion One component is used to characterize the irregularity of the user's pinching behavior.

[0129] Furthermore, the processing module also performs the following operations: it executes an online adaptive update mechanism to update the emotion state discrimination model. :

[0130]

[0131]

[0132] in, The model parameters are at time t;

[0133] The learning rate decays over time. The initial learning rate;

[0134] loss function Gradients with respect to model parameters;

[0135] The emotion-related features extracted at time t;

[0136] The emotion category label for time t;

[0137] The decay time constant;

[0138] loss function Using cross-entropy loss:

[0139]

[0140] in, For indicator functions; when the emotion category C equals the first... kind The value is 1 if it is true, and 0 otherwise. For the model to predict belonging to the first The probability of a class.

[0141] Working principle: The user takes out the hand-held pressure-relief device. The device integrates a pressure sensing module, a physiological signal acquisition module, and a tactile feedback module. A processing module is also located inside the device, communicating with the other three modules. When the user holds the device, the system automatically starts and enters standby mode.

[0142] The user begins applying pressure to the kneading device, and the pressure sensing module collects the pressure applied by the user and its change over time in real time. Simultaneously, the physiological signal acquisition module begins synchronously acquiring the user's ductal skin activity (ESI) and heart rate signals. The ESI is acquired through two electrode pads on the surface of the kneading device, while the heart rate is acquired through an optical sensor. These three signals—pressure value, ESI, and heart rate—are recorded synchronously on the time axis, forming a time-correlated multimodal data stream.

[0143] After receiving the aforementioned multimodal data stream, the processing module begins extracting multiple emotion-related features. First, it calculates the frequency domain features of pressure fluctuations, performing a Fourier transform on the curve of the squeezing pressure value over time to analyze the energy distribution of the pressure signal at different frequencies. Simultaneously, it calculates the latency features of the skin conductance response, detecting the moment the user's squeezing action begins and then monitoring the time required for the skin conductance activity signal to first rise above a preset threshold from that moment. The module also calculates the phase difference features between heart rate variability and the squeezing rhythm, extracting high-frequency components from the heart rate signal, performing a Hilbert transform on the pressure signal, and then calculating the phase difference between the two. Finally, the module calculates the time delay between the squeezing action and the skin conductance response, using a Gaussian weighted cross-correlation algorithm to calculate the correlation between the squeezing pressure value and the skin conductance activity signal within a time window before and after the current analysis moment, finding the time offset that maximizes the correlation. The processing module also calculates the correlation coefficient between the peak and trough frequencies of pressure and the ratio of low-frequency to high-frequency power in heart rate variability. This is achieved by detecting the time interval between local maxima and minima in the pinch pressure signal to obtain the peak and trough frequencies. Simultaneously, frequency domain analysis of the heart rate signal is performed to calculate the ratio of low-frequency to high-frequency power, and then the ratio of the covariance to the standard deviation between these two quantities is calculated. Finally, the processing module calculates the sample entropy of the pinch pressure sequence. This involves arranging the pinch pressure values ​​in chronological order, setting an embedding dimension and similarity tolerance, counting the number of similar subsequence pairs with a length equal to the embedding dimension and the length plus one, calculating the ratio of the two and taking the negative logarithm. This sample entropy characterizes the irregularity of the user's pinching behavior. These six features together constitute the emotion-related feature vector.

[0144] The processing module then calculates the cross-modal consistency index. Specifically, it calculates the average pinch pressure value within a time window as the average pinch pressure, and simultaneously calculates the average ductal skin activity signal within the same time window as the average ductal skin signal. Then, it calculates the ratio of the pinch pressure value to the average pinch pressure and the ratio of the ductal skin activity signal to the average ductal skin signal. The absolute value of the difference between these two values ​​is integrated over the time window, and half of the integral result is subtracted from one to obtain the cross-modal consistency index. If the index is less than a preset consistency threshold, the processing module determines that the user is faking emotions and marks this interaction data for subsequent model correction; if the index is greater than or equal to the preset threshold, it is determined to be a genuine emotional expression, and the subsequent steps continue.

[0145] The processing module inputs the extracted emotion-related feature vectors into a pre-trained emotion state discrimination model. This model is a temporal classification model based on a long short-term memory network, and its input is a historical multi-modal feature sequence of a preset length. The model outputs the current user's emotion category and intensity level. The emotion category includes venting emotions, anxious emotions, calm emotions, etc., and the intensity level is a continuous value from zero to one, with higher values ​​representing greater emotion intensity.

[0146] The processing module generates feedback control signals based on the emotion category and intensity level using a preset feedback mapping function, driving the tactile feedback module to execute the corresponding tactile output mode. When the emotion intensity level is greater than or equal to the high-intensity threshold and the emotion category belongs to the cathartic emotion set, the processing module generates a control signal in the same frequency mode. The feedback force output by the tactile feedback module is proportional to the user's current squeezing pressure value; that is, the feedback force increases with increasing squeezing pressure and decreases with decreasing squeezing pressure, resonating with the squeezing action and thus reinforcing the user's cathartic behavior. When the emotion intensity level is greater than or equal to the high-intensity threshold and the emotion category belongs to the anxious emotion set, the processing module generates a control signal in the opposite phase guidance mode. The feedback force output by the tactile feedback module is inversely related to the user's current squeezing pressure value; that is, the feedback force decreases when the user's squeezing pressure increases and increases when the user's squeezing pressure decreases. Simultaneously, the cycle of feedback force change is synchronized with a preset breathing guidance cycle, guiding the user to adjust their breathing according to the rhythm of the feedback force, thereby alleviating anxiety. When the emotional intensity level is less than the high intensity threshold, the processing module generates a control signal for a progressive relaxation mode. The feedback force output by the haptic feedback module presents a sinusoidal decay waveform, meaning that the feedback force fluctuates according to a sinusoidal law and the amplitude gradually decreases over time, inducing the user to enter a state of deep relaxation.

[0147] The processing module stores the emotion category, intensity level, and corresponding multimodal data stream from the current interaction into the user's emotion profile. The user emotion profile includes an incrementally updated user feature vector, which is updated with each interaction: the new user feature vector equals the difference between the previous user feature vector multiplied by one and the learning rate, plus the extracted emotion-related features multiplied by the learning rate. Simultaneously, based on the updated user feature vector, the processing module uses a regularized loss function to update the emotion state discrimination model online. This loss function consists of cross-entropy loss and a regularization term, and the model parameters are optimized using gradient descent by minimizing this loss function. The learning rate decays over time; that is, the current learning rate equals the initial learning rate multiplied by one plus the time interval divided by the reciprocal of the decay time constant. This design allows the model to learn quickly in the early stages and gradually stabilize later.

[0148] Users can continue using the tool, and the system will return to the data collection phase to begin a new round of collection, processing, and feedback. When the user stops using the tool, the system will automatically save all data from this interaction and enter standby or shutdown mode.

[0149] This embodiment combines the physical tactile channel of a handheld pressure-relief device with multimodal biosignal processing and closed-loop feedback control to achieve real-time, adaptive adjustment from emotion perception to tactile intervention. Compared with traditional solutions, this solution has the ability to perform cross-modal correlation analysis between the pinching action and physiological signals, thus identifying the user's true emotional state and possessing anti-faking capabilities; it has adaptive tactile feedback capabilities based on emotion category and intensity level, thus providing personalized intervention strategies according to the user's current state; it has the ability to detect faking behavior and label data driven by cross-modal consistency index, thus effectively improving the model's stability in real-world usage scenarios; and it has incrementally updated user feature vectors and an online adaptive learning mechanism, thus enabling personalized model optimization while protecting user privacy.

[0150] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A handheld stress-relief tool for student emotion management and psychological counseling, characterized in that, Including the mold body, which integrates internal components, The pressure sensing module is used to collect the pinching pressure value P(t) applied by the user to the kneading body and its time change curve in real time, where t represents time; The physiological signal acquisition module is used to acquire the user's skin conductance activity signal G(t) and heart rate H(t) in real time. The signal tactile feedback module is used to output variable vibration and damping force feedback F to the user. fb (t); The processing module is communicatively connected to the pressure sensing module, the physiological signal acquisition module, and the tactile feedback module. The processing module performs the following operations: The pinch pressure value P(t), skin conductance signal G(t), and heart rate H(t) are received and recorded synchronously to form a time-correlated multimode data stream. ; Based on the multimodal data stream Extract at least one emotion-related feature. ,in The number of features; The emotion-related features Input a pre-trained emotion state discrimination model and output the current user's emotion category. and intensity level λ∈[0,1], where K is the total number of emotion categories; According to the emotion category And the intensity level λ, dynamically generating feedback signals. This drives the haptic feedback module to execute the corresponding haptic output mode, wherein... This is a preset feedback mapping function; The processing module also performs the following operations: classifying the emotion category Intensity level λ and corresponding multimode data stream The emotion profile is stored in the user's emotion profile and used to update the emotion state discrimination model.

2. The handheld stress-relief device for student emotion management and psychological counseling according to claim 1, characterized in that, The emotion-related features include: Pressure fluctuation frequency domain characteristics ,in [x] represents the Fourier transform, and T is the sampling time window; Characteristics of skin conductance response latency ,in This refers to the initial moment of the pinching action. The preset skin conductance response threshold; Heart rate variability and phase difference characteristics of pinching rhythm: in The high-frequency component is obtained after the heart rate signal is high-pass filtered. Let be the Hilbert transform of the pressure signal P(t), and arg denote the complex argument.

3. A handheld stress-relief device for student emotion management and psychological counseling as described in claim 1, characterized in that, The emotion state discrimination model is a temporal classification model based on a long short-term memory network, and its input is a multimodal feature sequence of length L. The output is: in Let be the hidden state vector of the LSTM network at time t. To output the weight matrix, For bias terms, It is a normalized exponential function.

4. A handheld stress-relief device for student emotion management and psychological counseling as described in claim 1, characterized in that, The tactile output mode is controlled by a feedback control signal. The decision specifically includes: Same frequency mode: When and hour, Where α is the feedback gain coefficient, For high intensity threshold, A collection of venting emotions; Inverse guidance mode: when and hour, ,in For the guiding gain coefficient, This is a breathing guidance cycle. This is a collection of anxiety-related emotions. Gradual relaxation mode: when hour, ,in The relaxation amplitude coefficient, To relax the guiding frequency, is the relaxation decay time constant.

5. A handheld stress-relief device for student emotion management and psychological counseling according to claim 1, characterized in that, The processing module also performs the following operations: calculating the cross-modal consistency index. : in To achieve average kneading pressure, The average skin conductance signal; when When the processing module determines that the user is faking emotions, it marks the interaction data for model correction. This is the preset consistency threshold.

6. A handheld stress-relief device for student emotion management and psychological counseling according to claim 1, characterized in that, The user sentiment profile includes an incrementally updated user feature vector. : in This is the user feature vector after the nth interaction. The emotion-related features extracted during the nth interaction. The learning rate; The update of the emotion state discrimination model is based on the user feature vector. The following loss function is used: in Let i be the true sentiment label of the i-th sample. Here, N represents the probability predicted by the model, and N is the sample size. The regularization coefficient is . This is the group reference feature vector.

7. A handheld stress-relief device for student emotion management and psychological counseling according to claim 1, characterized in that, The processing module also performs the following operations: calculating the time delay between the pinching action and the skin conductance response. : in, This is the current analysis time. For time delay variables, W is the maximum allowable delay; W is the width of half the window. It is a Gaussian weighted function. The standard deviation is the window value; the processing module will use the time delay. As the emotion-related feature One component.

8. A handheld stress-relief device for student emotion management and psychological counseling according to claim 1, characterized in that, The processing module also performs the following operations: calculating the correlation coefficient between the peak and trough frequencies of pressure and the low-frequency and high-frequency power ratio of heart rate variability: in, The peak-valley frequency of the pressure signal P(t) is calculated by detecting the time interval between local maxima and minima of the pressure signal. The ratio of low-frequency power to high-frequency power of heart rate H(t), where low-frequency power... High-frequency power ; This represents the covariance operation. This represents the standard deviation calculation; The processing module uses the correlation coefficient as the emotion association feature. One component.

9. A handheld stress-relief device for student emotion management and psychological counseling according to claim 1, characterized in that, The processing module also performs the following operations: calculating the sample entropy of the pinching pressure sequence. : Where P is the sequence of kneading pressure values. N is the sequence length; m is the embedding dimension, representing the length of the subsequence used for comparison; r is the similarity tolerance, which takes the value of 0.1 to 0.25 times the standard deviation of the pressure signal; A is the number of similar subsequence pairs of length m+1; B is the number of similar subsequence pairs of length m; The processing module will process the sample entropy. As the emotion-related feature One component is used to characterize the irregularity of the user's pinching behavior.

10. A handheld stress-relief device for student emotion management and psychological counseling according to claim 1, characterized in that, The processing module also performs the following operations: executes an online adaptive update mechanism to update the emotion state discrimination model. : in, The model parameters are at time t; The learning rate decays over time. The initial learning rate; loss function Gradients with respect to model parameters; The emotion-related features extracted at time t; The emotion category label for time t; The decay time constant; The loss function Using cross-entropy loss: in, For indicator functions; when the emotion category C equals the first... kind The value is 1 if the condition is met, and 0 otherwise. For the model to predict belonging to the first The probability of a class.

Images