Emotion measurement and control method and wearable device

Abstract

The invention discloses an emotion measurement and control method and wearable equipment. The method comprises the following steps: acquiring an electroencephalogram signal and an electrocardiosignal of a subject in real time; performing feature extraction and feature fusion on the electroencephalogram signal and the electrocardiosignal to obtain fused feature data; analyzing the fused feature data to obtain emotion test data; generating intervention control parameters according to the emotion test data; and outputting an emotion intervention wave to the subject according to the intervention control parameter so as to regulate and control the current emotion of the subject. According to the technical scheme, accurate recognition of the emotional state can be achieved, emotion intervention is triggered on the basis of the recognition result, the intervention content is dynamically adjusted, an integrated closed-loop scheme of emotion measurement and emotion intervention is formed, and physical and psychological health development of people is promoted. And the wearable equipment provided by the invention is good in portability.

Description

[0001] This application claims priority to Chinese Patent Application No. 202411355246.0, filed on September 27, 2024. TECHNICAL FIELD

[0002] The present application relates to the field of emotion measurement and intervention, and more particularly, to an emotion measurement and control method and a wearable device. BACKGROUND

[0003] Emotion measurement is a core means of assessing an individual's psychological state, which can be achieved through various ways such as subjective questionnaire survey and objective physiological indicator monitoring. These methods help identify emotional states such as anxiety, depression or excitement, providing a data basis for mental health intervention. Emotion intervention is based on measurement results to take targeted actions, such as cognitive behavioral therapy to alleviate emotional problems by adjusting negative thought patterns, meditation exercises to enhance attention and reduce stress, and oxygen exercise to promote the release of endorphins to enhance positive emotions.

[0004] However, existing emotion measurement and intervention technologies have many limitations. First, traditional questionnaire surveys are highly subjective, affected by recall bias, and cannot be monitored in real time. Physiological indicator monitoring devices are often bulky and not portable, limiting their daily use. Second, existing intervention methods are mostly static, resulting in suboptimal intervention effectiveness. In addition, emotion measurement and intervention are usually performed separately, lacking integrated design, and cannot achieve a closed-loop system of measuring and intervening simultaneously. This results in users being unable to obtain timely and personalized support when facing emotional fluctuations.

[0005] Therefore, how to design an accurate emotion measurement method and its portable device, integrating personalized real-time intervention strategies, to promote the physical and mental health development of people, has become a technical challenge that needs to be solved in the field. SUMMARY

[0006] In view of the above, in a first aspect, the present application provides an emotion measurement and control method, comprising:

[0007] obtaining brain electrical signals and electrocardiogram signals of a subject in real time;

[0008] extracting features and fusing features from the brain electrical signals and electrocardiogram signals to obtain fused feature data;

[0009] analyzing the fused feature data to obtain emotion test data;

[0010] generating intervention control parameters according to the emotion test data;

[0011] outputting an emotion intervention wave to the subject according to the intervention control parameters to regulate the current emotion of the subject.

[0012] Preferably, before performing feature extraction and feature fusion on the electroencephalogram (EEG) and electrocardiogram (ECG) signals, the method further includes:

[0013] One or more steps are performed on the electroencephalogram (EEG) signal, including bandpass filtering, power frequency notch filtering, and independent component analysis.

[0014] One or more steps are performed on the electrocardiogram signal, including dynamic quality monitoring, bandpass filtering, and baseline drift correction.

[0015] Preferably, feature extraction and feature fusion are performed on the electroencephalogram (EEG) signals and electrocardiogram (ECG) signals to obtain fused feature data, including:

[0016] Extracting EEG features from the EEG signal; the EEG features include one or more of the following: power of each frequency band, the ratio of power of each frequency band to the total power, Herchos parameters, and nonlinear indices;

[0017] Extract electrocardiogram (ECG) features from the ECG signal; the ECG features include one or more of the time-domain and frequency-domain indices of heart rate variability.

[0018] The extracted EEG and ECG features are fused using an attention mechanism to obtain fused feature data.

[0019] Preferably, the fused feature data is parsed to obtain the subject's current emotion test data, including:

[0020] The fused feature data is analyzed using a deep learning model to obtain the subject's current emotional numerical index, the maximum confidence level corresponding to the emotional category, and the short-term emotional trend.

[0021] Further preferably, intervention control parameters are generated based on the subject's current emotional test data, including:

[0022] Under the rule-based mapping, intervention control parameters are generated based on the subject's current numerical emotion indicators, the maximum confidence level corresponding to the emotion category, the short-term emotion trend, and the user preference profile. The calculation formula is as follows:

[0023] M(t) = G(y(t),s(t),T(t),U);

[0024] Where t represents the current frame, M represents the intervention control parameters, G represents the regularization mapping, U represents the user preference profile, y(t) represents the numerical sentiment index of the previous frame, s(t) represents the maximum confidence level corresponding to the sentiment category of the current frame, and T(t) represents the short-term sentiment trend of the current frame.

[0025] More preferably, the formula for calculating the short-term sentiment trend is:

[0026] T(t)=β·T(t-1)+(1-β)·y(t);

[0027] Where T(t) represents the short-term sentiment trend of the current frame t, T(t) represents the short-term sentiment trend of the previous frame, β represents the smoothing coefficient, and y(t) represents the numerical sentiment index of the current frame t.

[0028] More preferably, the regularized mapping includes:

[0029] If the emotion category is negative and the corresponding maximum confidence value is greater than the first threshold, then a soothing intervention is performed.

[0030] If the emotion category is positive, then increase the volume as an intervention.

[0031] More preferably, the regularized mapping further includes:

[0032] If the current emotion category is negative or positive, and the corresponding maximum confidence value is consistently above or not less than the second threshold for the first time, then it is determined that intervention needs to be initiated.

[0033] If the current emotion category is negative or positive, and the corresponding maximum confidence value remains no greater than the third threshold for a second period of time, then the intervention is determined to be reduced or stopped; the second threshold is greater than the third threshold.

[0034] Further preferably, parsing the fused feature data to obtain the subject's current emotion test data also includes:

[0035] The fused feature data is analyzed using a deep learning model to obtain the subject's current mood improvement rate;

[0036] The regularized mapping also includes:

[0037] If the mood improvement rate is less than the predetermined improvement threshold, the music style will be changed, the intervention intensity will be increased, or the intervention mode will be switched to voice accompaniment.

[0038] If the mood improvement rate is not less than the predetermined improvement threshold, the intervention intensity is maintained or reduced.

[0039] In a second aspect, this application proposes a wearable device, characterized in that the wearable device comprises:

[0040] The data acquisition device is used to collect the subject's electroencephalogram (EEG) and electrocardiogram (ECG) signals in real time.

[0041] The processor is used to acquire the subject's electroencephalogram (EEG) and electrocardiogram (ECG) signals in real time.

[0042] Feature extraction and feature fusion are performed on the electroencephalogram (EEG) and electrocardiogram (ECG) signals to obtain fused feature data.

[0043] The fused feature data is analyzed to obtain emotion test data;

[0044] Based on emotion test data, generate intervention control parameters;

[0045] An intervention device is used to output an emotional intervention wave to the subject according to the intervention control parameters, so as to regulate the subject's current emotion.

[0046] Preferably, the wearable device is an earphone-type device, comprising: a headband and earphones;

[0047] The headgear includes an elastic clamping mechanism to fit the subject's head.

[0048] More preferably, the acquisition device includes: an electroencephalogram (EEG) signal acquisition device;

[0049] The EEG signal acquisition device is placed inside the headband and collects EEG signals from the subject's forehead region.

[0050] More preferably, the acquisition device further includes: an electrocardiogram (ECG) signal acquisition device;

[0051] The electrocardiogram (ECG) signal acquisition device is placed inside the headband and collects ECG signals from the temporal region of the subject.

[0052] Alternatively, the ECG signal acquisition device is an external chest patch ECG acquisition device.

[0053] More preferably, the processor is placed inside the headpiece.

[0054] More preferably, the intervention device is an earphone.

[0055] The technical solution provided in this application acquires the subject's electroencephalogram (EEG) and electrocardiogram (ECG) signals in real time. After analyzing the EEG and ECG signals, emotion test data is obtained. Intervention control parameters are generated based on the emotion test data to regulate the subject's current emotion. Through a multimodal feature fusion algorithm of EEG and ECG signals, the emotional state is accurately identified. Based on the identification results, emotion intervention is triggered and the intervention content is dynamically adjusted, forming an integrated closed-loop solution of emotion measurement and emotion intervention to promote people's physical and mental health development.

[0056] Other features and advantages of this application will be described in detail in the following detailed description section. Attached Figure Description

[0057] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application, and the illustrative embodiments and descriptions thereof are used to explain this application. In the drawings:

[0058] Figure 1 This is a flowchart of an emotion measurement and control method according to a preferred embodiment of this application;

[0059] Figure 2 This is a schematic diagram of a wearable device according to a preferred embodiment of this application;

[0060] Figure 3 This is a schematic diagram of a headgear according to a preferred embodiment of this application;

[0061] Figure 4 This is a circuit diagram of a wearable device according to a preferred embodiment of this application;

[0062] Figure 5 This is a schematic diagram of a processor according to a preferred embodiment of the present application. Detailed Implementation

[0063] The technical solution of this application will now be described in detail with reference to the accompanying drawings and embodiments.

[0064] Accurate emotion measurement not only provides a scientific basis for intervention strategies but also optimizes intervention effects through real-time feedback, thereby significantly improving an individual's overall mental health. Therefore, this application first proposes an emotion measurement and control method for measuring and intervening in the emotions of subjects, such as... Figure 1 As shown, the method includes steps 110-150:

[0065] Step 110: Acquire the subject's electroencephalogram (EEG) and electrocardiogram (ECG) signals in real time;

[0066] Specifically, electroencephalogram (EEG) signals and electrocardiogram (ECG) signals are two different bioelectrical signals. EEG signals mainly reflect the collective activity of neurons in the brain, while ECG signals mainly reflect the electrical excitation conduction state of the heart. Both can reflect emotional fluctuations. Therefore, this application simultaneously acquires the subject's EEG and ECG signals to more comprehensively and accurately reflect the subject's emotional fluctuations.

[0067] Step 120: Extract and fuse features from EEG and ECG signals to obtain fused feature data;

[0068] Specifically, before feature extraction from EEG and ECG signals, preprocessing is required to enhance data quality and usability.

[0069] In one specific implementation, the preprocessing includes one or more steps of bandpass filtering, power frequency notch filtering, and independent component analysis of the EEG signal, and one or more steps of dynamic quality monitoring, bandpass filtering, and baseline drift correction of the ECG signal.

[0070] For EEG signals, the raw acquired signals are bandpass filtered at 0.5Hz–45Hz and subjected to power frequency notch filtering at 50Hz or 60Hz. Then, artifact detection and removal for EEG and EMG are performed in the time domain. For persistent artifacts, a blind source separation method based on Independent Component Analysis (ICA) is used to correct them. ICA treats multi-channel observed signals as an instantaneous linear mixture of several statistically independent source signals. Utilizing only the assumption that the sources are independent, and by maximizing non-Gaussianity or minimizing mutual information, the mixture matrix can be simultaneously solved and each independent source recovered without prior knowledge, achieving "blind" source separation and ensuring the effectiveness and robustness of subsequent feature extraction.

[0071] For ECG signals, the sampling rate can be 250Hz–1000Hz, preferably 500Hz. The original acquired signal is then subjected to bandpass filtering (0.5Hz–40Hz) and baseline drift correction. Baseline drift correction removes slow, low-frequency fluctuations in the signal caused by respiration, electrode movement, or amplifier temperature drift, ensuring accurate detection of feature points and amplitudes. Simultaneously, dynamic signal quality index (SQI) monitoring is performed. When contact quality or motion artifacts exceed a threshold, such as sudden amplitude changes, noise power fluctuations, or impedance jumps, a prompt is triggered, and the identification results are weighted or suppressed.

[0072] After preprocessing the EEG and ECG signals, EEG features from the EEG signals and ECG features from the ECG signals are extracted separately. Then, the extracted EEG and ECG features are fused in a multimodal manner to obtain the fused feature data.

[0073] In one specific implementation, EEG characteristics include one or more of the following: power in each frequency band (δ, θ, α, β, γ), the ratio of power in each frequency band to total power, Hjorth parameters, and nonlinear indices. ECG characteristics include one or more of the following: time-domain and frequency-domain indices of heart rate variability (HRV).

[0074] The Herchos parameters involve three sub-parameters as time-domain complexity indicators: Activity, Mobility, and Complexity. Nonlinear indicators include one or more of sample entropy, approximate entropy, and spectral entropy, enabling the data to possess nonlinear expressive capabilities. Power P in each frequency band. band Based on intra-frame sampling points, it can be defined by the following formula (1):

[0075]

[0076] Where n represents the index of the nth sampling point in the current frame, N represents the total number of sampling points in the frame, and X band (n) represents the sample sequence of the corresponding frequency band after filtering.

[0077] Feature extraction of ECG signals includes: detecting the QRS wavecomplex of left and right ventricular depolarization potentials and times; identifying the peak point of the R wave within the QRS complex and recording its precise time; subtracting the times of two adjacent R peaks to obtain a series of RR intervals; and arranging these intervals into a time series according to heartbeat sequence. Then, the time-domain and frequency-domain characteristics of HRV are calculated from this time series.

[0078] Among them, the time-domain features of the electrocardiogram signal include: standard deviation SDNN and root mean square error RMSSD, defined as follows (2) and (3):

[0079]

[0080] Where N represents the total number of RR intervals involved in the calculation, NN i Let NN represent the i-th RR interval. mean Represents the arithmetic mean of N RR intervals, RR i RR represents the interval between heartbeats of the i-th heartbeat. i+1 This represents the interval between heartbeats of the (i+1)th heartbeat.

[0081] The frequency domain characteristics of an electrocardiogram signal can be represented by the energy ratio between low-frequency (LF) and high-frequency (HF) signals, and can be calculated using the Welch mean modified periodogram method or an autoregressive method.

[0082] In addition, during feature fusion, attention mechanisms can be used to perform multimodal feature fusion, as defined in equations (4) and (5):

[0083]

[0084] F fusion =∑ i a i ·F i (5)

[0085] Where i represents the EEG mode and ECG mode, a i u represents the dynamic weight assigned to each feature channel. i The feature importance score, u j F represents a set of importance scores corresponding to all modalities. i This represents the original features, namely EEG and ECG features, fused from two Fs. i The vector represents two modes, F fusion This represents the fused feature data.

[0086] Step 130: Analyze the fused feature data to obtain emotion test data;

[0087] Specifically, a deep learning model is used to analyze the fused feature data. The resulting current emotion test data for the subjects includes: a numerical emotion index, the maximum confidence score corresponding to the emotion category, and a short-term emotion trend. The emotion categories are positive, neutral, and negative. The numerical emotion index represents the emotion category; for example, negative emotion is represented by -1, neutral emotion by 0, and positive emotion by +1. The maximum confidence score corresponding to the emotion category can be understood as the model's maximum certainty regarding which emotion the sample is most likely to have. The short-term emotion trend represents the trend of emotion change.

[0088] In one specific implementation, the deep learning model in this application is a convolutional neural network (CNN) model. The model training can be based on publicly available emotion datasets, such as the DEAP (Database for Emotion Analysis using Physiological Signals) dataset and the SEED (SJTU Emotion EEG Dataset) dataset. Using this model to analyze the fused feature data, the numerical emotion index y(t) and the maximum confidence s corresponding to the emotion category can be obtained. Then, based on the numerical emotion index y(t), the short-term emotion trend T(t) can be obtained, and its calculation formula is shown in equation (6):

[0089] T(t)=β·T(t―1)+(1―β)·y(t) (6)

[0090] Where T(t) represents the short-term sentiment trend of the current frame t, T(t-1) represents the short-term sentiment trend of the previous frame, β represents the smoothing coefficient, 0<β<1, which can be set as needed, and y(t) represents the numerical sentiment index of the current frame t.

[0091] Step 140: Generate intervention control parameters based on the emotion test data;

[0092] Specifically, under the rule-based mapping, based on the subject's current numerical emotion index, the maximum confidence level corresponding to the emotion category, the short-term emotion trend, and the user preference profile, the calculation formula for the intervention control parameters is generated as shown in formula (7):

[0093] M(t)=G(y(t),s(t),T(t),U) (7)

[0094] Where t represents the current frame, M represents the intervention control parameters, G represents the regularization mapping, U represents the user preference profile, y(t) represents the numerical sentiment index of the previous frame, s(t) represents the maximum confidence level corresponding to the sentiment category of the previous frame, and T(t) represents the short-term sentiment trend of the current frame.

[0095] In one specific implementation, the emotion regulation in this application is regulated using music and voice. The user preference profile includes personalized music parameters, historical responses, and other data. The obtained pre-control parameters include a music parameter vector M: M = [tempo, pitch, harmonic, reverb, volume].

[0096] Regarding the specific scheme for rule-based mapping, in the first implementation, if the emotion category represented by the numerical emotion index is negative, and its corresponding maximum confidence value is greater than a first threshold, it indicates that the subject's current emotion is likely negative, requiring soothing intervention. This soothing intervention may include: lowering the music tempo, enhancing the warm timbre, and deepening the music reverberation effect to create a soothing atmosphere. If the emotion category represented by the numerical emotion index is positive, then while maintaining the music tempo, timbre, and reverberation effect, a volume increase intervention is implemented within a preset range, maintaining motivation by only slightly increasing the volume.

[0097] The first threshold can be set as needed. Taking frequency f as an example, the fine-tuning parameters for adjusting audio can be set using the following formula (8):

[0098] f′(t)=f0·(1+α·(s ref ―s(t))) (8)

[0099] Where f′(t) represents the adjustment frequency, f0 represents the reference frequency, and s ref Let s(t) represent the target (required) confidence level, s(t) represent the maximum confidence level corresponding to the sentiment category of the current frame, and α represent the moderating coefficient. α and s ref The value can be set as needed.

[0100] Similarly, parameters such as rhythm and harmonic proportions can be adjusted using the same logic.

[0101] In addition, adjustments can be made based on digital signal processing (DSP) algorithms, and existing open-source tools or self-developed modules can be used to perform time-based audio stretching, pitch shifting, digital equalization, and reverberation.

[0102] To increase the stability of the intervention, the second implementation scheme further includes the following: If the current numerical emotion index represents a negative or positive emotion category, and its corresponding maximum confidence value remains above the second threshold for a given period of time, it indicates that the subject is likely in a negative or positive emotional state and this state has lasted for a relatively long time, thus requiring intervention. The intervention method must correspond to the positive or negative category. Conversely, if the current numerical emotion index represents a negative or positive emotion category, and its corresponding maximum confidence value remains above the third threshold for a given period of time, it indicates that the subject is not likely in a negative or positive emotional state, but this state has lasted for a period of time, thus requiring intervention to be reduced or stopped. The first frame count, second frame count, second threshold, and third threshold can be determined by the user or through individualized calibration during initialization. It is understood that the second threshold must be greater than the third threshold to create hysteresis and avoid frequent switching of intervention modes.

[0103] In addition, for neutral emotion categories, although they are not included in the above-mentioned judgment logic for initiating, weakening, or stopping intervention, targeted interventions can be initiated in the following scenarios: when the emotion category turns neutral after intervention for negative or positive emotions, and its maximum confidence level remains at or above the second threshold for more than two consecutive periods, it is necessary to consolidate the intervention effect and maintain emotional stability, so mild intervention can be continued; if the subject's previous emotional fluctuations were more frequent in the neutral emotion state, mild intervention can be initiated as a buffer to help stabilize the emotional baseline and reduce the possibility of repeated emotional fluctuations.

[0104] To better achieve adaptive adjustment, in the third implementation, when analyzing the fused feature data using a deep learning model, it is also necessary to obtain the emotion improvement rate E(t) based on the short-term emotion trend before intervention and the subject's current short-term emotion trend to evaluate the intervention effect. The calculation formula is as shown in formula (9):

[0105]

[0106] Wherein, T(t) pre T(t) represents the short-term sentiment trend before intervention. post This represents the current short-term emotional trend of the subject.

[0107] The rule-based mapping scheme also includes: if the emotion improvement rate is less than the predetermined improvement threshold, indicating that the current intervention is not effective, then the music style will be changed, the intervention intensity will be increased, or the intervention mode will be switched to voice accompaniment. If the emotion improvement rate is not less than the predetermined improvement threshold, then the intervention intensity will be maintained or slightly reduced. Increasing or decreasing the intervention intensity includes: increasing or decreasing the volume, increasing or decreasing the tempo adjustment value, and increasing or decreasing the reverberation effect, etc.

[0108] It is understandable that the three implementations of the above-mentioned rule-based mapping scheme are compatible with each other and can be applied simultaneously. Furthermore, when implementing the above three schemes, user preference profiles can be considered simultaneously. For example, one or more music tracks can be selected from the user's preferred music tracks for intervention, or one or more music tracks can be selected from the historical response music tracks for intervention.

[0109] Step 150: Output an emotional intervention wave to the subject based on the intervention control parameters to regulate the subject's current emotion;

[0110] Specifically, as mentioned above, the emotional intervention wave in this application includes audio such as music and voice, which allows the person receiving the emotional support to improve their mood by listening to music or voice.

[0111] In addition, this application can also output visual signals when conducting emotional interventions on subjects. For example, it can use LED light strips to output colors and flashing patterns set according to emotional states, and use screens to provide personalized settings interfaces. Among them, the personalized settings interface can provide data interaction, such as displaying sampling and analysis results, displaying emotional history curves, setting privacy preferences, selecting music styles, etc., and can also support users to annotate intervention records to assist in personalized training of the model.

[0112] The emotion measurement and control method provided in this application acquires the subject's electroencephalogram (EEG) and electrocardiogram (ECG) signals in real time. After analyzing the EEG and ECG signals, emotion test data is obtained, and intervention control parameters are generated based on the emotion test data to regulate the subject's current emotion. Through a multimodal feature fusion algorithm of EEG and ECG signals, the method achieves accurate identification of emotional state. Based on the identification results, it triggers emotion intervention and dynamically adjusts the intervention content, forming an integrated closed-loop scheme of emotion measurement and emotion intervention to promote people's physical and mental health development.

[0113] On the other hand, this application also provides a wearable device for implementing the emotion measurement and control method described in the first aspect above, such as... Figures 2-5As shown, the device includes a data acquisition unit, a processor, and an intervention unit. The data acquisition unit collects the subject's electroencephalogram (EEG) and electrocardiogram (ECG) signals in real time. The processor acquires these signals in real time, extracts and fuses features, and then analyzes the fused feature data to obtain emotion test data. Finally, it generates intervention control parameters based on the emotion test data. The intervention unit outputs an emotion intervention wave based on the intervention control parameters to regulate the subject's current emotion.

[0114] To enhance portability, this wearable device is designed as an earphone-style device, comprising a headband and earphones.

[0115] The headgear includes an elastic clamping mechanism to keep it in close contact with the subject's head.

[0116] The acquisition devices include: an electroencephalogram (EEG) signal acquisition device and an electrocardiogram (ECG) signal acquisition device.

[0117] The EEG signal acquisition device is placed inside the headpiece to collect EEG signals from the subject's forehead region. Preferably, the EEG signal acquisition device uses two acquisition channels, Fp1 and Fp2, to ensure effective capture of emotion-related EEG activity in the forehead. Furthermore, the electrodes used in the EEG signal acquisition device are flexible, adhesive electrodes, preferably made of conductive silicone, medical-grade silver coating, or silver chloride coating. An elastic clamping mechanism ensures stable contact and comfort for subjects with different head shapes in various scenarios. Simultaneously, to improve wearing stability and signal-to-noise ratio, conductive pads are used to cover the electrode positions, and a low-impedance coupling layer is pre-placed between the electrodes and the skin. The electrode layout and mechanical connection structure are designed in conjunction with the device's wide handle and downward curvature to ensure reliable contact between the forehead acquisition unit and the forehead skin without displacement during wear.

[0118] At the circuit level, the EEG signal acquisition unit can incorporate a high-input-impedance preamplifier and common-mode rejection design, employing differential amplification and multi-stage filtering hardware channels to acquire microvolt (μV) EEG signals. Preferably, an instrumentation amplifier or a custom preamplifier chip with an input bias current below 1 nA and an input impedance greater than 1 GΩ is used to reduce signal distortion caused by contact impedance variations. The sampling module can employ a 16-bit or 24-bit analog-to-digital converter (ADC) with a sampling rate in the range of 250Hz–1000Hz, preferably 500Hz, to balance bandwidth coverage and power consumption. To prevent 50Hz or 60Hz interference, the front-end design includes hardware notch filtering or digital compensation preprocessing, and multiple conditioning circuits are reserved at the hardware level, including high-pass, low-pass, notch filtering, and configurable gain circuits for parameter adjustment in different scenarios or experimental requirements.

[0119] The ECG signal acquisition device is placed inside the headband, either at the bottom of the device or inside the wide handle, to acquire ECG signals from the temporal region of the subject. Alternatively, the ECG signal acquisition device can be an external chest patch ECG acquisition device.

[0120] The ECG acquisition circuit uses a differential preamplifier with millivolt (mV) sampling, combined with a high-precision ADC (16-bit / 24-bit) and anti-saturation design to adapt to changes in human movement and contact.

[0121] The processor is preferably a low-power embedded processor, such as an ARM single-chip CPU running a Linux kernel, housed within the headset to support local real-time processing and edge inference. The processor can integrate memory, external storage interfaces, a WiFi chip, and a Bluetooth module, and work in conjunction with hardware peripherals such as multi-channel conditioning circuitry, audio decoding and amplification circuitry, and LED light strips. The processor may also include power management functions, hardware security mechanisms, and expansion interfaces, such as controls for the rechargeable lithium battery, a charging management IC, low-power modes, hardware encryption modules, firmware upgrades, or external sensors. The processor's circuit board size and layout prioritize miniaturization and optimized thermal design to ensure wearing comfort and safety.

[0122] At the signal processing level, the processor is responsible for reading the data stream from the ADC module and performing real-time preprocessing, including filtering, artifact detection, segment marking, etc., while dividing the sliding window, for example, with a window length of 2s and a step size of 1s, to perform frame-level feature extraction.

[0123] The processor simultaneously implements feature normalization, missing value imputation, and statistically significant feature selection logic to reduce computational load and improve model robustness. To support offline operation, the processor can deploy lightweight deep learning models and can optionally perform more complex computations or model updates in the cloud.

[0124] In addition, the processor also handles device hardware coordination tasks, including: driving the audio decoding chip and power amplifier circuit, controlling the LED light strip to display emotional states, managing secure communication channels for Bluetooth and wireless networks, performing encrypted data storage, and implementing low-power strategies, such as downsampling the sampling rate or entering sleep mode during non-critical periods. To ensure data privacy, the processor supports end-to-end encryption and provides two operating strategies: "local processing without uploading" and "uploading with user authorization for model optimization."

[0125] During emotional intervention, the processor also includes an audio decoding chip, a digital-to-analog converter (DAC), a power amplifier, and a speaker. The speaker is positioned facing the ear canal, prioritizing the frequency response range and sound pressure level required for music therapy, and its structure, with its downward-curving arc and wide handle, ensures clear and comfortable sound transmission when the device fits snugly in the ear canal. The audio decoder supports multiple audio formats and can load local or cloud music libraries.

[0126] The intervention device is an earphone used to output music or speech to the subject in order to achieve the intervention output.

[0127] In addition, to improve intervention effectiveness, the device can be equipped with LED light strips to provide visual emotional feedback and enhance immersion. The device can also be equipped with a mobile app for user settings, data visualization and historical record retrieval, and a cloud server for model updates and long-term data analysis. The LED light strips map preset colors and flashing patterns according to emotional states to provide users with immediate visual cues. The mobile app provides a personalized settings interface, including sampling and recognition parameters, privacy preferences, music style selection, calibration procedures, and display of emotional history curves. It also supports user annotation of intervention records to assist in personalized model training.

[0128] In terms of communication and cloud architecture, the device supports user-authorized data uploads. Uploaded data should use encryption protocols such as TLS during transmission and undergo de-identification processing in the cloud. The cloud can provide more complex model training, cross-user statistical analysis, and music track recommendation services, and push model weight updates or music library updates to the local device. To meet privacy requirements, the device also provides a local-first operation mode, enabling all key emotion recognition and intervention logic to be completed locally, with data uploaded only when explicitly permitted by the user.

[0129] To improve recognition accuracy and intervention effectiveness, this device supports an initial calibration process that includes: recording the individual's baseline EEG and ECG signals through several minutes of guided stimulation and self-assessment labels or standardized tasks, and then fine-tuning the data to determine individualized thresholds.

[0130] Below, in a specific implementation, the process of the above method operating in a closed loop within a wearable device is described:

[0131] When the user wears the device and enters working mode, the EEG signal collector and ECG signal collector collect data synchronously. The processor preprocesses the raw data and extracts features by sliding window. Based on the fused features, it classifies and outputs data such as category, confidence level and short-term trend.

[0132] If the recognition result meets the trigger condition (e.g., the maximum confidence score of negative emotion s ≥ the second threshold T), start And continuously (immediately), the emotion intervention module generates audio parameters according to the current user profile and the rule-based mapping G(·) and plays them immediately. During the intervention process, the system continuously collects and evaluates the intervention effect. If the emotion improvement rate E(t) < the predetermined improvement threshold E th If the pre-control parameter M(t) is adjusted iteratively, otherwise the intervention intensity is gradually reduced or maintained.

[0133] When emotions return to a safe or target range (e.g., the maximum confidence level of negative emotions s ≤ the third threshold T), stop (And continuing for a second time), the system enters observation or sleep mode, ending the current intervention sequence and saving the session record. The entire process employs anti-shake, smoothing, and redundancy judgment mechanisms to avoid false triggers and frequent switching, ensuring the continuity and comfort of the user experience.

[0134] The wearable device provided in this application offers a non-invasive, portable solution. Other preferred embodiments and technical effects of this wearable device are the same as the emotion measurement and control method described in the first aspect above, and will not be repeated here.

[0135] The preferred embodiments of this application have been described in detail above. However, this application is not limited to the specific details of the above embodiments. Within the scope of the technical concept of this application, various simple modifications can be made to the technical solution of this application, and these simple modifications all fall within the protection scope of this application.

[0136] It should also be noted that the various specific technical features described in the above embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, this application will not describe the various possible combinations separately.

[0137] Furthermore, various different embodiments of this application can be combined in any way, as long as they do not violate the spirit of this application, they should also be regarded as the content disclosed by this invention.

Claims

1. An emotion measurement and control method, characterized in that, The method includes: Real-time acquisition of subjects' electroencephalogram (EEG) and electrocardiogram (ECG) signals; Feature extraction and feature fusion are performed on the electroencephalogram (EEG) and electrocardiogram (ECG) signals to obtain fused feature data. The fused feature data is analyzed to obtain emotion test data; Based on emotion test data, generate intervention control parameters; Based on the intervention control parameters, an emotional intervention wave is output to the subject to regulate the subject's current emotion.

2. The method according to claim 1, characterized in that, Before performing feature extraction and feature fusion on the electroencephalogram (EEG) and electrocardiogram (ECG) signals, the method further includes: One or more steps are performed on the electroencephalogram (EEG) signal, including bandpass filtering, power frequency notch filtering, and independent component analysis. One or more steps are performed on the electrocardiogram signal, including dynamic quality monitoring, bandpass filtering, and baseline drift correction.

3. The method according to claim 1, characterized in that, Feature extraction and feature fusion are performed on the electroencephalogram (EEG) and electrocardiogram (ECG) signals to obtain fused feature data, including: Extracting EEG features from the EEG signal; the EEG features include one or more of the following: power of each frequency band, the ratio of power of each frequency band to the total power, Herchos parameters, and nonlinear indices; Extract electrocardiogram (ECG) features from the ECG signal; the ECG features include one or more of the time-domain and frequency-domain indices of heart rate variability. The extracted EEG and ECG features are fused using an attention mechanism to obtain fused feature data.

4. The method according to claim 1, characterized in that, The fused feature data is parsed to obtain the subject's current emotion test data, including: The fused feature data is analyzed using a deep learning model to obtain the subject's current emotional numerical index, the maximum confidence level corresponding to the emotional category, and the short-term emotional trend.

5. The method according to claim 4, characterized in that, Based on the subjects' current emotional test data, intervention control parameters are generated, including: Under the rule-based mapping, intervention control parameters are generated based on the subject's current numerical emotion indicators, the maximum confidence level corresponding to the emotion category, the short-term emotion trend, and the user preference profile. The calculation formula is as follows: M(t) = G(y(t),s(t),T(t),U); Where t represents the current frame, M represents the intervention control parameters, G represents the regularization mapping, U represents the user preference profile, y(t) represents the numerical sentiment index of the previous frame, s(t) represents the maximum confidence level corresponding to the sentiment category of the current frame, and T(t) represents the short-term sentiment trend of the current frame.

6. The method according to claim 5, characterized in that, The formula for calculating the short-term sentiment trend is: T(t)=β·T(t-1)+(1-β)·y(t); Where T(t) represents the short-term sentiment trend of the current frame t, T(t) represents the short-term sentiment trend of the previous frame, β represents the smoothing coefficient, and y(t) represents the numerical sentiment index of the current frame t.

7. The method according to claim 5, characterized in that, The regularized mapping includes: If the emotion category is negative and the corresponding maximum confidence value is greater than the first threshold, then a soothing intervention is performed. If the emotion category is positive, then increase the volume as an intervention.

8. The method according to claim 7, characterized in that, The regularized mapping also includes: If the current emotion category is negative or positive, and the corresponding maximum confidence value is consistently above or not less than the second threshold for the first time, then it is determined that intervention needs to be initiated. If the current emotion category is negative or positive, and the corresponding maximum confidence value remains no greater than the third threshold for a second period of time, then the intervention is determined to be reduced or stopped; the second threshold is greater than the third threshold.

9. The method according to claim 4, characterized in that: The fused feature data is parsed to obtain the subject's current emotion test data, and the analysis also includes: The fused feature data is analyzed using a deep learning model to obtain the subject's current mood improvement rate; The regularized mapping also includes: If the mood improvement rate is less than the predetermined improvement threshold, the music style will be changed, the intervention intensity will be increased, or the intervention mode will be switched to voice accompaniment. If the mood improvement rate is not less than the predetermined improvement threshold, the intervention intensity is maintained or reduced.

10. A wearable device, characterized in that, The wearable device includes: The data acquisition device is used to collect the subject's electroencephalogram (EEG) and electrocardiogram (ECG) signals in real time. The processor is used to acquire the subject's electroencephalogram (EEG) and electrocardiogram (ECG) signals in real time. Feature extraction and feature fusion are performed on the electroencephalogram (EEG) and electrocardiogram (ECG) signals to obtain fused feature data. The fused feature data is analyzed to obtain emotion test data; Based on emotion test data, generate intervention control parameters; An intervention device is used to output an emotional intervention wave to the subject according to the intervention control parameters, so as to regulate the subject's current emotion.

11. The device according to claim 10, characterized in that, The wearable device is an earphone-type device, including: a headband and earphones; The headgear includes an elastic clamping mechanism to fit the subject's head.

12. The device according to claim 11, characterized in that, The acquisition device includes: an electroencephalogram (EEG) signal acquisition device; The EEG signal acquisition device is placed inside the headband and collects EEG signals from the subject's forehead region.

13. The device according to claim 11, characterized in that, The data acquisition device also includes: an electrocardiogram (ECG) signal acquisition device; The electrocardiogram (ECG) signal acquisition device is placed inside the headband and collects ECG signals from the temporal region of the subject. Alternatively, the ECG signal acquisition device is an external chest patch ECG acquisition device.

14. The device according to claim 11, characterized in that, The processor is placed inside the headpiece.

15. The device according to claim 11, characterized in that, The intervention device is an earphone.

Images