Pressure sensitivity classification evaluation method based on heart rate variability and wearable device

Abstract

The invention discloses a pressure sensitivity classification evaluation method based on heart rate variability and wearable equipment, and relates to the technical field of biomedical signal processing. According to the method, the collected signals can be processed to form the RR interval sequence and the resultant acceleration, the motion state is judged, the motion state mark is set, the RR interval sequence and the motion state mark are aligned according to the timestamp, the structured data are generated, the pressure sensitivity level of an individual is judged based on the structured data, and the pressure sensitivity level of the individual is calculated. The subjective influence of pressure sensitivity evaluation can be avoided, and the classification precision and adaptability are improved.

Description

Technical Field

[0001] The present application relates to the field of biomedical signal processing technology, and in particular to a stress sensitivity classification and assessment method based on heart rate variability and a wearable device. Background Art

[0002] Currently, stress sensitivity assessment has problems such as strong subjectivity, low classification accuracy, and dynamic adaptability, which are mainly manifested in: the current mainstream stress assessment method mainly relies on standardized psychological scales, which are subject to the influence of cognitive bias of the form fillers, different context sensitivity and poor cross-cultural applicability; the disconnection between physiological signals and subjective reports has the defects of subjective signal interpretation, delayed dynamic response, and fragmentation of multimodal data; there are verification difficulties in clinical practice. Although the popularity of wearable devices provides continuous physiological data, the existing system has not been able to break through the bottleneck of subjective assessment. Summary of the Invention

[0003] The embodiments of the present application provide a pressure sensitivity classification and assessment method based on heart rate variability and a wearable device, which are used to solve the technical problems in the prior art of pressure sensitivity assessment, such as high subjectivity, low classification accuracy, and poor dynamic adaptability.

[0004] The present embodiment provides a stress sensitivity classification and assessment method based on heart rate variability, including: collecting a photoplethysmography signal, and processing the photoplethysmography signal to form an R_R interval sequence; Collect acceleration signals in three directions, calculate the vector sum of the accelerations in the three directions, obtain the total acceleration, determine the motion state and set the motion state flag; Aligning the RR interval sequence with the motion state marker according to timestamps to generate structured data; extracting a heart rate variability (HRV) feature vector from the R_R interval sequence in the structured data after preprocessing, and extracting an acceleration feature vector from the combined acceleration in the structured data; The individual's stress sensitivity level is determined based on the heart rate variability (HRV) feature vector and the acceleration feature vector.

[0005] Furthermore, the collecting of the photoplethysmography signal and processing the photoplethysmography signal to form an R-R interval sequence includes: collecting blood volume change signals through a photoplethysmography sensor module, and digitally processing the blood volume change signals to form time-series photoplethysmography signals; Smoothing the photoplethysmography signal to remove high-frequency noise and form a heartbeat signal; By analyzing the first-order derivative changes of the heartbeat signal, the peak points representing the heartbeat events are identified, and the time intervals between adjacent peak points are calculated to obtain the R_R interval sequence.

[0006] Furthermore, the collecting acceleration signals in three directions, calculating the vector sum of the accelerations in the three directions to obtain the total acceleration, determining the motion state and setting the motion state flag includes: Detect the three-axis acceleration value through the acceleration sensor module ; pass Calculate the resultant acceleration; pass Get the acceleration amplitude change rate of each time step, where steps_{t-1} is the total acceleration corresponding to time step t-1, and steps_t is the total acceleration corresponding to time step t; The motion state is determined according to the acceleration amplitude change rate and a motion state flag is set.

[0007] Furthermore, judging the motion state according to the acceleration amplitude change rate and setting the motion state flag includes: When the rate of change of the combined acceleration amplitude is greater than or equal to a first threshold, it is determined that the current motion state is a running state, and the motion state flag is set to a running state flag; When the rate of change of the combined acceleration amplitude is between a first threshold and a second threshold, it is determined that the current motion state is a walking state, and the motion state flag is set to a walking state flag; When the rate of change of the combined acceleration amplitude is less than or equal to a second threshold, it is determined that the current motion state is a resting state, and the motion state flag is set to a resting state flag.

[0008] Furthermore, the extracting of a heart rate variability (HRV) feature vector after preprocessing the R_R interval sequence in the structured data includes: performing sliding window filtering and outlier detection processing on the R_R interval sequence to form clean data, extracting a heart rate variability (HRV) feature vector from the clean data, and classifying the heart rate variability (HRV) feature vector into time domain features, frequency domain features, and nonlinear features; The time domain features include the standard deviation of the R_R interval, the root mean square of the difference between adjacent R_R intervals, the proportion of R_R intervals with a difference greater than 50 ms, the area-to-height ratio when the R_R interval histogram is fitted into a triangle, and the relationship between short-term and long-term variability is analyzed through a scatter plot of the R_R intervals; Setting the frequency domain features to include total power and LF / HF ratio, wherein the LF / HF ratio is a ratio of low frequency power LF to high frequency power HF; The nonlinear features are set to include sample entropy and multi-scale entropy.

[0009] Furthermore, performing sliding window filtering and outlier detection processing on the RR interval sequence to form clean data includes: Setting a sliding window to adjust the size of the sliding window according to the motion state mark, setting the sliding window to a first duration window when the motion state mark is a resting state, setting the sliding window to a second duration window when the motion state mark is a walking state, and setting the sliding window to a third duration window when the motion state mark is a running state, wherein the first duration is greater than the second duration, and the second duration is greater than the third duration; Performing sliding window filtering on the R_R interval sequence to remove motion artifacts; Obtaining a valid range of an R_R interval of physiological constraints, and marking a value of the R_R interval sequence as an abnormal point when the value exceeds the valid range of the R_R interval; When the movement state is marked as a resting state, when three consecutive points deviate from the mean by more than three times the standard deviation, they are marked as abnormal segments; When the motion state is marked as a walking state or a running state, when three consecutive points deviate from the mean by more than five times the standard deviation, they are marked as abnormal segments; The values ​​of the abnormal points are deleted and the missing values ​​are filled by linear interpolation; Manually review the abnormal segment and replace the value of the abnormal segment with the mean value; The photoplethysmography signal after smoothing and removing high-frequency noise is standardized or normalized to form clean data.

[0010] Furthermore, the determining of an individual's stress sensitivity level based on the heart rate variability (HRV) feature vector and the acceleration feature vector includes: A multi-level classification architecture is used to construct a classification model, wherein the classification model includes a first-level classification and a second-level classification. The first-level classification uses a logistic regression algorithm to distinguish between normal state and stress state. The second-level classification uses an XGBoost algorithm to assess an individual's stress sensitivity level, wherein the stress sensitivity level includes low sensitivity, medium sensitivity, and high sensitivity. Acquire training data, set normal state labels, low sensitivity labels, medium sensitivity labels, and high sensitivity labels for the training data, and perform important feature screening on the training data; Using the training data to train the classification model, and dynamically adjusting the weights of the important features based on SHAP values; The trained classification model is used to determine the stress sensitivity level of an individual according to the heart rate variability (HRV) feature vector and the acceleration feature vector, and stress assessment results corresponding to the first level classification and the second level classification are output.

[0011] Furthermore, the dynamically adjusting the weights of the important features based on the SHAP values ​​includes: Collecting an EEG signal and an ECG signal, processing the EEG signal and the ECG signal to obtain a speech feature vector, processing the heart rate variability (HRV) feature vector to obtain a photoplethysmography feature vector, and using the photoplethysmography feature vector, the acceleration feature vector, and the speech feature vector as important features; A weighted fusion model is constructed based on the important features. The weighted fusion model is: , where w1 is the weight of the photoplethysmography feature vector, F_ppg is the photoplethysmography feature vector, w2 is the weight of the acceleration feature vector, F_acc is the acceleration feature vector, w3 is the weight of the voice feature vector, and F_voice is the voice feature vector; When the performance of the classification model drops by more than a preset ratio, the SHAP value of all training data is calculated using the current classification model; the importance score of each feature is recalculated based on the SHAP value; and the weight of the important features is adjusted during the training of the current classification model.

[0012] Furthermore, the determining of an individual's stress sensitivity level based on the heart rate variability (HRV) feature vector and the acceleration feature vector using the trained classification model includes: Obtaining a probability P1 that the output individual has stress according to the heart rate variability (HRV) feature vector; Obtaining a probability P2 that the output individual has stress according to the acceleration feature vector; The total probability P of the individual experiencing stress is calculated and outputted by P=0.7P1+0.3P2, and the individual's stress sensitivity level is determined based on the total probability P.

[0013] Furthermore, the stress sensitivity classification and assessment method based on heart rate variability also includes: Maintain a personalized weight vector for each user and dynamically adjust the personalized weight based on the user's latest stress assessment results; The formula for adjusting the personalization weight is set as ; where w_i is the weight of the i-th feature, w_i_base is the base weight, a is the adjustment coefficient, 0.1≤a≤0.3, performance_j is the accuracy of the latest stress assessment result of user j, and baseline is the baseline accuracy.

[0014] The present application also provides a wearable device for implementing the aforementioned stress sensitivity classification and assessment method based on heart rate variability, the wearable device comprising: Microprocessor (MCU); a pulse count (PPG) sensor connected to the microprocessor (MCU) via an I2C / SPI interface for collecting photoplethysmography signals; An acceleration sensor connected to the microprocessor (MCU) via an I2C interface, configured to detect three-dimensional motion state data of the device and process the three-dimensional motion state data to obtain acceleration signals in three directions; An electroencephalogram (EEG) module is connected to the microprocessor (MCU) via an SPI / UART interface for collecting EEG signals; an electrocardiogram (ECG) module, connected to the microprocessor (MCU) via a UART interface, for collecting ECG signals; A wireless transmission module (Bluetooth), connected to the microprocessor (MCU) via a UART interface, for transmitting data to external electronic devices; Wherein, the microprocessor (MCU) includes: a pulse number (PPG) processing unit, connected to the pulse number (PPG) sensor, for processing the photoplethysmography signal into an R_R interval sequence; an acceleration processing unit connected to the acceleration sensor, for calculating the vector sum of the accelerations in the three directions according to the acceleration signals in the three directions, obtaining a resultant acceleration, determining a motion state, and setting a motion state flag; a data acquisition module connected to the pulse number (PPG) processing unit and the acceleration processing unit, aligning the RR interval sequence with the motion state mark according to timestamps to generate structured data; A storage module, connected to the data acquisition module, is used to store the structured data, extract a heart rate variability (HRV) feature vector after preprocessing the R_R interval sequence in the structured data, extract an acceleration feature vector from the combined acceleration in the structured data, and determine the individual's stress sensitivity level based on the heart rate variability (HRV) feature vector and the acceleration feature vector.

[0015] The heart rate variability-based pressure sensitivity classification and assessment method and wearable device provided in the embodiments of the present application can process the collected signals to form an R_R interval sequence and a combined acceleration, determine the motion state and set a motion state mark, align the R_R interval sequence with the motion state mark according to the timestamp, generate structured data, and determine the individual's pressure sensitivity level based on the structured data, thereby avoiding the subjective influence of pressure sensitivity assessment and improving classification accuracy and adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The following detailed description of the specific embodiments of the present application in conjunction with the accompanying drawings will make the technical solutions and other beneficial effects of the present application apparent.

[0017] Figure 1 Schematic diagram of an application scenario of the stress sensitivity classification and assessment method based on heart rate variability provided in an embodiment of the present application.

[0018] Figure 2 Schematic diagram of the pressure sensitivity classification and assessment method based on heart rate variability provided in an embodiment of the present application.

[0019] Figure 3 A schematic diagram of the data preprocessing process provided in an embodiment of the present application. DETAILED DESCRIPTION

[0020] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the embodiments described are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0021] As described in the background art, current pressure sensitivity assessment has the problems of strong subjectivity, low classification accuracy, and dynamic adaptability.

[0022] 1. Inherent flaws in subjective assessment tools The current mainstream stress assessment methods mainly rely on standardized psychological scales (such as the PSS scale and DASS-21). The core problems are: 1) Cognitive bias: Self-reports are susceptible to influences such as current emotional state, comprehension, and social desirability. For example, completing a questionnaire immediately after an acute stressful event may lead to inflated scores due to enhanced emotional memory, while those experiencing chronic stress may underestimate their scores due to adaptation or numbing reactions.

[0023] 2) Contextual sensitivity: Scale results are highly dependent on the testing context. Laboratory-based assessments can lead to biased responses due to behavioral changes caused by participants' awareness of being observed, while self-assessments at home, lacking professional guidance, can easily lead to random responses.

[0024] 3) Poor cross-cultural applicability: Some scale items have significant semantic differences across different cultural contexts. For example, in the Chinese context, the word "tension" can encompass both physical and psychological meanings, while the original English version only refers to the psychological dimension, resulting in inconsistent scoring standards.

[0025] 2. Disconnect between physiological signals and subjective reports Although existing studies have attempted to incorporate physiological indicators such as HRV (heart rate variability) to assist in assessment, key contradictions still exist: 1) Subjective signal interpretation: Clinicians' interpretation of the relationship between HRV parameters (such as the LF / HF ratio) and stress relies on personal experience and lacks objective quantitative standards. For example, the same LF / HF ratio may correspond to completely different stress response patterns in different individuals (e.g., an athlete vs. a patient with anxiety).

[0026] 2) Dynamic response lag: Traditional scales capture stress states only at specific points in time, whereas the autonomic nervous system's stress response is dynamic at the millisecond level. For example, a stress spike triggered by a sudden traffic incident may have subsided before the questionnaire was completed, leading to missed diagnosis.

[0027] 3) Multimodal data fragmentation: Existing systems fail to effectively integrate the temporal correlation between HRV and behavioral data (such as a sudden drop in step count and voice tremor frequency), and are unable to capture the complex physiological-behavioral response chain induced by stress.

[0028] 3. Verification Dilemma in Clinical Practice Empirical research shows that subjective assessments have significant limitations: 1) Validity controversy: Analysis shows that the AUC (area under the curve) of the PSS scale for predicting clinical stress-related disorders is only 0.65-0.72, far below the conventional standard of medical diagnostic tools (>0.85).

[0029] 2) Low test-retest reliability: The coefficient of variation of the results of the same scale filled out by the same subject on different days can reach 15-20%, reflecting the fragility of its state capture.

[0030] 3) Ignoring individual differences: Existing scales use group-standardized norms and fail to account for baseline differences. For example, someone with a naturally high resting HRV may be misclassified as having "low stress sensitivity" when, in reality, their stress response threshold is significantly higher than that of the average person.

[0031] 4. Unresolved issues in technological evolution Despite the widespread availability of wearable devices providing continuous physiological data, existing systems still fail to break through the bottleneck of subjective assessment: 1) Black box interpretation of data: The numerical results output by the HRV analysis algorithm lack clinical operability and need to rely on secondary interpretation by doctors, which essentially continues the chain of subjective judgment.

[0032] 2) Lack of dynamic baseline: The device only provides absolute numerical alarms (such as HRV < 30ms triggering an alarm), and does not establish an individualized stress response baseline model, resulting in a high false alarm rate.

[0033] 3) Inability to differentiate between stressors: Unable to distinguish the physiological patterns of work stress (sustained low HRV) and post-exercise recovery (transient low HRV), resulting in insufficient assessment specificity.

[0034] To address the above issues, the present invention provides a stress sensitivity classification and assessment method based on heart rate variability, which specifically includes the following technical issues and corresponding solutions: 1. Overcoming the limitations of subjective assessment Current issues: Existing stress assessments rely on subjective questionnaires (such as the PSS scale), which are easily affected by individual cognitive biases, emotional states, and environmental interference. They lack the support of objective physiological indicators, resulting in unreliable assessment results.

[0035] The validity AUC of commonly used clinical scales (such as DASS-21) is only 0.65-0.72 and the test-retest reliability (coefficient of variation 15-20%) is low, making it difficult to accurately reflect an individual's true stress sensitivity.

[0036] The solution to be adopted in this embodiment is: HRV physiological signals were used to replace subjective questionnaires, and an objective stress sensitivity assessment index was established by analyzing the time domain, frequency domain and nonlinear characteristics of heart rate variability (HRV).

[0037] Combined with machine learning algorithms, it integrates multi-dimensional physiological data to reduce interference from human factors and improve the objectivity and consistency of the assessment.

[0038] 2. Improve the classification accuracy of HRV analysis Current issues: Traditional HRV research simplifies stress sensitivity into a binary state, namely, high / low HRV, ignoring the continuous spectrum characteristics of people's response to stress and failing to achieve fine grading.

[0039] The existing algorithms have not established a quantitative mapping relationship between HRV parameters and pressure sensitivity levels, resulting in rough classification results and limited clinical guidance value.

[0040] The solution to be adopted in this embodiment is: Build a multi-level classification model and perform fine classification based on HRV features (SDNN, LF / HF, sample entropy, etc.) to improve the accuracy of the assessment.

[0041] Feature engineering optimization is introduced to screen key features through methods such as mutual information and recursive feature elimination (RFE) to enhance the model's discriminative ability.

[0042] 3. Enhance dynamic adaptability Current issues: Existing systems mostly use static threshold judgments and cannot adapt to the dynamic changes of individuals in different physiological states, such as after exercise, during sleep, or under environmental stressors, such as work pressure and social pressure.

[0043] It lacks the ability to adaptively learn user-personalized baselines, resulting in a high false alarm rate, which affects the actual application effect.

[0044] The solution to be adopted in this embodiment is: A dynamic calibration mechanism is designed to establish a personalized baseline model based on user historical data, and an exponentially weighted moving average (EWMA) is used for online updates to adapt to individual differences.

[0045] An environmental stressor identification module is introduced, which combines metadata such as timestamps and geographic locations to distinguish different types of stressors, such as work stress and exercise stress, to improve the adaptability of the model.

[0046] 4. Realize collaborative analysis of multimodal data Current issues: Existing technologies analyze HRV data in isolation without integrating behavioral data such as step count, voice characteristics, or environmental information, making it difficult to capture the complex physiological-behavioral response chain induced by stress.

[0047] Insufficient data standardization across devices and indicators makes it difficult for home health monitoring systems to form a complete stress assessment portrait.

[0048] The solution to be adopted in this embodiment is: Build a multimodal data fusion framework to integrate multi-dimensional data such as HRV, motion sensors, and voice signals, and use the attention mechanism to enhance the weight of key features to improve the comprehensiveness of the evaluation.

[0049] Establish a unified data standardization process, and use methods such as sliding window standardization and Z-score normalization to eliminate dimensional differences between different devices and different indicators to ensure data comparability.

[0050] The effects of this application are as follows: 1. Address the limitations of subjective evaluation and enhance the objectivity of evaluation.

[0051] HRV physiological signals + machine learning are used to replace subjective questionnaires.

[0052] Eliminate subjective reporting bias, and the evaluation results are no longer affected by individual cognitive bias, emotional state or environmental interference.

[0053] Clinical validation has shown that HRV-based assessment results are more consistent with professional psychological assessments."

[0054] The coefficient of variation of the evaluation results of the same subject at different time points was reduced, significantly improving the stability across time.

[0055] 2. Improve classification accuracy and achieve fine grading.

[0056] Build a multi-level classification model + feature engineering optimization.

[0057] Test results on public datasets show that the overall accuracy of the three-level classification (low / medium / high sensitivity) is significantly improved compared to the traditional binary classification (high / low).

[0058] The recognition recall rate (Recall) of highly sensitive populations is improved, reducing the risk of missed diagnosis.

[0059] The correlation coefficients between the classification results and stress-related physiological indicators (such as cortisol levels) increased.

[0060] 3. Enhance dynamic adaptability and reduce false alarm rate.

[0061] Dynamic calibration mechanism + personalized baseline model.

[0062] After dynamic calibration, the system's false alarm rate is lower than traditional methods, and it performs particularly well in non-steady-state scenarios such as after exercise and during sleep.

[0063] The user-personalized baseline model improves cross-scenario applicability and adapts to people of different ages, genders, and health statuses.

[0064] 4. Realize multimodal data fusion to improve the comprehensiveness of assessment.

[0065] Collaborative analysis of HRV+behavioral data+environmental information.

[0066] After fusing motion sensors and voice signals, the detection rate of "hidden stress" (such as social anxiety) is improved compared to single HRV analysis.

[0067] Multimodal feature fusion improves the model AUC.

[0068] The sliding window normalization method improves the comparability of data collected by different devices (such as smart watches vs. medical-grade ECGs) and solves cross-platform compatibility issues.

[0069] 5. Social and economic benefits.

[0070] Early stress warning can reduce stress-induced cardiovascular events by more than 30% (estimated based on WHO stress-related disease burden data).

[0071] Improve the accessibility of mental health management and increase the coverage of home stress monitoring.

[0072] Reduce waste of medical resources, accurately classify and reduce unnecessary clinical examinations, and save medical expenses.

[0073] Commercialization potential: After the algorithm is integrated into wearable devices, user stickiness is expected to increase by 40%, and annual subscription service revenue growth potential will reach $2.3 billion (MarketsandMarkets industry report).

[0074] And feature screening, it removes a lot of irrelevant or redundant information, and avoids misjudgment caused by interference from these factors.

[0075] Example 1 The stress sensitivity classification and evaluation method based on heart rate variability in this embodiment is applicable to Figure 1 Wearable devices can connect to electronic devices such as smartphones to transmit data, and the wearable devices include a sensor layer, a main control layer, and a communication layer.

[0076] 1) Sensor layer: PPG sensor: Connected to the MCU via the I2C / SPI interface, responsible for collecting photoplethysmography signals (basic HRV data).

[0077] Accelerometer: Connected to the MCU via the I2C interface, it is used to detect the device's motion state and distinguish HRV data between resting and exercising states.

[0078] EEG (electroencephalogram) module: connected to the MCU via the SPI / UART interface to collect EEG signals and enhance the accuracy of stress assessment.

[0079] ECG (electrocardiogram) module: Connects to the MCU via the UART interface to provide more accurate ECG signals and verify HRV data.

[0080] 2) Main control layer MCU main control unit: core processor, responsible for: Data acquisition: obtain raw data from PPG, accelerometer, EEG module, and ECG module; Preprocessing: Perform sliding window filtering and outlier detection; Communication: Interact with the BLE module through the UART interface.

[0081] 3) Communication layer BLE 5.0 (Bluetooth) module: Connects to the MCU via the UART interface and is responsible for wirelessly transmitting processed HRV and motion data to the smartphone.

[0082] The smartphone receives data transmitted by the wearable device via BLE 5.0, runs a classification algorithm to assess pressure sensitivity, and provides a visual interface and early warning notifications.

[0083] like Figure 2 As shown, Figure 2 Provides a data collection process interaction mode for the core processor.

[0084] 1) Signal acquisition layer PPG sensor module: Function: Collect blood volume change signals through photoplethysmography technology; Output: digitized PPG signal (time series data).

[0085] Accelerometer module: Function: Detect the three-dimensional motion state of the equipment; Output: Three-axis acceleration value .

[0086] 2) Signal processing PPG data processing unit: Core features: Peak detection algorithm: identifies heartbeat events; R_R interval calculation: calculate the interval between adjacent heart beats; Noise filtering: remove motion artifacts; Output: Accurate sequence of R-R intervals. The R-R interval is the time interval between two consecutive R waves in an electrocardiogram, reflecting the heart's rhythm and frequency. The normal range is typically 0.6 to 1.2 seconds (corresponding to a heart rate of 50 to 100 beats per minute), but this should be determined based on age, physical condition, and measurement scenario.

[0087] Acceleration processing unit: Core features: Calculation of total acceleration: ; Motion state determination: Determine the motion state based on the threshold (1.5g); Output: motion status flag (True / False).

[0088] 3) Data Collection Layer The RR interval sequence and the motion state marker are aligned by timestamp to generate a structured data record (timestamp + RR interval + motion state).

[0089] Output: preprocessed data stream.

[0090] 4) Data storage and transmission layer Temporary storage of acquired data (flash memory).

[0091] BLE 5.0 communication module: Wirelessly transmit data to smartphone app.

[0092] like Figure 3 As shown, Figure 3 This is the architecture diagram of the data preprocessing system.

[0093] 1) Original R_R interval sequence Input: Raw heartbeat interval data (time series) collected from the PPG sensor; Characteristics: Contains noise, motion artifacts, and possible outliers.

[0094] 2) Sliding window filter module Function: Smooth the signal and remove the noise caused by motion interference; Method: Dynamically adjusted median filtering (window size changes with motion state); Output: Preliminary purified R_R interval sequence.

[0095] 3) Outlier Detection Module Purpose: Identify and address data points with physiological anomalies or measurement errors; Methods: Physiological constraint checks (e.g., R-R interval range 600-1200ms), statistical anomaly detection (3σ principle), and adaptive thresholding of movement states; Output: Labeled or corrected clean data.

[0096] 4) Clean data output Content: High-quality R_R interval series after filtering and outlier processing; Features: Can be directly used for subsequent HRV analysis or feature extraction.

[0097] 5) Feature extraction module Input: clean data; Function: Extract HRV features (such as SDNN, LF / HF, etc.) from the cleaned data; Output: Normalized feature vector for stress assessment.

[0098] The hardware composition and connection relationship are as follows: 1) Wearable HRV monitoring module PPG sensor: Type: Photoplethysmography (PPG) sensor Function: Obtain heart rate signals by detecting changes in capillary blood volume; parameter: Sampling rate: ≥250Hz (meeting the Nyquist sampling theorem for HRV analysis); Wavelength: Dual wavelength LED to improve signal quality; Resolution: 12-bit ADC (Analog-to-Digital Converter).

[0099] Accelerometer: Type: three-axis MEMS accelerometer; Function: Detect the device's motion status and distinguish HRV data between resting and exercising states.

[0100] EEG module: Type: Dry electrode EEG headband; Function: Collect EEG signals for more accurate stress assessment; parameter: Sampling rate: 256Hz; Number of channels: 8-16 channels; Resolution: 16-bit ADC.

[0101] ECG Module: Type: Single-lead ECG patch; Function: Provide more accurate ECG signals for verifying HRV data; parameter: Sampling rate: 500Hz Resolution: 12-bit ADC Main control unit (MCU) Type: low-power microcontroller; Function: Data acquisition: Get raw data from PPG and accelerometer; Preprocessing: Perform sliding window filtering and outlier detection; Communication: The processed data is transmitted to the mobile phone APP via BLE 5.0; parameter: Operating frequency: 300MHz Memory: 256KB RAM + 1MB Flash Power consumption: <10mW (supports long-term continuous monitoring) Wireless communication module Type: Bluetooth Low Energy 5.0 (BLE 5.0); Function: Data transmission: transmit HRV and exercise data to mobile phone APP in real time; Protocol: Support GATT (General Attribute Protocol) for standardized data exchange; parameter: Transmission rate: 2Mbps (maximum); Range: 10-20 meters indoors, 30-50 meters outdoors (depending on the environment); Latency: <200ms (meeting real-time monitoring requirements).

[0102] 2) Smartphone (user terminal) Function: Data reception: Receive data transmitted by wearable devices via BLE 5.0; Data processing: running classification algorithms for pressure sensitivity assessment; User interaction: Provides visual interface and early warning notifications.

[0103] Among them, the internal connection method of the wearable device is as follows.

[0104] PPG sensor → MCU: Interface: I2C or SPI (depending on the sensor model); Data flow: original PPG signal → MCU for pre-processing; Accelerometer → MCU: Interface: I2C; Data flow: Raw acceleration data → MCU for motion state detection; EEG module → MCU: Interface: SPI or UART; Data flow: raw EEG signal → wearable device MCU for pre-processing; ECG module → MCU: Interface: UART; Data flow: raw ECG signal → wearable device MCU for verification; MCU → BLE 5.0 module: Interface: UART (Universal Asynchronous Receiver / Transmitter); Data flow: processed HRV and motion data → wireless transmission via BLE module; Wearable devices connect to smartphones; BLE 5.0 module → smartphone: Protocol: GATT (Generic Attribute Protocol); Data flow: Wearable device → smartphone: HRV data, exercise data, device status; Smartphone → Wearable device: configuration parameters, firmware update instructions.

[0105] Traditional HRV monitoring devices typically consist of only a PPG sensor and a simple MCU, lacking an accelerometer or wireless communication module. They typically connect to a computer or dedicated device via USB, resulting in high data transmission latency (>500ms). Limitations include the inability to distinguish HRV data from motion disturbances and a lack of real-time data transmission capabilities.

[0106] The advantages of this embodiment include: It integrates PPG, accelerometer, EEG, and ECG modules for multimodal data acquisition. Low-latency wireless transmission: BLE 5.0 provides real-time data transmission of <200ms, supporting continuous monitoring and real-time early warning. Scalability: The modular design supports future expansion of EEG / ECG modules, and standardized interfaces (such as SPI, UART, and GATT) facilitate integration.

[0107] The process of a stress sensitivity classification and assessment method based on heart rate variability provided in an embodiment of the present application is as follows: Step 1: PPG signal acquisition and RR interval extraction.

[0108] Hardware setup: The PPG sensor operates continuously at a fixed sampling rate of 250 Hz to ensure that sufficient heartbeat waveform details can be captured. Each sampling point accurately records the timestamp and corresponding light intensity value.

[0109] RR interval extraction method: A peak detection technique based on waveform characteristics is used: first, the continuously collected light intensity signal is smoothed to remove high-frequency noise, then the significant peak points representing heartbeat events are identified by analyzing the first-order derivative changes of the signal, and finally the time intervals between adjacent peak points are calculated to obtain the RR interval sequence. This process can accurately reflect the rhythm of the heartbeat and provide basic data for subsequent HRV analysis.

[0110] Step 2: Mark the movement status.

[0111] Accelerometer data processing: The three-axis accelerometer continuously collects acceleration values ​​in three directions at a sampling rate of 100 Hz. The system calculates the vector sum of the accelerations in the three directions in real time to obtain the total acceleration value.

[0112] Motion status determination rules: Clear thresholds are set, such that when the combined acceleration exceeds 1.5g, the device is considered in motion. This ensures accurate motion status identification while effectively reducing false positives. The system continuously monitors acceleration changes and updates the motion status marker in real time.

[0113] Step 3: Data storage.

[0114] Temporary storage content: The system stores the collected RR interval sequence and the corresponding motion state mark in chronological order. Each data record contains an accurate timestamp, RR interval value and motion state mark.

[0115] Storage format features: It uses an easy-to-parse table format, with timestamps accurate to the millisecond level. Data records maintain a strict time correspondence to ensure the accuracy of subsequent analysis.

[0116] Storage media: Wearable devices are equipped with local flash memory for temporary storage of collected data. At the same time, the data is regularly transmitted to the mobile app and stored in the mobile phone database.

[0117] Table 1 Key parameter settings

[0118] Traditional HRV devices: They usually only focus on the collection of PPG signals and lack the monitoring and marking of motion status. The data storage format is relatively simple, which is not conducive to subsequent complex data analysis. The sampling rate is relatively low, which may affect the accuracy of HRV analysis.

[0119] Advantages of this embodiment: 1) Comprehensive data collection: It not only collects PPG signals but also records motion status information synchronously, providing a more comprehensive basis for subsequent data analysis.

[0120] 2) Accurate time synchronization: Millisecond-level timestamps ensure the precise correspondence between the R_R interval and motion state data, improving data quality.

[0121] 3) Flexible storage format: Data storage in tabular form facilitates data organization and analysis, and is conducive to the development and optimization of subsequent algorithms.

[0122] Among them, the actual collection example scenario: users perform daily activities, including walking and sitting.

[0123] Table 2 Data record example

[0124] Analysis and explanation: During the time period corresponding to the "exercise" state, the R_R interval fluctuated. This may be due to the impact of exercise on the heart rhythm. In the subsequent data processing, the data can be screened and processed accordingly based on the exercise state marker.

[0125] Among them, the optimization measures for data collection are as follows.

[0126] Dynamic sampling rate adjustment: The system can automatically adjust the sampling rate of the PPG sensor according to the motion status. When motion is detected, the sampling rate can be temporarily increased to 500Hz to more accurately capture heartbeat signals. In the resting state, it returns to the regular sampling rate of 250Hz to reduce power consumption.

[0127] Data compression: For data that is in a static state for a long time, the system will use a segmented storage method to store statistical information within a certain period of time (such as 5 minutes) to reduce the amount of data storage. For data in motion, since it is more important for analysis, the original sampling rate will be retained for storage.

[0128] Anomaly detection: The system monitors the RR interval in real time to determine whether there are any abnormal values. RR intervals that are significantly outside the normal range (such as less than 300ms or greater than 2000ms) will be marked. These marked data can be manually reviewed in the subsequent analysis process to ensure data accuracy.

[0129] Among them, the verification of data collection is as follows.

[0130] Sampling rate verification: By comparing data collected at different sampling rates (125Hz and 250Hz), and analyzing the changes in HRV indicators, experimental results show that the HRV indicator at a sampling rate of 250Hz is more stable and can more accurately reflect changes in heart rhythm.

[0131] Motion Interference Test: This test simulates different motion states (such as walking and running) to verify the accuracy of motion state labeling. Experimental results show that the accuracy of motion state detection exceeds 95%, and it can effectively distinguish between motion and resting states.

[0132] Time synchronization test: The timestamp deviation between the wearable device and the mobile phone app is measured. The experimental results show that the timestamp deviation between the two is less than 1ms, which can meet the accuracy requirements of data alignment.

[0133] The data preprocessing process is as follows: Figure 3 shown.

[0134] Step 11: Sliding window filtering (motion artifact removal) 1) Algorithm principle: Adopt the improved median filtering algorithm, dynamically adjust the window size, and the filtering formula is: ; Where t is the middle moment of the filtering calculation, 2w is the time step of the dynamic adjustment window, For filtering calculation, is the R_R interval sequence of the start time of the dynamic adjustment window, It is the R_R interval sequence for dynamically adjusting the end time of the window.

[0135] Window parameters: Base window size: 30 seconds (covering a typical HRV analysis period).

[0136] Dynamic adjustment rule: When the motion state is marked as True: To improve the response speed, the window is reduced to 15 seconds; During the resting state: To ensure stability, the 30-second window is restored.

[0137] 2) Processing effect: This filtering method can improve the signal-to-noise ratio in the walking state and better retain the physiological characteristics of HRV compared to traditional filtering.

[0138] Step 12: Outlier detection (noise removal) 1) Detection rules: Physiological constraint check: Effective range of R_R interval: 600ms < R_R < 1200ms, and values outside this range are directly marked as abnormal.

[0139] Statistical outlier detection: Using the 3σ principle: A segment is considered an abnormal segment if three consecutive points deviate from the mean by more than three times the standard deviation.

[0140] Motion-related outliers: During the period when the motion state is marked as True, the threshold is relaxed to 5σ to avoid misjudgment.

[0141] 2) Processing method: <​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​Traditional preprocessing methods: usually only use filtering algorithms with fixed parameters, lack dynamic adaptation to motion states, and outlier detection is too simple (such as simple thresholds).

[0147] Advantages of this embodiment: Adaptive filtering: Dynamically adjusts the window size according to the motion state, improving real-time performance while ensuring signal quality. Intelligent anomaly detection: Combines physiological constraints and statistical methods to reduce the false positive rate.

[0148] Feature engineering is the process of converting raw physiological signals (such as R-R interval sequences) into quantitative indicators with clear physiological significance. It is a critical bridge between data acquisition and intelligent analysis. Its core goal is to extract key characteristic parameters that reflect the state of the human autonomic nervous system and stress sensitivity from time series data through mathematical transformation and statistical analysis.

[0149] Among them, the time domain feature extraction is as follows: The calculation formula of SDNN (standard deviation of all R_R intervals) is: ; Physiological significance: Reflects the overall variability of HRV and is negatively correlated with stress level, that is, the smaller the value, the greater the stress.

[0150] Clinical reference value: normal range 100-150ms (resting state) The calculation formula of RMSSD (root mean square difference between adjacent R_R intervals) is: ; Physiological significance: Mainly reflects the short-term variability of parasympathetic nervous regulation.

[0151] Advantages: Sensitive to respiratory sinus arrhythmia (RSA).

[0152] The calculation formula for pNN50 (the proportion of R_R intervals with a difference > 50ms) is: *100%; Application scenario: Assess changes in vagal tone.

[0153] Among them, the advanced time domain features are as follows: Triangular Index: Definition: The ratio of the area to the height of the R_R interval histogram when fitted as a triangle.

[0154] Significance: Reflects the overall distribution of HRV.

[0155] SD1 / SD2 ratio: Calculation method: The relationship between short-term and long-term variability was analyzed by scatter plot of R_R interval.

[0156] Advantages: Visual display of autonomic nervous system balance state.

[0157] Among them, the frequency domain feature extraction is as follows: Power spectral density analysis method: Use the Welch method to estimate the segmented power spectrum.

[0158] The window function of the parameter setting is: Hanning window.

[0159] Number of segments: 8 (256 points per segment, 50% overlap).

[0160] Frequency range: covers ultra-low frequency to high frequency.

[0161] Table 4 Key frequency band division

[0162] Among them, the frequency domain characteristic parameters are as follows: Total power (TP): LF+HF+VLF (reflects overall autonomic nervous activity).

[0163] LF / HF ratio: ; Clinical significance: When LH_HF<1: parasympathetic dominance, that is, in a relaxed state; When LH_HF>2: sympathetic dominance, i.e. stress / stress state; Normal range: 0.8-2.0.

[0164] Among them, nonlinear feature extraction is as follows: 1) Sample entropy analysis How it works: Complexity is assessed by comparing the probability of pattern repetition in a time series.

[0165] Parameter settings: embedding dimension m=2; tolerance r=0.2×SDNN; Physiological significance: Higher values ​​indicate more complex HRV signals and lower entropy under stress conditions.

[0166] 2) Multiscale entropy analysis Methods: Sample entropy was calculated at different time scales (1-20).

[0167] Advantages: Captures the dynamic changes of HRV at different time scales and distinguishes physiological from pathological HRV reduction.

[0168] Among them, multimodal feature fusion is as follows: 1) Motion state characteristics Source: Accelerometer data; Extraction method: Step rate of change: .

[0169] Exercise intensity classification: divided into resting / walking / running states based on acceleration amplitude.

[0170] The classification model is the core intelligent decision-making module of this embodiment. Its function is to automatically determine the individual's stress sensitivity level (low / medium / high) based on the HRV features and multimodal data extracted by feature engineering.

[0171] This module uses advanced machine learning algorithms combined with a dynamic weight adjustment mechanism to adapt to the physiological characteristics and stress response patterns of different individuals and achieve high-precision stress assessment.

[0172] This classification model adopts a multi-level classification architecture, which includes two levels of decision-making process: 1) First-level classification (coarse classification): binary classification task (normal / stress state).

[0173] Purpose: To quickly screen out individuals who may have stress problems; Algorithm selection: Logistic Regression; Advantages: High computational efficiency, suitable as a preliminary screening tool.

[0174] 2) Second-level classification (fine classification): multi-classification tasks (low / medium / high sensitivity).

[0175] Purpose: To accurately assess an individual's level of stress sensitivity; Algorithm selection: XGBoost (Extreme Gradient Boosting); Advantages: Strong ability to handle nonlinear relationships and support feature importance analysis.

[0176] Among them, model training and optimization are as follows.

[0177] 1) Training data preparation Data source: PhysioNet HRV database (public dataset), data collected from real users (n=500, including different ages, genders, and health statuses).

[0178] Tag definition: Normal state: SDNN>100ms, LF / HF<1.5; Low sensitivity: SDNN>50ms, LF / HF<1.0; Medium sensitivity: 30ms <SDNN≤50ms,1.0<LF / HF<2.5; High sensitivity: SDNN≤30ms, LF / HF>2.5.

[0179] 2) Feature selection and dimensionality reduction Feature screening method: Mutual Information Analysis Recursive Feature Elimination (RFE); Retain features, 8 most important features: SDNN; LF / HF ratio; RMSSD; Sample entropy; pNN50; LF power; HF power; Step change rate (multimodal fusion feature).

[0180] 3) Model training parameters XGBoost parameter configuration: Tree depth (max_depth): 6; Learning rate (learning_rate): 0.1; Subsample: 0.8 Minimum number of child node samples (min_child_weight): 5; Regularization parameter (lambda): 1.0; Logistic regression parameter configuration: Regularization type: L2; Regularization strength (C): 1.0.

[0181] The dynamic weight adjustment mechanism is as follows: 1) Feature importance update Method: Dynamic weight adjustment based on SHAP values ​​(SHapley Additive exPlanations).

[0182] Update frequency: When the model performance drops by more than 5%.

[0183] Calculation process: (1) Calculate the SHAP value of all training data using the current model; (2) Recalculate the importance score of each feature based on the SHAP value; (3) Adjust the feature weights during model training (features with high importance are given higher weights); 2) Individualized weight adjustment Method: Personalized weight optimization driven by user historical data.

[0184] Implementation: Maintain a personalized weight vector for each user and dynamically adjust the weight based on the user's latest stress assessment results.

[0185] formula: ; in: w_i: weight of the i-th feature; w_i_base: base weight; a: adjustment coefficient (0.1-0.3); performance_j: the latest evaluation accuracy of user j; baseline: baseline accuracy.

[0186] Among them, the implementation of modal data fusion includes: data synchronization mechanism and feature fusion algorithm.

[0187] In the data synchronization mechanism, timestamp alignment: the wearable device and the mobile app synchronize time using the NTP protocol, and the data packet contains a timestamp accurate to the millisecond. Motion status marking: accelerometer data and PPG data are aligned by timestamp, with a motion status marking accuracy of ±50ms.

[0188] In the feature fusion algorithm, weighted fusion model: ; in w1,w2,w3: weights dynamically adjusted based on SHAP values; F_ppg: PPG feature vector; F_acc: accelerometer feature vector; F_voice: voice feature vector.

[0189] Among them, decision-level fusion: The PPG model outputs the pressure probability P1; The accelerometer model outputs the pressure probability P2; Final probability: P=0.7P1+0.3P2.

[0190] Example 2 Embodiment 2 includes all the technical features of embodiment 1.

[0191] The present embodiment provides a stress sensitivity classification and assessment method based on heart rate variability, including: collecting a photoplethysmography signal, and processing the photoplethysmography signal to form an R_R interval sequence; Collect acceleration signals in three directions, calculate the vector sum of the accelerations in the three directions, obtain the total acceleration, determine the motion state and set the motion state flag; Aligning the RR interval sequence with the motion state marker according to timestamps to generate structured data; extracting a heart rate variability (HRV) feature vector from the R_R interval sequence in the structured data after preprocessing, and extracting an acceleration feature vector from the combined acceleration in the structured data; The individual's stress sensitivity level is determined according to the heart rate variability (HRV) feature vector and the acceleration feature vector.

[0192] Furthermore, the collecting of the photoplethysmography signal and processing the photoplethysmography signal to form an R_R interval sequence includes: collecting blood volume change signals through a photoplethysmography sensor module, and digitally processing the blood volume change signals to form time-series photoplethysmography signals; Smoothing the photoplethysmography signal to remove high-frequency noise and form a heartbeat signal; By analyzing the first-order derivative changes of the heartbeat signal, the peak points representing the heartbeat events are identified, and the time intervals between adjacent peak points are calculated to obtain the R_R interval sequence.

[0193] Furthermore, the collecting acceleration signals in three directions, calculating the vector sum of the accelerations in the three directions to obtain the total acceleration, determining the motion state and setting the motion state flag includes: Detect the three-axis acceleration value through the acceleration sensor module ; pass Calculate the resultant acceleration; pass Get the acceleration amplitude change rate of each time step, where steps_{t-1} is the total acceleration corresponding to time step t-1, and steps_t is the total acceleration corresponding to time step t; The motion state is determined according to the acceleration amplitude change rate and a motion state flag is set.

[0194] Furthermore, judging the motion state according to the acceleration amplitude change rate and setting the motion state flag includes: When the rate of change of the combined acceleration amplitude is greater than or equal to a first threshold, it is determined that the current motion state is a running state, and the motion state flag is set to a running state flag; When the rate of change of the combined acceleration amplitude is between a first threshold and a second threshold, it is determined that the current motion state is a walking state, and the motion state flag is set to a walking state flag; When the rate of change of the combined acceleration amplitude is less than or equal to a second threshold, it is determined that the current motion state is a resting state, and the motion state flag is set to a resting state flag.

[0195] Furthermore, the extracting of a heart rate variability (HRV) feature vector after preprocessing the R_R interval sequence in the structured data includes: performing sliding window filtering and outlier detection processing on the R_R interval sequence to form clean data, extracting a heart rate variability (HRV) feature vector from the clean data, and classifying the heart rate variability (HRV) feature vector into time domain features, frequency domain features, and nonlinear features; The time domain features include the standard deviation of the R_R interval, the root mean square of the difference between adjacent R_R intervals, the proportion of R_R intervals with a difference greater than 50 ms, the area-to-height ratio when the R_R interval histogram is fitted into a triangle, and the relationship between short-term and long-term variability is analyzed through a scatter plot of the R_R intervals; Setting the frequency domain features to include total power and LF / HF ratio, wherein the LF / HF ratio is a ratio of low frequency power LF to high frequency power HF; The nonlinear features are set to include sample entropy and multi-scale entropy.

[0196] Furthermore, performing sliding window filtering and outlier detection processing on the RR interval sequence to form clean data includes: Setting a sliding window to adjust the size of the sliding window according to the motion state mark, setting the sliding window to a first duration window when the motion state mark is a resting state, setting the sliding window to a second duration window when the motion state mark is a walking state, and setting the sliding window to a third duration window when the motion state mark is a running state, wherein the first duration is greater than the second duration, and the second duration is greater than the third duration; Performing sliding window filtering on the R_R interval sequence to remove motion artifacts; Obtaining a valid range of an R_R interval of physiological constraints, and marking a value of the R_R interval sequence as an abnormal point when the value exceeds the valid range of the R_R interval; When the movement state is marked as a resting state, when three consecutive points deviate from the mean by more than three times the standard deviation, they are marked as abnormal segments; When the motion state is marked as a walking state or a running state, when three consecutive points deviate from the mean by more than five times the standard deviation, they are marked as abnormal segments; The values ​​of the abnormal points are deleted and the missing values ​​are filled by linear interpolation; Manually review the abnormal segment and replace the value of the abnormal segment with the mean value; The photoplethysmography signal after smoothing and removing high-frequency noise is standardized or normalized to form clean data.

[0197] Furthermore, the determining of an individual's stress sensitivity level based on the heart rate variability (HRV) feature vector and the acceleration feature vector includes: A multi-level classification architecture is used to construct a classification model, wherein the classification model includes a first-level classification and a second-level classification. The first-level classification uses a logistic regression algorithm to distinguish between normal state and stress state. The second-level classification uses an XGBoost algorithm to assess an individual's stress sensitivity level, wherein the stress sensitivity level includes low sensitivity, medium sensitivity, and high sensitivity. Acquire training data, set normal state labels, low sensitivity labels, medium sensitivity labels, and high sensitivity labels for the training data, and perform important feature screening on the training data; Using the training data to train the classification model, and dynamically adjusting the weights of the important features based on SHAP values; The trained classification model is used to determine the stress sensitivity level of an individual according to the heart rate variability (HRV) feature vector and the acceleration feature vector, and stress assessment results corresponding to the first level classification and the second level classification are output.

[0198] Furthermore, the dynamically adjusting the weights of the important features based on the SHAP values ​​includes: Collecting an EEG signal and an ECG signal, processing the EEG signal and the ECG signal to obtain a speech feature vector, processing the heart rate variability (HRV) feature vector to obtain a photoplethysmography feature vector, and using the photoplethysmography feature vector, the acceleration feature vector, and the speech feature vector as important features; A weighted fusion model is constructed based on the important features. The weighted fusion model is: , where w1 is the weight of the photoplethysmography feature vector, F_ppg is the photoplethysmography feature vector, w2 is the weight of the acceleration feature vector, F_acc is the acceleration feature vector, w3 is the weight of the voice feature vector, and F_voice is the voice feature vector; When the performance of the classification model drops by more than a preset ratio, the SHAP value of all training data is calculated using the current classification model; the importance score of each feature is recalculated based on the SHAP value; and the weight of the important features is adjusted during the training of the current classification model.

[0199] Furthermore, the determining of an individual's stress sensitivity level based on the heart rate variability (HRV) feature vector and the acceleration feature vector using the trained classification model includes: Obtaining a probability P1 that the output individual has stress according to the heart rate variability (HRV) feature vector; Obtaining a probability P2 that the output individual has stress according to the acceleration feature vector; The total probability P of the individual experiencing stress is calculated and outputted by P=0.7P1+0.3P2, and the individual's stress sensitivity level is determined based on the total probability P.

[0200] Furthermore, the stress sensitivity classification and assessment method based on heart rate variability also includes: Maintain a personalized weight vector for each user and dynamically adjust the personalized weight based on the user's latest stress assessment results; The formula for adjusting the personalization weight is set as ; where w_i is the weight of the i-th feature, w_i_base is the base weight, a is the adjustment coefficient, 0.1≤a≤0.3, performance_j is the accuracy of the latest stress assessment result of user j, and baseline is the baseline accuracy.

[0201] The present application also provides a wearable device for implementing the aforementioned stress sensitivity classification and assessment method based on heart rate variability, the wearable device comprising: Microprocessor (MCU); a pulse count (PPG) sensor connected to the microprocessor (MCU) via an I2C / SPI interface for collecting photoplethysmography signals; An acceleration sensor connected to the microprocessor (MCU) via an I2C interface, configured to detect three-dimensional motion state data of the device and process the three-dimensional motion state data to obtain acceleration signals in three directions; An electroencephalogram (EEG) module is connected to the microprocessor (MCU) via an SPI / UART interface for collecting EEG signals; an electrocardiogram (ECG) module, connected to the microprocessor (MCU) via a UART interface, for collecting ECG signals; A wireless transmission module (Bluetooth), connected to the microprocessor (MCU) via a UART interface, for transmitting data to external electronic devices; Wherein, the microprocessor (MCU) includes: a pulse number (PPG) processing unit, connected to the pulse number (PPG) sensor, for processing the photoplethysmography signal into an R_R interval sequence; an acceleration processing unit connected to the acceleration sensor, for calculating the vector sum of the accelerations in the three directions according to the acceleration signals in the three directions, obtaining a resultant acceleration, determining a motion state, and setting a motion state flag; a data acquisition module connected to the pulse number (PPG) processing unit and the acceleration processing unit, aligning the RR interval sequence with the motion state mark according to timestamps to generate structured data; A storage module, connected to the data acquisition module, is used to store the structured data, extract a heart rate variability (HRV) feature vector after preprocessing the R_R interval sequence in the structured data, extract an acceleration feature vector from the combined acceleration in the structured data, and determine the individual's stress sensitivity level based on the heart rate variability (HRV) feature vector and the acceleration feature vector.

[0202] The heart rate variability-based pressure sensitivity classification and assessment method and wearable device provided in the embodiments of the present application can process the collected signals to form an R_R interval sequence and a combined acceleration, determine the motion state and set a motion state mark, align the R_R interval sequence with the motion state mark according to the timestamp, generate structured data, and determine the individual's pressure sensitivity level based on the structured data, thereby avoiding the subjective influence of pressure sensitivity assessment and improving classification accuracy and adaptability.

[0203] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0204] The above is a detailed introduction to a pressure sensitivity classification and evaluation method based on heart rate variability and a wearable device provided in the embodiments of the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the technical solutions and core ideas of the present application. Ordinary technicians in this field should understand that they can still modify the technical solutions recorded in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A stress sensitivity classification and assessment method based on heart rate variability, characterized in that: include: collecting a photoplethysmography signal, and processing the photoplethysmography signal to form an R_R interval sequence; Collect acceleration signals in three directions, calculate the vector sum of the accelerations in the three directions, obtain the total acceleration, determine the motion state and set the motion state flag; Aligning the RR interval sequence with the motion state marker according to timestamps to generate structured data; extracting a heart rate variability feature vector from the R_R interval sequence in the structured data after preprocessing, and extracting an acceleration feature vector from the combined acceleration in the structured data; Determining the stress sensitivity level of an individual according to the heart rate variability feature vector and the acceleration feature vector, including: constructing a classification model using a multi-level classification architecture, the classification model including a first-level classification and a second-level classification, the first-level classification distinguishing between a normal state and a stress state by a logistic regression algorithm, and the second-level classification using an XGBoost algorithm to evaluate the stress sensitivity level of an individual, the stress sensitivity level including low sensitivity, medium sensitivity, and high sensitivity; acquiring training data, setting a normal state label, a low sensitivity label, a medium sensitivity label, and a high sensitivity label for the training data, and screening important features of the training data; training the classification model using the training data, and dynamically adjusting the weights of the important features based on SHAP values; using the trained classification model to determine the stress sensitivity level of an individual according to the heart rate variability feature vector and the acceleration feature vector, and outputting stress assessment results corresponding to the first-level classification and the second-level classification; The dynamically adjusting the weights of the important features based on the SHAP values ​​includes: Collecting EEG signals and ECG signals, processing the EEG signals and ECG signals to obtain a speech feature vector, processing the heart rate variability feature vector to obtain a photoplethysmography feature vector, and using the photoplethysmography feature vector, the acceleration feature vector, and the speech feature vector as important features; A weighted fusion model is constructed based on the important features. The weighted fusion model is: , where w1 is the weight of the photoplethysmography feature vector, F_ppg is the photoplethysmography feature vector, w2 is the weight of the acceleration feature vector, F_acc is the acceleration feature vector, w3 is the weight of the voice feature vector, and F_voice is the voice feature vector; When the performance of the classification model drops by more than a preset ratio, the SHAP value of all training data is calculated using the current classification model; the importance score of each feature is recalculated based on the SHAP value; and the weight of the important features is adjusted during the training of the current classification model; The step of using the trained classification model to determine the individual's stress sensitivity level according to the heart rate variability feature vector and the acceleration feature vector includes: Obtaining a probability P1 of an output individual having stress according to the heart rate variability feature vector; Obtaining a probability P2 that the output individual has stress according to the acceleration feature vector; The total probability P of the individual experiencing stress is calculated and outputted by P=0.7P1+0.3P2, and the individual's stress sensitivity level is determined based on the total probability P.

2. The stress sensitivity classification and assessment method based on heart rate variability according to claim 1, wherein: The collecting of the photoplethysmography signal and processing the photoplethysmography signal to form an R-R interval sequence includes: collecting blood volume change signals through a photoplethysmography sensor module, and digitally processing the blood volume change signals to form time-series photoplethysmography signals; Smoothing the photoplethysmography signal to remove high-frequency noise and form a heartbeat signal; By analyzing the first-order derivative changes of the heartbeat signal, the peak points representing the heartbeat events are identified, and the time intervals between adjacent peak points are calculated to obtain the R_R interval sequence.

3. The stress sensitivity classification and assessment method based on heart rate variability according to claim 1, wherein: The collecting of acceleration signals in three directions, calculating the vector sum of the accelerations in the three directions to obtain the resultant acceleration, determining the motion state and setting the motion state flag comprises: Detect the three-axis acceleration value through the acceleration sensor module ; pass Calculate the resultant acceleration; pass Get the acceleration amplitude change rate of each time step, where steps_{t-1} is the total acceleration corresponding to time step t-1, and steps_t is the total acceleration corresponding to time step t; The motion state is determined according to the acceleration amplitude change rate and a motion state flag is set.

4. The stress sensitivity classification and assessment method based on heart rate variability according to claim 3, wherein: The determining the motion state according to the acceleration amplitude change rate and setting the motion state flag comprises: When the rate of change of the combined acceleration amplitude is greater than or equal to a first threshold, it is determined that the current motion state is a running state, and the motion state flag is set to a running state flag; When the rate of change of the combined acceleration amplitude is between a first threshold and a second threshold, it is determined that the current motion state is a walking state, and the motion state flag is set to a walking state flag; When the rate of change of the combined acceleration amplitude is less than or equal to a second threshold, it is determined that the current motion state is a resting state, and the motion state flag is set to a resting state flag.

5. The stress sensitivity classification and assessment method based on heart rate variability according to claim 1, wherein: The step of preprocessing the R_R interval sequence in the structured data and extracting the heart rate variability feature vector comprises: performing sliding window filtering and outlier detection processing on the R_R interval sequence to form clean data, extracting a heart rate variability feature vector from the clean data, and classifying the heart rate variability feature vector into time domain features, frequency domain features, and nonlinear features; The time domain features include the standard deviation of the R_R interval, the root mean square of the difference between adjacent R_R intervals, the proportion of R_R intervals with a difference greater than 50 ms, the area-to-height ratio when the R_R interval histogram is fitted into a triangle, and the relationship between short-term and long-term variability is analyzed through a scatter plot of the R_R intervals; Setting the frequency domain features to include total power and LF / HF ratio, wherein the LF / HF ratio is a ratio of low frequency power LF to high frequency power HF; The nonlinear features are set to include sample entropy and multi-scale entropy.

6. The stress sensitivity classification and assessment method based on heart rate variability according to claim 5, wherein: The performing sliding window filtering and outlier detection processing on the R_R interval sequence to form clean data comprises: Setting a sliding window to adjust the size of the sliding window according to the motion state mark, setting the sliding window to a first duration window when the motion state mark is a resting state, setting the sliding window to a second duration window when the motion state mark is a walking state, and setting the sliding window to a third duration window when the motion state mark is a running state, wherein the first duration is greater than the second duration, and the second duration is greater than the third duration; Performing sliding window filtering on the R_R interval sequence to remove motion artifacts; Obtaining a valid range of an R_R interval of physiological constraints, and marking a value of the R_R interval sequence as an abnormal point when the value exceeds the valid range of the R_R interval; When the movement state is marked as a resting state, when three consecutive points deviate from the mean by more than three times the standard deviation, they are marked as abnormal segments; When the motion state is marked as a walking state or a running state, when three consecutive points deviate from the mean by more than five times the standard deviation, they are marked as abnormal segments; The values ​​of the abnormal points are deleted and the missing values ​​are filled by linear interpolation; Manually review the abnormal segment and replace the value of the abnormal segment with the mean value; The photoplethysmography signal after smoothing and removing high-frequency noise is standardized or normalized to form clean data.

7. The stress sensitivity classification and assessment method based on heart rate variability according to claim 1, wherein: Also includes: Maintain a personalized weight vector for each user and dynamically adjust the personalized weight based on the user's latest stress assessment results; The formula for adjusting the personalization weight is set as ; where w_i is the weight of the i-th feature, w_i_base is the base weight, a is the adjustment coefficient, 0.1≤a≤0.3, performance_j is the accuracy of the latest stress assessment result of user j, and baseline is the baseline accuracy.

8. A wearable device for implementing the stress sensitivity classification and assessment method based on heart rate variability according to any one of claims 1 to 7, characterized in that: The wearable device includes: microprocessor; a pulse number sensor connected to the microprocessor via an I2C / SPI interface for collecting photoplethysmography signals; An acceleration sensor connected to the microprocessor via an I2C interface, configured to detect three-dimensional motion state data of the device and process the three-dimensional motion state data to obtain acceleration signals in three directions; An EEG module, connected to the microprocessor via an SPI / UART interface, for collecting EEG signals; an electrocardiogram module, connected to the microprocessor via a UART interface, for collecting electrocardiogram signals; A wireless transmission module, connected to the microprocessor via a UART interface, for transmitting data to external electronic devices; Wherein, the microprocessor includes: a pulse number processing unit, connected to the pulse number sensor, for processing the photoplethysmography signal into an R_R interval sequence; an acceleration processing unit connected to the acceleration sensor, for calculating the vector sum of the accelerations in the three directions according to the acceleration signals in the three directions, obtaining a resultant acceleration, determining a motion state, and setting a motion state flag; a data acquisition module connected to the pulse number processing unit and the acceleration processing unit, aligning the RR interval sequence with the motion state mark according to timestamps to generate structured data; A storage module, connected to the data acquisition module, is used to store the structured data, extract a heart rate variability feature vector after preprocessing the R_R interval sequence in the structured data, extract an acceleration feature vector from the combined acceleration in the structured data, and determine the individual's stress sensitivity level based on the heart rate variability feature vector and the acceleration feature vector.

Images