Non-contact physiological pressure monitoring system and method

Abstract

The application provides a kind of non-contact physiological pressure monitoring system and method, comprising: multi-modal signal acquisition module, for the synchronous acquisition of the multispectral rPPG signal of measured person, thermal imaging signal and millimeter wave radar signal;Signal pre-processing module, for the time alignment of collected signal, noise filtering, baseline drift correction and motion artifact elimination;Feature extraction and fusion analysis module, for extracting time domain feature, frequency domain feature, nonlinear feature and HPA interaxial indirect representation feature from pre-processed signal;Pressure value quantification module, for calculating stress index based on dynamic baseline adaptive calibration and integrated learning model;Output and early warning interaction module, for outputting stress level determination result and timing evolution early warning information, the application effectively distinguishes low-frequency baseline drift and effective pulse wave signal, while removing the drift caused by ambient light change and slow displacement, the effective frequency band of pulse wave and its morphological characteristics are maximized.

Description

Technical Field

[0001] This invention relates to the field of physiological monitoring technology, and in particular to a non-contact physiological stress monitoring system and method. Background Technology

[0002] With the fast pace of modern life, psychological stress has become a significant factor affecting public health. Prolonged or excessive psychological stress can lead to autonomic nervous system dysfunction, resulting in various health problems such as cardiovascular disease, weakened immune system, and mental disorders. Therefore, real-time and accurate monitoring of an individual's physiological stress level is crucial for early intervention and health management.

[0003] Stress level refers to the intensity of the body's non-specific stress response to internal and external environmental stimuli, mainly reflected in the dynamic imbalance of the autonomic nervous system (especially the sympathetic-parasympathetic balance). It reflects the combined physiological effects of the activation level of the hypothalamus-pituitary-adrenal axis (HPA axis) and the increased excitability of the sympathetic nervous system. It is a psychophysiological load indicator. Acute or chronic stress states are usually characterized by a relative increase in sympathetic nerve activity and a relative inhibition of parasympathetic nerve activity, leading to measurable adaptive changes in the endocrine system.

[0004] Existing physiological stress monitoring technologies are mainly divided into two categories: contact and non-contact. Although contact monitoring (such as electrocardiogram ECG, skin conductance response GSR, etc.) has high accuracy, it requires wearing sensors or electrodes, which not only causes discomfort to users but also restricts their activities in a natural state, making it difficult to achieve long-term continuous monitoring, and is prone to artifacts due to loosening of the wearer.

[0005] Non-contact monitoring technology has developed rapidly in recent years, mainly based on remote photoplethysmography (rPPG) technology. This technology uses a camera to capture weak blood flow changes in the face to deduce heart rate variability (HRV) and thus assess stress levels. However, existing non-contact stress monitoring technologies still have the following significant shortcomings in signal processing: 1. Severe Baseline Drift Issue: rPPG signals are highly susceptible to slow changes in ambient light, slight subject displacement, and sensor thermal noise, resulting in low-frequency baseline drift. Traditional denoising methods (such as high-pass filtering with a fixed cutoff frequency or simple polynomial fitting) often use fixed thresholds or parameters, which cannot adapt to dynamic changes in different individuals and environments. When the drift frequency overlaps with the effective pulse wave frequency band, traditional methods easily lead to distortion of the effective signal or incomplete drift removal, severely affecting the accuracy of subsequent feature extraction.

[0006] 2. Insufficient Motion Artifact Removal Capability: In real-world applications, subjects inevitably exhibit subtle movements such as head rotation and facial expression changes. The artifacts generated by these movements are often much larger than the actual pulse wave signal. Existing artifact removal techniques mostly employ blind source separation (BSS) or independent component analysis (ICA) algorithms. However, the single BSS algorithm performs poorly when the assumption of statistical independence of source signals is not met, while the single ICA algorithm is sensitive to initial values ​​and prone to getting trapped in local optima. Furthermore, existing technologies lack robust discrimination criteria when identifying and retaining the true pulse wave components from multiple independent components. They often mistakenly retain high-frequency noise or low-frequency drift as valid signals or incorrectly remove components containing real physiological information, resulting in a low signal-to-noise ratio of the reconstructed signal and unreliable stress assessment results.

[0007] Therefore, it is necessary to provide a new non-contact physiological stress monitoring system and method to solve the above-mentioned technical problems. Summary of the Invention

[0008] To address the aforementioned technical problems, this invention provides a non-contact physiological pressure monitoring system and method.

[0009] The non-contact physiological pressure monitoring system provided by this invention includes: The multimodal signal acquisition module is used to simultaneously acquire the multispectral rPPG signal, thermal imaging signal and millimeter-wave radar signal of the subject, and to synchronize the timestamps of each acquisition module through the PTP precise time protocol; The signal preprocessing module, connected to the multimodal signal acquisition module, is used to perform time alignment, noise filtering, baseline drift correction, and motion artifact elimination on the acquired signals. The baseline drift removal employs an adaptive wavelet threshold denoising method, which performs multi-layer wavelet decomposition on the original rPPG signal, reconstructs the signal after setting a dynamic threshold, and retains the effective frequency band of the pulse wave. The motion artifact removal employs a fusion algorithm of blind source separation and independent component analysis to construct a multi-region rPPG signal matrix. After calculating the covariance matrix, whitening is performed, and independent component analysis is used to extract independent components. Based on the kurtosis criterion, pulse wave components are identified and retained before reconstructing the artifact-free signal. The feature extraction and fusion analysis module is connected to the signal preprocessing module and is used to extract time-domain features, frequency-domain features, nonlinear features and HPA axis indirect characterization features from the preprocessed signal. The stress value quantification module, connected to the feature extraction and fusion analysis module, is used to calculate the stress index based on dynamic baseline adaptive calibration and ensemble learning model; The output and early warning interaction module is connected to the pressure value quantification module and is used to output the pressure level determination result and the time-series evolution early warning information.

[0010] Furthermore, the multimodal signal acquisition module includes: The multispectral rPPG acquisition unit is used to simultaneously acquire pulse wave signals in the green, red and near-infrared bands, and independently extract signals from the forehead, left cheek, right cheek and the area around the nose. The thermal imaging acquisition unit is used to acquire temperature change signals in the nose tip, around the eyes, and cheek areas. The millimeter-wave radar acquisition unit is used to acquire thoracic cavity micro-motion signals using the frequency-modulated continuous wave band. An ambient light compensation unit is used to correct ambient light interference in conjunction with an ambient light sensor.

[0011] Furthermore, the time-domain features include the mean pulse interval, the standard deviation of the pulse interval, the root mean square of the difference between adjacent pulse intervals, the proportion of the difference between adjacent pulse intervals exceeding a preset threshold, the pulse interval triangular index, and the stress volatility. The frequency domain feature extraction uses the periodogram method to calculate the power of the extremely low frequency, low frequency and high frequency bands, and calculates the low frequency to high frequency ratio, normalized low frequency, normalized high frequency and stress spectrum index. The nonlinear features include sample entropy, multi-scale entropy, detrended fluctuation analysis scaling index, Poincaré plot short-to-long axis ratio, and recursive quantitative analysis parameters. The indirect characterization features of the HPA axis include the skin sympathetic response index, the HPA expression index, the respiratory-heart rate coupling index, and the comprehensive indirect index of the HPA axis.

[0012] Furthermore, the stress volatility is the ratio of the standard deviation of the stress state window to the standard deviation of the baseline state window; The stress spectrum index is the ratio of the stress-state low-frequency power to the baseline low-frequency power divided by the ratio of the stress-state high-frequency power to the baseline high-frequency power. The skin sympathetic response index is a negative value of the rate of change in nasal temperature. The HPA expression index is the product of the duration, frequency, and intensity coefficient of micro-expressions; The respiratory-heart rate coupling index is the coherence between the respiratory signal spectrum and the heart rate variability spectrum in a preset frequency band. The HPA axis comprehensive indirect index is a weighted sum of the skin sympathetic response index, HPA expression index, respiratory-heart rate coupling index, and low-frequency-high-frequency ratio.

[0013] Furthermore, the pressure value quantification module performs the following processing: Dynamic baseline adaptive calibration establishes an initial baseline upon first use. The initial baseline is a combination of the median and the absolute deviation of the median of the eigenvalues. It is updated using a rolling baseline update formula, and a diurnal rhythm baseline curve is established for rhythm correction. Feature vector construction combines time-domain features, frequency-domain features, nonlinear features, and HPA axis indirect characterization features into a multi-dimensional feature vector. The stress index is calculated using an ensemble learning stress regression model, which includes a random forest regressor, a long short-term memory network temporal model, and a temporal attention mechanism. After standardizing the features, static stress scores and temporal stress evolution scores are calculated. The final stress index is obtained by attention-weighted fusion and then normalized. The stress level assessment divides the stress index into multiple levels, corresponding to a relaxed state, mild stress, moderate stress, high stress, and extreme stress.

[0014] Furthermore, the output and early warning interaction module includes: The pressure trend prediction unit uses a time-series prediction model, takes historical pressure index sequences as input, and outputs pressure prediction values ​​for a preset time period in the future. The abnormal stress event detection unit trains a normal stress pattern based on the isolated forest algorithm, detects abnormal deviations in real time, and triggers an early warning when the abnormal score exceeds a preset threshold. The Chronic Stress Accumulation Index (CSCI) calculation unit calculates the CSCI as the integral of the difference between the stress index and the baseline over time. When the CSCI exceeds a preset threshold, it indicates a risk of chronic stress.

[0015] Another aspect of the present invention provides a non-contact physiological pressure monitoring method, the method comprising the following steps: Step 1: Synchronous acquisition of multimodal non-contact signals. Multi-band pulse wave signals are acquired synchronously through the multispectral rPPG acquisition module, facial temperature change signals are acquired through the thermal imaging acquisition module, and chest cavity micro-motion signals are acquired through the millimeter-wave radar acquisition module. The time synchronization protocol is used to synchronize the timestamps of each module. Step 2, signal preprocessing and quality assessment: baseline drift removal is performed on the acquired signal, denoising is carried out using wavelet decomposition and adaptive thresholding, motion artifact elimination is performed, and after constructing a multi-region signal matrix, pulse wave components are extracted using independent component analysis algorithm, and signal quality index is calculated. When the signal quality index is lower than the preset threshold, it is marked as low quality data. Step 3: Pulse interval sequence feature extraction. The multi-scale differential thresholding method is used to detect the peak value of the pulse wave, calculate the first-order difference of the signal and find the zero-crossing point within the sliding window, set an adaptive threshold to verify the physiological interval between peaks, output the pulse interval sequence, and extract time-domain features, frequency-domain features and nonlinear features from the pulse interval sequence. Step 4: Non-contact indirect characterization of HPA axis activation level. The skin sympathetic response index is calculated based on the rate of change of nasal temperature in thermal imaging. Stress-related micro-expressions are detected through facial video analysis and the HPA expression index is calculated. Respiratory signals are extracted synchronously from rPPG and radar signals and the respiratory-heart rate coupling index is calculated. The comprehensive indirect index of HPA axis is calculated. Step 5: Multimodal feature fusion and stress value calculation. Construct a multidimensional feature vector, perform dynamic baseline adaptive calibration, establish an initial baseline for the first time and update it using a rolling baseline update formula, establish a diurnal rhythm baseline curve for correction, use an ensemble learning stress regression model to standardize the features and calculate the static stress score and time-series stress evolution score, and obtain the final stress index through attention-weighted fusion and normalization. Step 6: Temporal evolution modeling and early warning. A time-series prediction model is used to input the historical stress index sequence and output the predicted future stress value. The normal stress pattern is trained based on the isolated forest algorithm and abnormal deviations are detected in real time. When the abnormal score exceeds the preset threshold, an early warning is triggered. The chronic stress accumulation index is calculated. When the chronic stress accumulation index exceeds the preset threshold, a chronic stress risk is indicated.

[0016] Furthermore, the multi-scale differential thresholding method for pulse wave peak detection in step three specifically includes: Calculate the first-order difference of the preprocessed rPPG signal; Find the zero-crossing point where the first-order difference changes from positive to negative within the sliding window; Set an adaptive threshold, which is the sum of the mean difference within the window and the coefficient multiplied by the standard deviation of the difference within the window; Verify the minimum physiological interval between detected peaks and remove spurious peaks that do not meet physiological constraints; The pulse interval sequence is obtained by subtracting adjacent peak timestamps. Outlier removal is performed on the pulse interval sequence. When the difference between the pulse interval value and the median exceeds a preset multiple of the absolute deviation of the median, it is marked as an outlier and replaced with linear interpolation.

[0017] Furthermore, the training process of the integrated learning stress regression model in step five includes: Data collection and labeling: Recruit subjects to conduct stress tests and simultaneously collect salivary cortisol samples as a stress standard label; Feature standardization involves standardizing multidimensional feature vectors. Train a random forest regressor, set the number of decision trees and maximum depth, optimize hyperparameters using cross-validation, and output a static stress score; Training a temporal model of a Long Short-Term Memory Network: Input a stress feature sequence, set the hidden layer dimension and the number of network layers, and output a temporal stress evolution score. A temporal attention mechanism is used to calculate attention weights and perform weighted fusion of static stress scores and temporal stress evolution scores. The model output is normalized, and the fused score is limited to a preset range to obtain the stress index. Model validation: Cross-validation was used to verify the correlation between stress index and cortisol and the accuracy of stress level determination.

[0018] Furthermore, the calculation of the chronic stress accumulation index in step six specifically includes: The pressure index is discretized and sampled using a preset time unit to obtain a time series; Calculate the pressure deviation value at each time point. The pressure deviation value is the difference between the pressure index at that time point and the dynamic baseline value for the corresponding period. The daily chronic stress accumulation index is obtained by approximating the deviation value through time integration. The cumulative index is obtained by summing up the daily chronic stress cumulative index over multiple consecutive days; Set risk thresholds. When the cumulative chronic stress index exceeds the first threshold, it is marked as a high stress load for the day. When the cumulative index exceeds the second threshold, a chronic stress risk is indicated. When the cumulative index exceeds the third threshold, a high-risk warning is triggered.

[0019] Compared with related technologies, the non-contact physiological pressure monitoring system and method provided by the present invention have the following beneficial effects: 1. The adaptive wavelet threshold denoising method of this invention removes baseline drift. By performing multi-layer wavelet decomposition on the original rPPG signal, the signal can be finely characterized at different scales. Unlike the traditional fixed threshold method, the dynamic threshold set by this invention can be adjusted in real time according to the local noise level and energy distribution of the signal. It can effectively distinguish between low-frequency baseline drift and effective pulse wave signal. While removing drift caused by changes in ambient light and slow displacement, it preserves the effective frequency band and morphological characteristics of the pulse wave to the maximum extent.

[0020] 2. This invention employs a motion artifact elimination algorithm that integrates blind source separation and independent component analysis. First, it constructs an rPPG signal matrix encompassing multiple facial regions, leveraging the spatial diversity of these signals to increase information redundancy. Next, it calculates the covariance matrix and performs whitening to eliminate correlations between signals. Then, it extracts statistically independent source components. Crucially, this invention introduces an automatic identification mechanism based on the kurtosis criterion. Utilizing the unique super-Gaussian distribution characteristics of pulse wave signals, it accurately identifies and filters out genuine pulse wave components from numerous independent components, automatically eliminating motion-induced sub-Gaussian or Gaussian artifacts and noise. This fusion strategy combines the blind separation advantages of BSS with the high-order statistical properties of ICA, enabling the reconstruction of a high signal-to-noise ratio pure pulse wave signal even under severe facial motion interference, significantly improving the system's robustness in unconstrained scenarios. Attached Figure Description

[0021] Figure 1 This is a structural block diagram of the non-contact physiological pressure monitoring system provided by the present invention; Figure 2 This is a schematic diagram of the structure of the multimodal signal acquisition module provided by the present invention; Figure 3 This is a schematic diagram of the output and early warning interaction module provided by the present invention; Figure 4 This is a flowchart of the non-contact physiological pressure monitoring method provided by the present invention; Figure 5 The flowchart of the multi-scale differential thresholding method for pulse wave peak detection provided by the present invention is shown below. Figure 6 A flowchart illustrating the training process of the ensemble learning stress regression model provided by this invention; Figure 7 The flowchart for calculating the chronic stress accumulation index provided by this invention is shown. Detailed Implementation

[0022] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0023] Please refer to the following: Figure 1 , Figure 2 , Figure 3 , Figure 4 , Figure 5 , Figure 6 as well as Figure 7 ,in, Figure 1 This is a structural block diagram of the non-contact physiological pressure monitoring system provided by the present invention; Figure 2 This is a schematic diagram of the structure of the multimodal signal acquisition module provided by the present invention; Figure 3 This is a schematic diagram of the output and early warning interaction module provided by the present invention; Figure 4This is a flowchart of the non-contact physiological pressure monitoring method provided by the present invention; Figure 5 The flowchart of the multi-scale differential thresholding method for pulse wave peak detection provided by the present invention is shown below. Figure 6 A flowchart illustrating the training process of the ensemble learning stress regression model provided by this invention; Figure 7 The flowchart for calculating the chronic stress accumulation index provided by this invention is shown.

[0024] Example 1 In the specific implementation process, such as Figure 1 As shown, the non-contact physiological stress monitoring system includes: The multimodal signal acquisition module is used to simultaneously acquire the multispectral rPPG signal, thermal imaging signal and millimeter-wave radar signal of the subject, and to synchronize the timestamps of each acquisition module through the PTP precise time protocol; The signal preprocessing module, connected to the multimodal signal acquisition module, is used to perform time alignment, noise filtering, baseline drift correction, and motion artifact elimination on the acquired signals. Baseline drift removal employs an adaptive wavelet thresholding denoising method, performing multi-level Daubechies-4 wavelet decomposition on the original rPPG signal and setting a dynamic threshold. ,in As a time-adaptive adjustment factor, the reconstructed signal retains the effective frequency band of the pulse wave, 0.5-4Hz; Motion artifact removal employs a fusion algorithm combining blind source separation and independent component analysis to construct a multi-region rPPG signal matrix. Calculate the covariance matrix Whitening treatment The FastICA algorithm was used to extract independent components, and the pulse wave components were identified and preserved based on the kurtosis criterion before reconstructing the artifact-free signal. The feature extraction and fusion analysis module is connected to the signal preprocessing module and is used to extract time-domain features, frequency-domain features, nonlinear features, and HPA axis indirect characterization features from the preprocessed signal. The stress value quantification module, connected to the feature extraction and fusion analysis module, is used to calculate the stress index based on dynamic baseline adaptive calibration and ensemble learning model; The output and early warning interaction module is connected to the pressure value quantification module and is used to output the pressure level determination result and the time-series evolution early warning information.

[0025] It should be noted that the reference Figure 2 As shown, the multimodal signal acquisition module includes: The multispectral rPPG acquisition unit is used to simultaneously acquire pulse wave signals in the green, red and near-infrared bands, and independently extract signals from the forehead, left cheek, right cheek and the area around the nose. The thermal imaging acquisition unit is used to acquire temperature change signals in the nose tip, around the eyes, and cheek areas. The millimeter-wave radar acquisition unit is used to acquire thoracic cavity micro-motion signals using the frequency-modulated continuous wave band. An ambient light compensation unit is used to correct ambient light interference in conjunction with an ambient light sensor.

[0026] It should be noted that the time-domain features include the mean pulse interval, the standard deviation of the pulse interval, the root mean square of the difference between adjacent pulse intervals, the proportion of the difference between adjacent pulse intervals exceeding a preset threshold, the pulse interval triangular index, and the stress volatility. Frequency domain feature extraction uses the periodogram method to calculate the power in the extremely low frequency, low frequency and high frequency bands, and calculates the low frequency to high frequency ratio, normalized low frequency, normalized high frequency and stress spectrum index. Nonlinear characteristics include sample entropy, multiscale entropy, detrended fluctuation analysis scaling index, Poincaré plot short-to-long axis ratio, and recursive quantitative analysis parameters; Indirect characteristics of the HPA axis include the skin sympathetic response index, HPA expression index, respiratory-heart rate coupling index, and comprehensive indirect indicators of the HPA axis.

[0027] It should be noted that stress volatility is the ratio of the standard deviation of the stress state window to the standard deviation of the baseline state window; The stress spectrum index is the ratio of the stress-state low-frequency power to the baseline low-frequency power divided by the ratio of the stress-state high-frequency power to the baseline high-frequency power. The skin sympathetic response index is a negative value of the rate of change in nasal temperature. The HPA expression index is the product of the duration, frequency, and intensity coefficients of micro-expressions. The respiratory-heart rate coupling index is the coherence between the respiratory signal spectrum and the heart rate variability spectrum in a preset frequency band. The comprehensive indirect index of the HPA axis is a weighted sum of the skin sympathetic response index, HPA expression index, respiratory-heart rate coupling index, and the low-frequency to high-frequency ratio.

[0028] It should be noted that the pressure value quantification module performs the following processing: Dynamic baseline adaptive calibration establishes an initial baseline upon first use. The initial baseline is a combination of the median and the absolute deviation of the median of the eigenvalues. It is updated using a rolling baseline update formula, and a 24-hour pressure baseline curve is established for circadian rhythm correction. Feature vector construction combines time-domain features, frequency-domain features, nonlinear features, and HPA axis indirect characterization features into a multi-dimensional feature vector. The stress index is calculated using an ensemble learning stress regression model, which includes a random forest regressor, a long short-term memory network temporal model, and a temporal attention mechanism. After standardizing the features, static stress scores and temporal stress evolution scores are calculated. The final stress index is obtained by attention-weighted fusion and then normalized. The stress level is determined by dividing the stress index into multiple levels, corresponding to relaxed state, mild stress, moderate stress, high stress and extreme stress. The stress index ranges are as follows: 0-20 Level 1 Relaxation, 21-40 Level 2 Mild Stress, 41-60 Level 3 Moderate Stress, 61-80 Level 4 High Stress, and 81-100 Level 5 Extreme Stress.

[0029] It should be noted that the reference Figure 3 As shown, the output and early warning interaction module includes: The pressure trend prediction unit uses a time-series prediction model, takes historical pressure index sequences as input, and outputs pressure prediction values ​​for a preset time period in the future. The abnormal stress event detection unit trains a normal stress pattern based on the isolated forest algorithm, detects abnormal deviations in real time, and triggers an early warning when the abnormal score exceeds a preset threshold. The Chronic Stress Accumulation Index (CSCI) calculation unit calculates the CSCI as the integral of the difference between the stress index and the baseline over time. When the CSCI exceeds a preset threshold, it indicates a risk of chronic stress. The preset threshold for the CSCI is 500.

[0030] Example 2 In a specific implementation process, refer to Figure 4 As shown, the non-contact physiological stress monitoring method includes the following steps: Step 1: Multimodal non-contact signal synchronous acquisition. The multispectral rPPG acquisition module synchronously acquires pulse wave signals in three bands of 530nm, 660nm, and 850nm at a frame rate of ≥60fps. The thermal imaging acquisition module acquires temperature change signals of the tip of the nose, periocular area, and cheeks at a temperature resolution of ≤0.05℃ and a sampling frequency of 10Hz. The millimeter-wave radar acquisition module acquires chest cavity micro-motion signals in the 60GHz FMCW band. The PTP precise time protocol is used to achieve module timestamp synchronization with a synchronization accuracy of ≤1ms. Step 2, signal preprocessing and quality assessment: Baseline drift removal is performed on the acquired signal, denoising is carried out using wavelet decomposition and adaptive thresholding, motion artifact elimination is performed, and after constructing a multi-region signal matrix, pulse wave components are extracted using independent component analysis algorithm, and signal quality index is calculated. When the signal quality index is lower than a preset threshold, it is marked as low-quality data. The preset threshold for the signal quality index is 0.6. Step 3: Pulse interval sequence feature extraction. The multi-scale differential thresholding method is used to detect the peak value of the pulse wave, calculate the first-order difference of the signal and find the zero-crossing point within the sliding window, set an adaptive threshold to verify the physiological interval between peaks, output the pulse interval sequence, and extract time-domain features, frequency-domain features and nonlinear features from the pulse interval sequence. Step four involves non-contact indirect characterization of HPA axis activation levels. The skin sympathetic response index is calculated based on the rate of change in nasal temperature using thermal imaging. Stress-related micro-expressions are detected through facial video analysis, and the duration of HPA expressions is calculated. and frequency The HPA expression index was obtained. Respiratory signals are extracted synchronously from rPPG and radar signals, and the respiratory-heart rate coupling index is calculated to calculate the HPA axis comprehensive indirect index. Step 5: Multimodal feature fusion and stress value calculation. Construct a multidimensional feature vector, perform dynamic baseline adaptive calibration, establish an initial baseline for the first time and update it using a rolling baseline update formula, establish a diurnal rhythm baseline curve for correction, use an ensemble learning stress regression model to standardize the features and calculate the static stress score and time-series stress evolution score, and obtain the final stress index through attention-weighted fusion and normalization. Step 6: Temporal evolution modeling and early warning. A time-series prediction model is used to input the historical stress index sequence and output the predicted future stress value. The normal stress pattern is trained based on the isolated forest algorithm and abnormal deviations are detected in real time. When the abnormal score exceeds the preset threshold, an early warning is triggered. The chronic stress accumulation index is calculated. When the chronic stress accumulation index exceeds the preset threshold, a chronic stress risk is indicated.

[0031] It should be noted that the reference Figure 5 As shown, the multi-scale differential thresholding method for pulse wave peak detection in step three specifically includes: Calculate the first-order difference of the preprocessed rPPG signal; Find the zero-crossing point where the first-order difference changes from positive to negative within the sliding window; Set an adaptive threshold, which is the sum of the mean difference within the window and the coefficient multiplied by the standard deviation of the difference within the window; Verify the minimum physiological interval between detected peaks and remove spurious peaks that do not meet physiological constraints; The pulse interval sequence is obtained by subtracting adjacent peak timestamps. Outlier removal is performed on the pulse interval sequence. When the difference between the pulse interval value and the median exceeds a preset multiple of the absolute deviation of the median, it is marked as an outlier and replaced with linear interpolation.

[0032] It should be noted that the reference Figure 6 As shown, the training process of the ensemble learning stress regression model in step five includes: Data collection and labeling: Recruit subjects to conduct stress tests and simultaneously collect salivary cortisol samples as a stress standard label; Feature standardization involves standardizing multidimensional feature vectors. Train a random forest regressor, set the number of decision trees and maximum depth, optimize hyperparameters using cross-validation, and output a static stress score; Training a temporal model of a Long Short-Term Memory Network: Input a stress feature sequence, set the hidden layer dimension and the number of network layers, and output a temporal stress evolution score. A temporal attention mechanism is used to calculate attention weights and perform weighted fusion of static stress scores and temporal stress evolution scores. The model output is normalized, and the fused score is limited to a preset range to obtain the stress index. Model validation: Cross-validation was used to verify the correlation between stress index and cortisol and the accuracy of stress level determination.

[0033] It should be noted that the reference Figure 7 As shown, the calculation of the Chronic Stress Accumulation Index in step six specifically includes: The pressure index is discretized and sampled using a preset time unit to obtain a time series; Calculate the pressure deviation value at each time point. The pressure deviation value is the difference between the pressure index at that time point and the dynamic baseline value for the corresponding period. The daily chronic stress accumulation index is obtained by approximating the deviation value through time integration. The cumulative index is obtained by summing up the daily chronic stress cumulative index over multiple consecutive days; Set risk thresholds. When the cumulative chronic stress index exceeds the first threshold, it is marked as a high stress load for the day. When the cumulative index exceeds the second threshold, a chronic stress risk is indicated. When the cumulative index exceeds the third threshold, a high-risk warning is triggered.

[0034] According to embodiments of the present invention, a computing device that can be used to implement the above method includes a processor and a memory; The processor can be a multi-core processor or include multiple processors. In some embodiments, the processor may include a general-purpose main processor and one or more special coprocessors, such as a graphics processing unit (GPU), a digital signal processor (DSP), etc. In some embodiments, the processor may be implemented using custom circuitry, such as an application-specific integrated circuit (ASIC) or a field-programmable gate array (FPGA).

[0035] Memory can include various types of storage units, such as system memory, read-only memory (ROM), and permanent storage devices. ROM can store static data or instructions required by the processor or other modules of the computer. Permanent storage devices can be read-write storage devices. Permanent storage devices can be non-volatile storage devices that retain stored instructions and data even when the computer is powered off. In some embodiments, permanent storage devices use mass storage devices (e.g., magnetic or optical disks, flash memory) as permanent storage devices. In other embodiments, permanent storage devices can be removable storage devices (e.g., floppy disks, optical drives). System memory can be read-write storage devices or volatile read-write storage devices, such as dynamic random access memory. System memory can store some or all of the instructions and data required by the processor during operation. Furthermore, memory can include any combination of computer-readable storage media, including various types of semiconductor memory chips (DRAM, SRAM, SDRAM, flash memory, programmable read-only memory), and disks and / or optical disks can also be used. In some implementations, the memory may include removable storage devices that are readable and / or writable, such as laser discs (CDs), read-only digital versatile optical discs (e.g., DVD-ROMs, dual-layer DVD-ROMs), read-only Blu-ray discs, ultra-high density optical discs, flash memory cards (e.g., SD cards, mini SD cards, Micro-SD cards, etc.), magnetic floppy disks, etc. Computer-readable storage media do not include carrier waves and transient electronic signals transmitted wirelessly or via wired connections.

[0036] It should be understood that, unless otherwise expressly stated herein, there is no strict order restriction on the execution of the above steps, and these steps may be executed in other orders. Moreover, at least some steps in the processes involved in the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but may be executed at different times. The execution order of these steps or stages is not necessarily sequential, but may be performed alternately or in turn with other steps or at least some of the steps or stages in other steps.

[0037] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or basic characteristics. Therefore, the embodiments should be considered illustrative and non-limiting in all respects. The scope of the invention is defined by the appended claims rather than the foregoing description. Therefore, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention, and no reference numerals in the claims should be construed as limiting the scope of the claims.

[0038] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment includes only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A non-contact physiological pressure monitoring system, characterized in that, include: The multimodal signal acquisition module is used to simultaneously acquire the multispectral rPPG signal, thermal imaging signal and millimeter-wave radar signal of the subject, and to synchronize the timestamps of each acquisition module through the PTP precise time protocol; The signal preprocessing module, connected to the multimodal signal acquisition module, is used to perform time alignment, noise filtering, baseline drift correction, and motion artifact elimination on the acquired signals. The baseline drift removal employs an adaptive wavelet threshold denoising method, which performs multi-layer wavelet decomposition on the original rPPG signal, reconstructs the signal after setting a dynamic threshold, and retains the effective frequency band of the pulse wave. The motion artifact removal employs a fusion algorithm of blind source separation and independent component analysis to construct a multi-region rPPG signal matrix. After calculating the covariance matrix, whitening is performed, and independent component analysis is used to extract independent components. Based on the kurtosis criterion, pulse wave components are identified and retained before reconstructing the artifact-free signal. The feature extraction and fusion analysis module is connected to the signal preprocessing module and is used to extract time-domain features, frequency-domain features, nonlinear features and HPA axis indirect characterization features from the preprocessed signal. The stress value quantification module, connected to the feature extraction and fusion analysis module, is used to calculate the stress index based on dynamic baseline adaptive calibration and ensemble learning model; The output and early warning interaction module is connected to the pressure value quantification module and is used to output the pressure level determination result and the time-series evolution early warning information.

2. The non-contact physiological pressure monitoring system according to claim 1, characterized in that, The multimodal signal acquisition module includes: The multispectral rPPG acquisition unit is used to simultaneously acquire pulse wave signals in the green, red and near-infrared bands, and independently extract signals from the forehead, left cheek, right cheek and the area around the nose. The thermal imaging acquisition unit is used to acquire temperature change signals in the nose tip, around the eyes, and cheek areas. The millimeter-wave radar acquisition unit is used to acquire thoracic cavity micro-motion signals using the frequency-modulated continuous wave band. An ambient light compensation unit is used to correct ambient light interference in conjunction with an ambient light sensor.

3. The non-contact physiological pressure monitoring system according to claim 2, characterized in that, The time-domain features include the mean pulse interval, the standard deviation of the pulse interval, the root mean square of the difference between adjacent pulse intervals, the proportion of the difference between adjacent pulse intervals exceeding a preset threshold, the pulse interval triangular index, and the stress volatility. The frequency domain feature extraction uses the periodogram method to calculate the power of the extremely low frequency, low frequency and high frequency bands, and calculates the low frequency to high frequency ratio, normalized low frequency, normalized high frequency and stress spectrum index. The nonlinear features include sample entropy, multi-scale entropy, detrended fluctuation analysis scaling index, Poincaré plot short-to-long axis ratio, and recursive quantitative analysis parameters. The indirect characterization features of the HPA axis include the skin sympathetic response index, the HPA expression index, the respiratory-heart rate coupling index, and the comprehensive indirect index of the HPA axis.

4. The non-contact physiological pressure monitoring system according to claim 3, characterized in that, The stress volatility rate is the ratio of the standard deviation of the stress state window to the standard deviation of the baseline state window. The stress spectrum index is the ratio of the stress-state low-frequency power to the baseline low-frequency power divided by the ratio of the stress-state high-frequency power to the baseline high-frequency power. The skin sympathetic response index is a negative value of the rate of change in nasal temperature. The HPA expression index is the product of the duration, frequency, and intensity coefficient of micro-expressions; The respiratory-heart rate coupling index is the coherence between the respiratory signal spectrum and the heart rate variability spectrum in a preset frequency band. The HPA axis comprehensive indirect index is a weighted sum of the skin sympathetic response index, HPA expression index, respiratory-heart rate coupling index, and low-frequency-high-frequency ratio.

5. The non-contact physiological pressure monitoring system according to claim 4, characterized in that, The pressure value quantification module performs the following processing: Dynamic baseline adaptive calibration establishes an initial baseline upon first use. The initial baseline is a combination of the median and the absolute deviation of the median of the eigenvalues. It is updated using a rolling baseline update formula, and a diurnal rhythm baseline curve is established for rhythm correction. Feature vector construction combines time-domain features, frequency-domain features, nonlinear features, and HPA axis indirect characterization features into a multi-dimensional feature vector. The stress index is calculated using an ensemble learning stress regression model, which includes a random forest regressor, a long short-term memory network temporal model, and a temporal attention mechanism. After standardizing the features, static stress scores and temporal stress evolution scores are calculated. The final stress index is obtained by attention-weighted fusion and then normalized. The stress level assessment divides the stress index into multiple levels, corresponding to a relaxed state, mild stress, moderate stress, high stress, and extreme stress.

6. The non-contact physiological pressure monitoring system according to claim 5, characterized in that, The output and early warning interaction module includes: The pressure trend prediction unit uses a time-series prediction model, takes historical pressure index sequences as input, and outputs pressure prediction values ​​for a preset time period in the future. The abnormal stress event detection unit trains a normal stress pattern based on the isolated forest algorithm, detects abnormal deviations in real time, and triggers an early warning when the abnormal score exceeds a preset threshold. The Chronic Stress Accumulation Index (CSCI) calculation unit calculates the CSCI as the integral of the difference between the stress index and the baseline over time. When the CSCI exceeds a preset threshold, it indicates a risk of chronic stress.

7. A non-contact physiological stress monitoring method, applicable to the non-contact physiological stress monitoring system according to any one of claims 1-6, characterized in that, The method includes the following steps: Step 1: Synchronous acquisition of multimodal non-contact signals. Multi-band pulse wave signals are acquired synchronously through the multispectral rPPG acquisition module, facial temperature change signals are acquired through the thermal imaging acquisition module, and chest cavity micro-motion signals are acquired through the millimeter-wave radar acquisition module. The time synchronization protocol is used to synchronize the timestamps of each module. Step 2, signal preprocessing and quality assessment: baseline drift removal is performed on the acquired signal, denoising is carried out using wavelet decomposition and adaptive thresholding, motion artifact elimination is performed, and after constructing a multi-region signal matrix, pulse wave components are extracted using independent component analysis algorithm, and signal quality index is calculated. When the signal quality index is lower than the preset threshold, it is marked as low quality data. Step 3: Pulse interval sequence feature extraction. The multi-scale differential thresholding method is used to detect the peak value of the pulse wave, calculate the first-order difference of the signal and find the zero-crossing point within the sliding window, set an adaptive threshold to verify the physiological interval between peaks, output the pulse interval sequence, and extract time-domain features, frequency-domain features and nonlinear features from the pulse interval sequence. Step 4: Non-contact indirect characterization of HPA axis activation level. The skin sympathetic response index is calculated based on the rate of change of nasal temperature in thermal imaging. Stress-related micro-expressions are detected through facial video analysis and the HPA expression index is calculated. Respiratory signals are extracted synchronously from rPPG and radar signals and the respiratory-heart rate coupling index is calculated. The comprehensive indirect index of HPA axis is calculated. Step 5: Multimodal feature fusion and stress value calculation. Construct a multidimensional feature vector, perform dynamic baseline adaptive calibration, establish an initial baseline for the first time and update it using a rolling baseline update formula, establish a diurnal rhythm baseline curve for correction, use an ensemble learning stress regression model to standardize the features and calculate the static stress score and time-series stress evolution score, and obtain the final stress index through attention-weighted fusion and normalization. Step 6: Temporal evolution modeling and early warning. A time-series prediction model is used to input the historical stress index sequence and output the predicted future stress value. The normal stress pattern is trained based on the isolated forest algorithm and abnormal deviations are detected in real time. When the abnormal score exceeds the preset threshold, an early warning is triggered. The chronic stress accumulation index is calculated. When the chronic stress accumulation index exceeds the preset threshold, a chronic stress risk is indicated.

8. The non-contact physiological pressure monitoring method according to claim 7, characterized in that, The multi-scale differential thresholding method for pulse wave peak detection in step three specifically includes: Calculate the first-order difference of the preprocessed rPPG signal; Find the zero-crossing point where the first-order difference changes from positive to negative within the sliding window; Set an adaptive threshold, which is the sum of the mean difference within the window and the coefficient multiplied by the standard deviation of the difference within the window; Verify the minimum physiological interval between detected peaks and remove spurious peaks that do not meet physiological constraints; The pulse interval sequence is obtained by subtracting adjacent peak timestamps. Outlier removal is performed on the pulse interval sequence. When the difference between the pulse interval value and the median exceeds a preset multiple of the absolute deviation of the median, it is marked as an outlier and replaced with linear interpolation.

9. The non-contact physiological stress monitoring method according to claim 7, characterized in that, The training process of the ensemble learning stress regression model in step five includes: Data collection and labeling: Recruit subjects to conduct stress tests and simultaneously collect salivary cortisol samples as a stress standard label; Feature standardization involves standardizing multidimensional feature vectors. Train a random forest regressor, set the number of decision trees and maximum depth, optimize hyperparameters using cross-validation, and output a static stress score; Training a temporal model of a Long Short-Term Memory Network: Input a stress feature sequence, set the hidden layer dimension and the number of network layers, and output a temporal stress evolution score. A temporal attention mechanism is used to calculate attention weights and perform weighted fusion of static stress scores and temporal stress evolution scores. The model output is normalized, and the fused score is limited to a preset range to obtain the stress index. Model validation: Cross-validation was used to verify the correlation between stress index and cortisol and the accuracy of stress level determination.

10. The non-contact physiological pressure monitoring method according to claim 7, characterized in that, The calculation of the chronic stress accumulation index in step six specifically includes: The pressure index is discretized and sampled using a preset time unit to obtain a time series; Calculate the pressure deviation value at each time point. The pressure deviation value is the difference between the pressure index at that time point and the dynamic baseline value for the corresponding period. The daily chronic stress accumulation index is obtained by approximating the deviation value through time integration. The cumulative index is obtained by summing up the daily chronic stress cumulative index over multiple consecutive days; Set risk thresholds. When the cumulative chronic stress index exceeds the first threshold, it is marked as a high stress load for the day. When the cumulative index exceeds the second threshold, a chronic stress risk is indicated. When the cumulative index exceeds the third threshold, a high-risk warning is triggered.

Images