A Data / Knowledge Collaborative-Driven Method and Apparatus for Pulse Graph Mixture Gaussian Feature Extraction

Abstract

This application provides a data / knowledge collaboratively driven method and apparatus for pulse map mixture Gaussian feature extraction, relating to the field of data processing. The method includes: acquiring a PPG signal; the PPG signal refers to a photoplethysmography (PPG) signal; extracting structural features based on the PPG signal using a Gaussian mixture model; extracting physiological features based on physiological feature points of the main wave, reflected wave, and diphtheria wave in the PPG signal; the physiological features have clear physiological significance; constructing a constraint space based on the distribution range of the physiological features, using the constraint space as the boundary condition for optimizing the parameters of the Gaussian mixture model, thus obtaining an optimized Gaussian mixture model; and extracting pulse map mixture Gaussian features of the PPG signal based on the optimized Gaussian mixture model. This application, used in the data / knowledge collaboratively driven pulse map mixture Gaussian feature extraction process, solves the technical problem of existing technologies struggling to balance the effectiveness and interpretability of PPG signal feature extraction.

Description

Technical Field

[0001] This application relates to the field of data processing, and in particular to a data / knowledge collaborative-driven pulse graph mixture Gaussian feature extraction method and apparatus. Background Technology

[0002] Current pulse wave feature extraction methods primarily rely on the physiological mechanisms of photoplethysmography (PPG) waveforms. These methods extract physiologically significant parameters by identifying feature points within the waveform. While offering good interpretability, these methods require high waveform quality and struggle to comprehensively reflect the overall signal characteristics. Signal processing and statistical analysis techniques can capture global signal features but lack interpretability. Deep learning models, such as convolutional neural networks or long short-term memory networks, are used to address the challenges of PPG signal feature extraction, learning feature representations directly from the raw PPG signal in an end-to-end manner. Although these deep learning models demonstrate excellent feature extraction performance, as typical "black box" models, their internal feature representations lack clear physiological interpretation, hindering practical application. Therefore, a novel PPG signal feature extraction method is urgently needed to address the technical challenge of balancing effectiveness and interpretability in existing PPG signal feature extraction techniques. Summary of the Invention

[0003] This application provides a data / knowledge collaborative-driven pulse map mixed Gaussian feature extraction method and apparatus, which solves the technical problem that existing technologies struggle to balance the effectiveness and interpretability of PPG signal feature extraction.

[0004] To achieve the above objectives, this application adopts the following technical solution: Firstly, a data / knowledge collaboratively driven method for extracting pulse map mixture Gaussian features is provided, comprising: acquiring PPG signals; PPG signals refer to photoplethysmography (PPG) signals; extracting structural features based on PPG signals using a Gaussian mixture model; extracting physiological features based on physiological feature points of the main wave, reflected wave, and diphtheria wave in the PPG signals; the physiological features have clear physiological significance; constructing a constraint space based on the distribution range of the physiological features, and using the constraint space as the boundary condition for optimizing the parameters of the Gaussian mixture model to obtain an optimized Gaussian mixture model; and extracting pulse map mixture Gaussian features of the PPG signals based on the optimized Gaussian mixture model.

[0005] In conjunction with the first aspect mentioned above, in one possible implementation, structural features are extracted based on the PPG signal using a Gaussian mixture model, including: treating the PPG signal as a two-dimensional continuous function, discretizing the PPG signal in time to obtain a time-discrete PPG signal; simulating the time-discrete PPG signal using a Gaussian mixture model containing three Gaussian distributions, and solving for the Gaussian parameter vector to obtain the structural features.

[0006] In conjunction with the first aspect mentioned above, in one possible implementation, physiological features are extracted based on the physiological feature points of the main wave, reflected wave, and dicrotic wave in the PPG signal. These features include: obtaining the systolic peak feature point of the main wave, the reflection peak feature point of the reflected wave, and the diastolic peak feature point of the dicrotic wave based on the PPG signal and its fourth derivative; and extracting the time from the onset of the PPG signal to the systolic peak, the time from the onset of the PPG signal to the reflected wave, the time from the onset of the PPG signal to the diastolic peak, the duration of systole, the duration of diastole, the amplitude of the systolic peak, the amplitude of the reflected wave, and the amplitude of the diastolic wave as physiological features based on the PPG signal, the systolic peak feature point, the reflection peak feature point, and the diastolic peak feature point.

[0007] In conjunction with the first aspect mentioned above, in one possible implementation, a constraint space is constructed based on the distribution range of physiological characteristics, and the constraint space is used as the boundary condition for optimizing the parameters of the Gaussian mixture model to obtain the optimized Gaussian mixture model. This includes: performing statistical analysis on physiological characteristics, calculating the distribution range of characteristic point amplitudes and time parameters, constructing a constraint space, and using the constraint space as the boundary condition for optimizing the parameters of the Gaussian mixture model to obtain the optimized Gaussian mixture model.

[0008] In one possible implementation, the optimized Gaussian function of the optimized Gaussian mixture model is used. Satisfy the following formula:

[0009] Where t is the input variable; The first optimized Gaussian function controls the amplitude. The mean of the first optimized Gaussian function, The standard deviation of the first optimized Gaussian function; The second optimization is to control the amplitude of the Gaussian function. The mean of the second optimized Gaussian function, The standard deviation of the second optimized Gaussian function; The third optimization is to control the amplitude of the Gaussian function. The mean of the third optimized Gaussian function, The standard deviation is the third optimized Gaussian function.

[0010] In conjunction with the first aspect mentioned above, in one possible implementation, optimizing the Gaussian mixture function further includes: solving for the Gaussian function vector. The Gaussian parameter vector is solved using a residual optimization model constructed by the least squares method. The residual optimization model satisfies the following formula:

[0011] in, The optimal parameters are those that minimize the error. The input is the PPG signal observation value at time t. Let be the predicted value of the PPG signal at point t using the Gaussian mixture model. This indicates the search for the parameter w that minimizes the objective function. Represents the parameter vector Belongs to the constraint set , This is a constraint space constructed based on physiological characteristics.

[0012] In conjunction with the first aspect mentioned above, in one possible implementation, before extracting structural features based on the PPG signal using a Gaussian mixture model, the method further includes: extracting a single-cycle signal based on the PPG signal; performing quality assessment on the single-cycle signal to obtain a high-quality single-cycle signal; and performing baseline removal processing on the high-quality single-cycle signal to obtain a processed PPG signal.

[0013] In conjunction with the first aspect mentioned above, in one possible implementation, to evaluate the quality of a single-cycle signal and obtain a high-quality single-cycle signal, the quality of the single-cycle signal is quantified using the perfusion signal quality index, and high-quality single-cycle signals are selected by using a preset threshold.

[0014] In conjunction with the first aspect mentioned above, in one possible implementation, after extracting the pulse map mixture Gaussian features of the PPG signal based on the optimized Gaussian mixture model, the method further includes: using a deep neural network and a support vector machine to predict blood pressure based on the pulse map mixture Gaussian features.

[0015] Secondly, a data / knowledge collaborative driven pulse graph hybrid Gaussian feature extraction device is provided, including: a communication unit and a processing unit; The system includes a communication unit for acquiring PPG signals; a processing unit for extracting structural features from PPG signals using a Gaussian mixture model; extracting physiological features from the physiological feature points of the main wave, reflected wave, and diphtheria wave in the PPG signals; these physiological features have clear physiological significance; constructing a constraint space based on the distribution range of the physiological features, and using the constraint space as the boundary condition for optimizing the parameters of the Gaussian mixture model to obtain an optimized Gaussian mixture model; and extracting pulse map mixture Gaussian features from the PPG signals based on the optimized Gaussian mixture model.

[0016] This application provides a data / knowledge-driven method and apparatus for extracting pulse map features using a mixture of Gaussian features. By synergistically fusing structural and physiological features, it effectively improves the effectiveness and interpretability of PPG pulse map feature extraction. On one hand, the optimized Gaussian mixture model can automatically mine global structural features from the original PPG signal, ensuring its ability to express complex waveform structures. On the other hand, physiological features constructed based on physiological feature points such as the main wave, reflected wave, and dicrotic wave have clear physiological meaning. Their distribution intervals form a constraint space and participate in the optimization of Gaussian model parameters, enabling the model to maintain its fitting ability while avoiding physiologically meaningless outliers, thus achieving an organic combination of data-driven and knowledge-driven approaches. Through this constraint optimization mechanism, the constructed mixture of Gaussian models can more stably capture key morphological features of the pulse map, enhance the model's generalization ability, and improve the robustness and reliability of pulse map feature extraction. This provides higher-quality feature input for subsequent tasks such as blood pressure prediction, solving the technical problem of existing technologies struggling to balance the effectiveness and interpretability of PPG signal feature extraction.

[0017] It should be understood that the descriptions of technical features, technical solutions, beneficial effects, or similar language in this application do not imply that all features and advantages can be achieved in any single embodiment. Rather, it is understood that the description of a feature or beneficial effect means that a specific technical feature, technical solution, or beneficial effect is included in at least one embodiment. Therefore, the descriptions of technical features, technical solutions, or beneficial effects in this specification do not necessarily refer to the same embodiment. Furthermore, the technical features, technical solutions, and beneficial effects described in this embodiment can be combined in any suitable manner. Those skilled in the art will understand that embodiments can be implemented without one or more specific technical features, technical solutions, or beneficial effects of a particular embodiment. In other embodiments, additional technical features and beneficial effects may be identified in specific embodiments that do not embody all embodiments. Attached Figure Description

[0018] Figure 1 A schematic flowchart illustrating a data / knowledge collaborative-driven pulse graph mixture Gaussian feature extraction method provided in this application embodiment; Figure 2 A flowchart illustrating another data / knowledge collaborative-driven pulse graph mixture Gaussian feature extraction method provided in this application embodiment; Figure 3 This application provides a Gaussian mixture model fitting piecewise line graph for a data / knowledge collaborative-driven pulse graph mixture Gaussian feature extraction method. Figure 4 A flowchart illustrating another data / knowledge collaborative-driven pulse graph mixture Gaussian feature extraction method provided in this application embodiment; Figure 5 This application provides a data / knowledge collaborative-driven pulse graph mixture Gaussian feature extraction method for PPG signals and fourth derivatives using a piecewise linear plot. Figure 6 A flowchart illustrating another data / knowledge collaborative-driven pulse graph mixture Gaussian feature extraction method provided in this application embodiment; Figure 7 A schematic diagram of a data / knowledge collaborative-driven pulse map mixture Gaussian feature extraction device provided in this application embodiment; Figure 8 A flowchart illustrating the feature extraction process of the pulse graph mixture Gaussian feature extraction device based on data / knowledge collaborative driving provided in this application embodiment. Detailed Implementation

[0019] In the description of this application, unless otherwise stated, " / " means "or," for example, A / B can mean A or B. The "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone. Furthermore, "at least one" means one or more, and "multiple" means two or more. The terms "first," "second," etc., do not limit the quantity or order of execution, and "first," "second," etc., do not necessarily imply differences.

[0020] It should be noted that, in this application, the terms "exemplary" or "for example" are used to indicate that something is being described as an example, illustration, or illustration. Any embodiment or design described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or design solutions. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0021] To address the technical challenge of balancing effectiveness and interpretability in PPG signal feature extraction in existing technologies, this application provides a data / knowledge collaboratively driven pulse map mixture Gaussian feature extraction method. This method includes: acquiring a PPG signal; the PPG signal refers to a photoplethysmography (PPG) signal; extracting structural features based on the PPG signal using a Gaussian mixture model; extracting physiological features based on physiological feature points of the main wave, reflected wave, and diphtheria wave in the PPG signal; the physiological features have clear physiological significance; constructing a constraint space based on the distribution range of the physiological features, using the constraint space as the boundary condition for optimizing the Gaussian mixture model parameters to obtain an optimized Gaussian mixture model; and extracting pulse map mixture Gaussian features of the PPG signal based on the optimized Gaussian mixture model.

[0022] Figure 1This is a flowchart illustrating the data / knowledge collaborative-driven pulse graph mixture Gaussian feature extraction method provided in the embodiments of this application, as shown below. Figure 1 As shown, the method includes: S101, Acquire PPG signal.

[0023] Among them, PPG signal refers to photoplethysmography signal. PPG signal is a sequence of light intensity reflection or transmission of blood volume over time measured by optical sensors at the fingertips, earlobes, or wrists. The PPG signal data in this application is obtained from the publicly available standard PPG-BP database.

[0024] In one possible implementation, the PPG signal is acquired. Raw PPG can be acquired using a sensing module containing a light source (red / infrared) and a photodiode, with the subject's consent. The signal is first amplified and anti-aliasing filtered by an analog front-end, then sampled by an ADC and stored in a buffer. The device or acquisition system simultaneously records the timestamp, sampling rate, and sensor position for subsequent alignment and analysis.

[0025] It should be noted that PPG signals are susceptible to various artifacts during acquisition, primarily including high-frequency noise and low-frequency baseline drift. These interferences can cause signal distortion and affect the accuracy of subsequent analysis. To improve signal quality, a Savitzky-Golay filter is used to smooth the PPG signal during preprocessing, suppressing noise while preserving high-frequency characteristics through local polynomial fitting. After acquiring the PPG signal, it is processed to extract high-quality single-cycle signals. Specifically: single-cycle signals are extracted based on the PPG signal; the quality of the single-cycle signals is assessed to obtain high-quality single-cycle signals; baseline removal is performed on the high-quality single-cycle signals to obtain the processed PPG signal. The quality assessment of the single-cycle signals to obtain high-quality single-cycle signals is achieved by quantifying the quality of the single-cycle signals using the perfusion signal quality index and filtering high-quality single-cycle signals using a preset threshold.

[0026] As an example, in an embodiment of this application, the perfusion signal quality index PSQI satisfies the following formula:

[0027] in, The maximum value of the filtered PPG signal within each period. The minimum value of the filtered PPG signal within each period. This represents the mean of the original PPG signal. Used to ensure that the denominator is positive.

[0028] It should also be noted that when The waveform integrity is significantly improved, so this application uses this threshold for high-quality single-cycle signal screening.

[0029] Based on the steps described above, this step ensures the provision of high-quality PPG raw data with time synchronization information, noise labeling, and metadata annotation, thus providing reliable input for subsequent model-based feature extraction. Simultaneously, to reduce contamination from low-frequency baseline drift, a signal quality screening and periodic baseline correction method are employed. This method subtracts the baseline from the raw PPG signal to obtain the corrected signal, effectively eliminating low-frequency interference while preserving key physiological features, thereby laying a solid foundation for subsequent feature extraction and analysis.

[0030] S102. Based on PPG signals, structural features are extracted using a Gaussian mixture model.

[0031] In this context, the Gaussian mixture model is used to approximate the PPG waveform of a single cardiac cycle as a linear superposition of several Gaussian functions representing the main peak, reflection peak, and diastolic peak, respectively. The model parameters include the amplitude, center position, and width of each Gaussian distribution.

[0032] In one possible implementation, the processed PPG signal is periodically segmented, time-series normalized and aligned with the baseline for each period, and then the parameter vectors of the three Gaussian components are solved using least squares or weighted least squares as the objective function and constrained gradient descent.

[0033] It should be noted that, in order to avoid local optima or parameter swapping problems, initialization can be performed using physiological priors in the first half, middle, and second half of the cycle or based on empirical initialization of peak detection. In addition, parameter order constraints and regularization terms are added during the optimization process to ensure physical interpretability.

[0034] Based on the steps described above, this step, considering the periodicity of the PPG signal, ensures that each data point in the acquired waveform corresponds to a specific physiological state and thus has clear physiological significance. The PPG is treated as a two-dimensional continuous function image, and a Gaussian mixture model is used for parametric modeling to characterize its geometric features. Through parametric geometric modeling, the original waveform is compressed into a low-dimensional, highly interpretable, and numerically stable feature representation, thereby improving the computational efficiency and interpretability of subsequent pattern recognition, population statistics, and long-term trend analysis.

[0035] S103. Based on the physiological feature points of the main wave, reflected wave and diphtheria wave in the PPG signal, extract physiological features.

[0036] Among them, physiological characteristics have clear physiological significance.

[0037] In one possible implementation, firstly, the first to fourth order differentials or filtered derivatives of the PPG within the preprocessed period are calculated to enhance the notch and secondary peak structure; then, the main peak, reflection peak and diabetic notch are located in the joint feature space of the original signal and the fourth derivative by using threshold and neighborhood extremum rules, and the corresponding amplitudes and time intervals are calculated. If necessary, interpolation or robust estimation is used for the noise segment.

[0038] Based on the above steps, this step explicitly transforms the physiological information of PPG into interpretable parameters, enabling subsequent models not only to mathematically fit the waveforms but also to verify and interpret the results in a medical sense, thereby improving the credibility of the application evaluation.

[0039] S104. Based on the distribution range of physiological characteristics, construct a constraint space and use the constraint space as the boundary condition for optimizing the parameters of the Gaussian mixture model to obtain the optimized Gaussian mixture model.

[0040] In one possible implementation, the distribution statistics of physiological characteristics are first calculated, and these statistical boundaries are mapped to the feasible region of the Gaussian mixture model parameters. During optimization, hard or soft constraints are added to the objective function, and the constrained optimal parameters are obtained using a constrained optimization algorithm.

[0041] It should be noted that the constraint space should take into account individual differences and differences in measurement conditions: on the one hand, a general global constraint can be constructed to ensure the physiological rationality of the baseline; on the other hand, individualized adjustments should be supported to avoid excessive constraints leading to fitting bias.

[0042] Based on the above steps, this step integrates data-driven fitting with physiological knowledge by transforming prior knowledge into explicit mathematical constraints and incorporating them into the parameter estimation process. This results in a model output that reduces physiologically unreasonable solutions and enhances support for clinical interpretability.

[0043] S105. Based on the optimized Gaussian mixture model, extract the pulse map mixture Gaussian features of the PPG signal.

[0044] In one possible implementation, a constrained optimizer is used for each cardiac cycle to find the final parameters of the optimized Gaussian mixture model, resulting in a fitted optimized Gaussian mixture model, and features are extracted.

[0045] As an example, in the embodiments of this application, Table 1 shows the pulse map mixture Gaussian feature results provided in the embodiments of this application. As shown in Table 1 below, the pulse map mixture Gaussian features of 9 dimensions, from F1 to F9, are extracted based on the parameters of the optimized Gaussian mixture model, and each feature has a corresponding physiological meaning.

[0046] Table 1

[0047] Based on the above steps, the pulse map mixed Gaussian features obtained in this step have high compression ratio, good interpretability and robustness, thereby improving the reliability and usability of physiological assessment, individualized model training and clinical indicator inference based on PPG signals.

[0048] In one possible implementation, after extracting the pulse map mixture Gaussian features of the PPG signal based on an optimized Gaussian mixture model, blood pressure prediction can be performed using a deep neural network and a support vector machine based on these features. This is one of the downstream tasks of the pulse map mixture Gaussian features extracted in the embodiments of this application.

[0049] This application, based on the original PPG signal, uses a Gaussian mixture model as the data-driven core, while integrating physiologically significant features such as the main wave, reflected wave, and dicrotic wave to achieve high-precision modeling of the pulse waveform structure. This provides realistic and complete hemodynamic information for subsequent feature decomposition, enabling the extraction of stable and discriminative structural features from complex pulse waves. Furthermore, by extracting physiological feature points to form physiological features, the model becomes interpretable and effectively reflects arterial structure and blood flow reflection mechanisms. A constraint space based on the statistical range of physiological features is constructed and used as the optimization boundary for the Gaussian mixture model parameters, preventing invalid peak distributions or solutions that do not conform to the ideal pulse waveform structure, thus improving the model's stability and generalization ability. The Gaussian mixture features extracted by the optimized Gaussian mixture model possess both data sensitivity and physiological consistency, improving the quality of pulse waveform representation and providing more reliable input features for applications such as blood pressure monitoring, pulse waveform analysis, and vascular status assessment. This solves the technical problem of existing technologies struggling to balance the effectiveness and interpretability of PPG signal feature extraction.

[0050] In one possible implementation of the embodiments of this application, combined with Figure 1 ,like Figure 2 As shown, the above S102 can be implemented through the following S201 and S202, which are explained in detail below: S201. Treat the PPG signal as a two-dimensional continuous function and discretize the PPG signal in time to obtain a time-discrete PPG signal.

[0051] Time discretization refers to transforming a continuously changing pulse waveform into a series of equally spaced numerical points at a fixed sampling frequency, thus converting the pulse wave from a continuous curve into a discrete sequence that is easier to process digitally. The discrete sequence contains waveform amplitude information at each moment, serving as the basis for subsequent feature modeling.

[0052] In one possible implementation, the processed PGG signal is sampled at fixed intervals to acquire continuous data points, forming a discrete time sequence with equal step sizes. To ensure alignment and comparability between different periods, each period is remapped to a uniform length point.

[0053] It should be noted that the sampling frequency for time discretization should be reasonably selected according to the rate of change of the pulse wave. Too low a sampling frequency will result in the loss of fine waveform structure, while too high a sampling frequency will increase the computational burden. Therefore, a balance should be struck within a reasonable range according to the actual scenario.

[0054] Based on the above steps, the beneficial effect of this step is that it can convert continuous pulse wave signals into a digital sequence with regular structure and easy processing, so that subsequent waveform modeling has a unified input format and clear time reference, improves the stability of calculation and the repeatability of feature extraction, and reduces the impact of noise on the model.

[0055] S202. The time-discrete PPG signal is simulated by a Gaussian mixture model containing three Gaussian distributions, and the Gaussian parameter vector is solved to obtain the structural characteristics.

[0056] The hybrid model in this application refers to using three bell-shaped base curves to model the main peak, reflection peak, and diastolic phase of the pulse wave, simulating the complete pulse waveform by superimposing the three base curves. The model parameters include the height, position, and width of each base curve, which together describe the overall shape characteristics of the pulse wave.

[0057] In one possible implementation, the device first normalizes the discrete pulse wave sequence, and then continuously adjusts the parameters of the three base curves by minimizing the fitting error, so that the superposition of the three curves closely matches the original pulse waveform. During the parameter solving process, the parameter values ​​can be updated iteratively, maintaining a reasonable range for each update.

[0058] As an example, in an embodiment of this application, Figure 3 The Gaussian mixture model provided in the embodiments of this application is fitted with a piecewise linear plot, such as... Figure 3 As shown, according to Figure 3 The fitted curves are used to extract the structural features with the same meaning from F1 to F9 as shown in Table 1 above. However, the structural features at this time are unconstrained structural features.

[0059] Based on the steps described above, this step precisely describes the complex pulse wave structure using a small number of parameters, compressing the pulse wave from high-dimensional waveform data into low-dimensional structural features with clear meaning. This not only improves the stability and interpretability of waveform modeling but also provides a highly refined and comparable structural description for subsequent physiological analysis, statistical learning, and health assessment.

[0060] This application's embodiments, by extracting physiological features, can promptly identify representative change patterns and key morphological information after acquiring the raw signal, making the data foundation upon which subsequent analysis relies more accurate and structurally clear. This process not only reduces the workload of manually selecting feature points and improves processing efficiency, but also maintains relatively stable recognition capabilities even when the signal contains noise or there are significant individual differences, thereby enhancing the overall reliability and adaptability of the method in practical applications.

[0061] In one possible implementation of the embodiments of this application, combined with Figure 1 ,like Figure 4 As shown, the above S103 can be specifically implemented through the following S401 to S402, which are explained in detail below: S401. Based on the PPG signal and its fourth derivative, the contraction peak characteristic points of the main wave, the reflection peak characteristic points of the reflected wave, and the diastolic peak characteristic points of the diphtheria wave are obtained.

[0062] In this embodiment, the fourth derivative refers to the numerical processing result of calculating the rate of change four times consecutively based on the original pulse wave signal. It can amplify the small changes in the signal, making the peak, notch and reflected wave of the waveform more obvious, thereby facilitating the accurate location of the characteristic points of the main wave, reflected wave and diphtheria wave.

[0063] In one possible implementation, the device smooths and filters the pulse wave signal, then calculates its fourth derivative to enhance the local detail structure of the waveform. By searching for the zero intersections of the fourth derivative and local extrema, the contraction peak of the main wave, the reflection peak of the reflected wave, and the diastolic peak of the diphtheria wave are identified sequentially. The time and amplitude corresponding to each feature point are recorded to form key markers that can be used for the next feature calculation.

[0064] As an example, in a wearable monitoring device, the device takes a preprocessed pulse wave signal as input, first calculates the fourth derivative of the continuous cardiac cycle, then identifies the position and amplitude of the systolic peak, reflection peak and diastolic peak in each cycle, and finally outputs a stable set of feature points for subsequent analysis.

[0065] Based on the above steps, the beneficial effect of this step is that it can accurately identify the key peaks of the pulse wave, quantify and label the main physiological events of each cardiac cycle, thereby providing a reliable basis for physiological characteristic calculation and improving the repeatability and physiological interpretability of the characteristics.

[0066] S402. Based on PPG signals, systolic peak feature points, reflective peak feature points, and diastolic peak feature points, extract the time from the start point to the systolic peak, the time from the start point to the reflective wave, the time from the start point to the diastolic peak, the duration of systole, the duration of diastole, the amplitude of the systolic peak, the amplitude of the reflective wave, and the amplitude of the diastolic wave as physiological features.

[0067] In one possible implementation, based on the characteristic point information of each cycle, the time interval from each peak to the waveform start point is calculated sequentially, and the duration of systole and diastole is calculated based on the time difference between the systolic and diastolic peaks. Simultaneously, the amplitude of each peak is recorded, and these time and amplitude parameters are combined to form a physiological feature vector to reflect the functional characteristics of each cardiac cycle.

[0068] It should be noted that the physiological characteristics in this application embodiment refer to the time and amplitude parameters related to cardiac contraction, blood flow reflection and diastole calculated by the pulse wave and its characteristic points. These mainly include the time from the waveform start point to the systolic peak, the time to the reflection peak, the time to the diastolic peak, and the duration of the systolic and diastolic periods. In addition, they also include the amplitude parameters of each peak, which are used to quantify the cardiovascular dynamic state.

[0069] As an example, in an embodiment of this application, Figure 5 Table 2 shows the line graphs of the PPG signal and fourth derivative provided in the embodiments of this application, and the physiological feature extraction results provided in the embodiments of this application. Figure 5 As shown in Table 3, according to Figure 5 The meaning of each point is extracted, and eight physiological features from K1 to K8 are shown in Table 3. The corresponding meanings are displayed in Table 3. For example, the time from the starting point to the contraction peak is extracted as the first physiological feature, corresponding to... Figure 2 The parameter t1 in the text.

[0070] Table 2

[0071] Based on the above steps, this step transforms the key peak information of the pulse wave into quantifiable and comparable physiological characteristics, enabling an objective description of cardiac contraction, blood flow reflex, and diastolic function.

[0072] This application precisely extracts physiological features of the cardiac cycle based on the pulse wave and its key peaks, including the time interval from each peak to the start point, the duration of systole and diastole, and the amplitude parameters of each peak, thereby achieving a quantitative description of cardiac contraction, blood flow reflex, and diastolic function. This process not only improves the accuracy and stability of feature extraction but also enhances the repeatability and physiological interpretability of the features, ensuring reliable labeling of key physiological events in each cardiac cycle. This provides high-quality input data for subsequent constraint space construction, model optimization, and health assessment, enhancing the reliability and clinical usability of the entire method in practical applications.

[0073] In one possible implementation of the embodiments of this application, combined with Figure 1 ,like Figure 6As shown, the above S104 can be specifically implemented through the following S601 to S602, which are explained in detail below: S601. Perform statistical analysis on physiological characteristics, calculate the distribution range of characteristic point amplitude and time parameter, and construct a constraint space.

[0074] In one possible implementation, the collected physiological characteristic data is statistically analyzed, for example, by calculating the minimum, maximum, average and standard deviation of the amplitude and time parameters of each characteristic point, and then setting reasonable upper and lower limits to form a constraint interval.

[0075] It should be noted that the constraint space in this application embodiment refers to the set of parameter value ranges established based on the statistical results of multiple cardiac cycles and the physiological characteristics of multiple subjects. This set is used to limit the parameters in the subsequent model fitting process to a reasonable physiological range, thereby ensuring the physiological consistency of pulse wave modeling. The construction of the constraint space should consider sample diversity, including characteristic differences among different individuals, genders, ages, or heart rate states, to ensure the applicability of the model to different subjects.

[0076] As an example, in this embodiment, the boundaries of the calculated distribution intervals are shown in Table 3.

[0077] Table 3

[0078] (where p∈(p low ,p up (p∈{2%,12%,...,99%),p low =50%-p / 2,p up =50%+p / 2}, where T is the period, Tsys is the duration of systole, and Tdia is the duration of diastole.

[0079] Based on the above steps, this step transforms the statistical distribution of physiological characteristics into a constraint space, which can prevent physiologically unreasonable fitting results during the optimization of Gaussian mixture models, improve the reliability and interpretability of the model and the effectiveness of feature extraction, and provide a solid foundation for the integration of data-driven and knowledge-driven approaches.

[0080] S602. Using the constraint space as the boundary condition for optimizing the parameters of the Gaussian mixture model, the optimized Gaussian mixture model is obtained.

[0081] In one possible implementation, when modeling the discrete pulse wave sequence, the constraint space is used as the boundary condition. The height, position, and broadening parameters of each base curve are adjusted to ensure that these parameters always fall within physiologically acceptable ranges. The optimization method can employ iterative solutions or numerical optimization algorithms to automatically adjust the parameters while minimizing the residuals, and to avoid unreasonable situations such as peak misalignment, abnormal width, or negative amplitude.

[0082] It should be noted that constrained optimization not only prevents model parameters from exceeding reasonable ranges, but also improves convergence speed and fitting stability. The more precise the constraint space, the higher the consistency between the model's output pulse wave fitting results and the true waveform, while also enhancing the clinical interpretability of the data features.

[0083] As an example, in an embodiment of this application, the optimized Gaussian function of the optimized Gaussian mixture model is used. Satisfy the following formula:

[0084] Where t is the input variable; The first optimized Gaussian function controls the amplitude. The mean of the first optimized Gaussian function, The standard deviation of the first optimized Gaussian function; The second optimization is to control the amplitude of the Gaussian function. The mean of the second optimized Gaussian function, The standard deviation of the second optimized Gaussian function; The third optimization is to control the amplitude of the Gaussian function. The mean of the third optimized Gaussian function, The standard deviation is the third optimized Gaussian function.

[0085] As an example, in this embodiment of the application, optimizing the Gaussian mixture function further includes: solving for the Gaussian function vector. The Gaussian parameter vector is solved using a residual optimization model constructed by the least squares method. The residual optimization model satisfies the following formula:

[0086] in, The optimal parameters are those that minimize the error. The input is the PPG signal observation value at time t. Let be the predicted value of the PPG signal at point t using the Gaussian mixture model. This indicates the search for the parameter w that minimizes the objective function. Represents the parameter vector Belongs to the constraint set , It is a constraint space constructed based on knowledge features.

[0087] This application's embodiments statistically analyze key peak amplitude and time parameters based on the physiological characteristics of multiple cycles and individuals, constructing a reasonable constraint space, and then applying this constraint space to optimize pulse wave model parameters. In this process, the optimized model not only accurately fits discrete pulse waveforms, avoiding unreasonable results such as peak misalignment or amplitude anomalies, but also ensures that the fitted parameters fall within physiologically acceptable ranges, improving the model's stability and interpretability. Therefore, the final Gaussian mixture model possesses both high fitting accuracy and conforms to actual physiological patterns, providing reliable and repeatable effective features for subsequent tasks.

[0088] The above primarily describes the solutions of the embodiments of this application from the perspective of device implementation. It is understood that each device, such as a data / knowledge collaboratively driven pulse map mixture Gaussian feature extraction device, includes at least one of the hardware structures and software modules corresponding to the execution of each function in order to achieve the above-mentioned functions. Those skilled in the art should readily recognize that, based on the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0089] This application embodiment can divide the data / knowledge collaboratively driven pulse map mixture Gaussian feature extraction device into functional units based on the above method example. For example, each function can be divided into separate functional units, or two or more functions can be integrated into one processing unit. The integrated unit can be implemented in hardware or as a software functional unit. It should be noted that the unit division in this application embodiment is illustrative and only represents one logical functional division; other division methods may be used in actual implementation.

[0090] When using integrated units, Figure 7 A possible structural schematic diagram of the data / knowledge collaboratively driven pulse map mixture Gaussian feature extraction device (referred to as data / knowledge collaboratively driven pulse map mixture Gaussian feature extraction device 70) involved in the above embodiments is shown. The data / knowledge collaboratively driven pulse map mixture Gaussian feature extraction device 70 includes a processing unit 701 and a communication unit 702, and may also include a storage unit 703. Figure 7The schematic diagram shown can be used to illustrate the structure of the data / knowledge collaboratively driven pulse map mixture Gaussian feature extraction device involved in the above embodiments.

[0091] when Figure 7 The schematic diagram shown illustrates the structure of the data / knowledge collaboratively driven pulse map mixture Gaussian feature extraction device involved in the above embodiments. The processing unit 701 is used to control and manage the operation of the data / knowledge collaboratively driven pulse map mixture Gaussian feature extraction device, the communication unit 702 is used for the data / knowledge collaboratively driven pulse map mixture Gaussian feature extraction device to communicate with other devices, and the storage unit 703 is used to store the program code and data of the data / knowledge collaboratively driven pulse map mixture Gaussian feature extraction device.

[0092] For example, communication unit 702 is used to acquire PPG signals; Processing unit 701 extracts data features based on PPG signals using a Gaussian mixture model; extracts knowledge features based on physiological feature points of the main wave, reflected wave, and diphtheria wave in the PPG signals; the knowledge features have clear physiological significance; constructs a constraint space based on the distribution range of the knowledge features, and uses the constraint space as the boundary condition for optimizing the parameters of the Gaussian mixture model to obtain an optimized Gaussian mixture model; and extracts pulse map mixture Gaussian features of the PPG signals based on the optimized Gaussian mixture model.

[0093] In one possible implementation, the processing unit 701 is further configured to extract structural features based on the PPG signal using a Gaussian mixture model, including: treating the PPG signal as a two-dimensional continuous function, discretizing the PPG signal in time to obtain a time-discrete PPG signal; simulating the time-discrete PPG signal using a Gaussian mixture model containing three Gaussian distributions, and solving for the Gaussian parameter vector to obtain the structural features.

[0094] In one possible implementation, the processing unit 701 is further configured to extract physiological features based on the physiological feature points of the main wave, reflected wave, and dicrotic wave in the PPG signal, including: obtaining the systolic peak feature point of the main wave, the reflection peak feature point of the reflected wave, and the diastolic peak feature point of the dicrotic wave based on the PPG signal and the fourth derivative of the PPG signal; and extracting the time from the start point to the systolic peak, the time from the start point to the reflected wave, the time from the start point to the diastolic peak, the duration of systole, the duration of diastole, the amplitude of the systolic peak, the amplitude of the reflected wave, and the amplitude of the diastolic wave as physiological features based on the PPG signal, the systolic peak feature point, the reflection peak feature point, and the diastolic peak feature point.

[0095] In one possible implementation, the processing unit 701 is further configured to construct a constraint space based on the distribution range of physiological characteristics, and use the constraint space as the boundary condition for optimizing the parameters of the Gaussian mixture model to obtain an optimized Gaussian mixture model. This includes: performing statistical analysis on the physiological characteristics, calculating the distribution range of the feature point amplitude and time parameter, constructing a constraint space, and using the constraint space as the boundary condition for optimizing the parameters of the Gaussian mixture model to obtain an optimized Gaussian mixture model.

[0096] In one possible implementation, the optimized Gaussian function of the optimized Gaussian mixture model is used. Satisfy the following formula:

[0097] Where t is the input variable; The first optimized Gaussian function controls the amplitude. The mean of the first optimized Gaussian function, The standard deviation of the first optimized Gaussian function; The second optimization is to control the amplitude of the Gaussian function. The mean of the second optimized Gaussian function, The standard deviation of the second optimized Gaussian function; The third optimization is to control the amplitude of the Gaussian function. The mean of the third optimized Gaussian function, The standard deviation is the third optimized Gaussian function.

[0098] In one possible implementation, the processing unit 701, for optimizing the Gaussian mixture function, further includes: solving for the Gaussian function vector. The Gaussian parameter vector is solved using a residual optimization model constructed by the least squares method. The residual optimization model satisfies the following formula:

[0099] in, The optimal parameters are those that minimize the error. The input is the PPG signal observation value at time t. Let be the predicted value of the PPG signal at point t using the Gaussian mixture model. This indicates the search for the parameter w that minimizes the objective function. Represents the parameter vector Belongs to the constraint set , This is a constraint space constructed based on physiological characteristics.

[0100] In one possible implementation, the processing unit 701 is further configured to, before extracting structural features based on the PPG signal using a Gaussian mixture model, include: extracting a single-cycle signal based on the PPG signal; performing quality assessment on the single-cycle signal to obtain a high-quality single-cycle signal; and performing baseline removal processing on the high-quality single-cycle signal to obtain a processed PPG signal.

[0101] In one possible implementation, the processing unit 701 is further used to perform quality assessment on the single-cycle signal to obtain a high-quality single-cycle signal. This is achieved by quantifying the quality of the single-cycle signal using the perfusion signal quality index and filtering the high-quality single-cycle signal using a preset threshold.

[0102] In one possible implementation, the processing unit 701 is further configured to, after extracting the pulse map mixture Gaussian features of the PPG signal based on the optimized Gaussian mixture model, also include: using a deep neural network and a support vector machine to predict blood pressure based on the pulse map mixture Gaussian features.

[0103] As an example, in an embodiment of this application, Figure 8 The flowchart for feature extraction using the pulse graph mixture Gaussian feature extraction device based on data / knowledge collaborative driving provided in the embodiments of this application is as follows: Figure 8 As shown, the PPG signal is first preprocessed. Preprocessed PPG signal data is obtained through filtering, single-cycle processing quality assessment, and baseline removal. Next, feature extraction and fusion are performed on the preprocessed PPG signal. Data features are extracted using a Gaussian mixture model, and knowledge features are extracted through feature point recognition. Finally, the knowledge features and data features are fused using an optimized Gaussian mixture model to obtain pulse map mixture Gaussian features, which are then used for modeling and comparative analysis in subsequent tasks.

[0104] The processing unit 701 can be a processor or a controller, and the communication unit 702 can be a communication interface, transceiver, transceiver circuit, transceiver device, etc. The term "communication interface" is a general term and may include one or more interfaces. The storage unit 703 can be a memory. When the data / knowledge collaboratively driven pulse map mixture Gaussian feature extraction device 70 is a chip, the processing unit 701 can be a processor or a controller, and the communication unit 702 can be an input interface and / or an output interface, pins, or circuits, etc. The storage unit 703 can be a storage unit within the chip (e.g., a register, cache, etc.) or a storage unit located outside the chip (e.g., read-only memory (ROM), random access memory (RAM, etc.).

[0105] The communication unit can also be referred to as a transceiver unit. The antenna and control circuit with transceiver functions in the data / knowledge collaboratively driven pulse map mixture Gaussian feature extraction device 70 can be considered as the communication unit 702 of the data / knowledge collaboratively driven pulse map mixture Gaussian feature extraction device 70, and the processor with processing functions can be considered as the processing unit 701 of the data / knowledge collaboratively driven pulse map mixture Gaussian feature extraction device 70. Optionally, the device in the communication unit 702 used to implement the receiving function can be considered as a communication unit, which is used to execute the receiving steps in the embodiments of this application. The communication unit can be a receiver, a receiver circuit, etc. The device in the communication unit 702 used to implement the transmitting function can be considered as a transmitting unit, which is used to execute the transmitting steps in the embodiments of this application. The transmitting unit can be a transmitter, a transmitter, a transmitting circuit, etc.

[0106] Figure 7 If the integrated units in the process are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, in essence, or the parts that contribute to the prior art, or all or part of the technical solutions, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods of the various embodiments of this application. Storage media for storing computer software products include various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory, random access memory, magnetic disks, or optical disks.

[0107] Figure 7 The units in the process can also be called modules; for example, a processing unit can be called a processing module.

[0108] The memory in the embodiments of this application may include at least one of the following types: read-only memory (ROM) or other types of static storage devices capable of storing static information and instructions; random access memory (RAM) or other types of dynamic storage devices capable of storing information and instructions; or electrically erasable programmable-only memory (EEPROM). In some scenarios, the memory may also be a compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media, or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures that can be accessed by a computer, but is not limited thereto.

[0109] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented using software programs, implementation can be, in whole or in part, in the form of a computer program product. This computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this application is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device containing one or more servers, data centers, etc., that can be integrated with the medium. The available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state disks (SSDs)).

[0110] Although this application has been described herein in conjunction with various embodiments, those skilled in the art, by reviewing the accompanying drawings, disclosure, and appended claims, will understand and implement other variations of the disclosed embodiments in carrying out the claimed application. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude multiple instances. A single processor or other unit can implement several functions listed in the claims. While different dependent claims may recite certain measures, this does not mean that these measures cannot be combined to produce good results.

[0111] Although this application has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made thereto without departing from the spirit and scope of this application. Accordingly, this specification and drawings are merely exemplary illustrations of this application as defined by the appended claims, and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from the spirit and scope of this application. Thus, if such modifications and modifications of this application fall within the scope of the claims of this application and their equivalents, this application is also intended to include such modifications and modifications.

Claims

1. A data / knowledge collaboratively driven pulse map mixture Gaussian feature extraction method, characterized in that, include: Acquire the PPG signal; the PPG signal refers to the photoplethysmography signal. Based on the PPG signal, structural features are extracted using a Gaussian mixture model containing three Gaussian distributions; the extraction of structural features based on the PPG signal using the Gaussian mixture model includes: The PPG signal is treated as a two-dimensional continuous function, and the PPG signal is discretized in time to obtain a time-discrete PPG signal. The time-discrete PPG signal was simulated using a Gaussian mixture model containing three Gaussian distributions, and the Gaussian parameter vector was solved to obtain the structural features. Physiological features are extracted based on the physiological feature points of the main wave, reflected wave, and dicrotic wave in the PPG signal; these physiological features have clear physiological significance; the extraction of physiological features based on the physiological feature points of the main wave, reflected wave, and dicrotic wave in the PPG signal includes: Based on the PPG signal and its fourth derivative, the contraction peak feature points of the main wave, the reflection peak feature points of the reflected wave, and the diastolic peak feature points of the diphtheria wave are obtained. Based on the PPG signal, the systolic peak feature point, the reflection peak feature point, and the diastolic peak feature point, the time from the start point to the systolic peak, the time from the start point to the reflection wave, the time from the start point to the diastolic peak, the duration of systole, the duration of diastole, the amplitude of the systolic peak, the amplitude of the reflection wave, and the amplitude of the diastolic wave are extracted as physiological features. Based on the distribution range of the physiological characteristics, a constraint space is constructed. The constraint space is used as the boundary condition for the optimization of Gaussian mixture model parameters. In the optimization process, parameter order constraints and regularization terms are added to obtain the optimized Gaussian mixture model. Based on the optimized Gaussian mixture model, pulse map mixture Gaussian features of the PPG signal are extracted.

2. The method according to claim 1, characterized in that, Based on the distribution interval of the physiological characteristics, a constraint space is constructed. This constraint space is used as the boundary condition for optimizing the parameters of the Gaussian mixture model, resulting in an optimized Gaussian mixture model, including: Statistical analysis is performed on the physiological characteristics to calculate the distribution range of the characteristic point amplitude and time parameter, and a constraint space is constructed. The constraint space is used as the boundary condition for optimizing the parameters of the Gaussian mixture model, thus obtaining the optimized Gaussian mixture model.

3. The method according to claim 2, characterized in that, The optimized Gaussian function of the optimized Gaussian mixture model Satisfy the following formula: Where t is the input variable; The first optimized Gaussian function controls the amplitude. The mean of the first optimized Gaussian function, The standard deviation of the first optimized Gaussian function; The second optimization is to control the amplitude of the Gaussian function. The mean of the second optimized Gaussian function, The standard deviation of the second optimized Gaussian function; The third optimization is to control the amplitude of the Gaussian function. The mean of the third optimized Gaussian function, The standard deviation is the third optimized Gaussian function.

4. The method according to claim 3, characterized in that, The optimization of the Gaussian mixture function further includes: solving for the Gaussian function vector. The Gaussian parameter vector is solved using a residual optimization model constructed by the least squares method. The residual optimization model satisfies the following formula: in, The optimal parameters are those that minimize the error. The input is the PPG signal observation value at time t. Let be the predicted value of the PPG signal at point t using the Gaussian mixture model. This indicates the search for the parameter w that minimizes the objective function. Represents the parameter vector Belongs to the constraint set , This is a constraint space constructed based on physiological characteristics.

5. The method according to claim 1, characterized in that, Before extracting structural features based on the PPG signal using a Gaussian mixture model, the method further includes: Extract single-cycle signals based on PPG signals; The quality of the single-cycle signal is evaluated to obtain a high-quality single-cycle signal; The high-quality single-cycle signal is subjected to baseline removal processing to obtain the processed PPG signal.

6. The method according to claim 5, characterized in that, The process of evaluating the quality of the single-cycle signal to obtain a high-quality single-cycle signal involves quantifying the quality of the single-cycle signal using the perfusion signal quality index and filtering high-quality single-cycle signals using a preset threshold.

7. The method according to claim 1, characterized in that, After extracting the pulse map mixture Gaussian features of the PPG signal based on the optimized Gaussian mixture model, the method further includes: Based on Gaussian mixture features of pulse map, a deep neural network and support vector machine are used for blood pressure prediction.

8. A data / knowledge collaboratively driven pulse map mixture Gaussian feature extraction device, characterized in that, The device includes: a communication unit and a processing unit; The communication unit is used to acquire PPG signals; The processing unit is configured to extract structural features based on the PPG signal using a Gaussian mixture model containing three Gaussian distributions. The extraction of structural features based on the PPG signal using the Gaussian mixture model includes: treating the PPG signal as a two-dimensional continuous function and discretizing the PPG signal over time to obtain a time-discrete PPG signal; simulating the time-discrete PPG signal using a Gaussian mixture model containing three Gaussian distributions and solving for the Gaussian parameter vector to obtain structural features; and extracting physiological features based on the physiological feature points of the main wave, reflected wave, and dicrotic wave in the PPG signal; the physiological features have clear physiological significance. The extraction of physiological features based on the physiological feature points of the main wave, reflected wave, and dicrotic wave in the PPG signal includes: based on the PPG signal and... The fourth derivative of the PPG signal is used to obtain the characteristic points of the contraction peak of the main wave, the reflection peak of the reflected wave, and the diastolic peak of the diphtheria wave. Based on the PPG signal, the characteristic points of the contraction peak, the reflection peak, and the diastolic peak, the time from the onset to the contraction peak, the time from the onset to the reflection wave, the time from the onset to the diastolic peak, the duration of the contraction phase, the duration of the diastolic phase, the amplitude of the contraction peak, the amplitude of the reflected wave, and the amplitude of the diastolic wave are extracted as physiological features. Based on the distribution range of the physiological features, a constraint space is constructed, and the constraint space is used as the boundary condition for the optimization of Gaussian mixture model parameters. Parameter order constraints and regularization terms are added during the optimization process to obtain the optimized Gaussian mixture model. Based on the optimized Gaussian mixture model, the pulse map mixture Gaussian features of the PPG signal are extracted.

Images