Non-contact heart rate variability analysis method and apparatus based on tasking parameter selection

Abstract

The present application relates to a non-contact heart rate variability analysis method and device based on task-based parameter selection, belonging to the technical field of radar vital sign detection. It includes: constructing a set of candidate parameters for respiration and heart rate respectively; using a respiration fitness function as the objective function, performing mixed integer Bayesian optimization on the set of candidate parameters for respiration to determine the optimal respiration parameter combination and output the respiration frequency; using the respiration frequency as the respiration harmonic prior and using a heart rate fitness function as the objective function, performing mixed integer Bayesian optimization on the set of candidate parameters for heart rate to determine the optimal heart rate parameter combination and output the heart rate frequency; using the respiration frequency and the heart rate frequency as the prior, performing heart rate variability fitness function scoring on the decomposition results of the optimal heart rate parameter combination to screen the optimal mode and generate the heart rate variability reconstruction signal; performing peak detection, peak position refinement and inter-beat interval quality control on the signal to obtain the heart rate variability index. The present application improves the accuracy of heart rate variability analysis in non-contact scenarios.

Description

Technical Field

[0001] This invention relates to the field of radar vital sign detection technology, and in particular to a non-contact heart rate variability analysis method and apparatus based on task-oriented parameter selection. Background Technology

[0002] In the field of non-contact vital sign monitoring, millimeter-wave radar, with its advantages of not requiring contact with the human body and being able to penetrate thin obstructions, is widely used in daily health assessment scenarios. Its core objective is to accurately extract key physiological parameters such as respiratory rhythm, heart rate rhythm, and heart rate variability (HRV) from radar echoes. HRV, as an important indicator reflecting the function of the autonomic nervous system, is of great significance for health status assessment. The prerequisite for accurately obtaining HRV is to first achieve effective separation and high-quality extraction of respiratory and heart rate signals, and then obtain accurate inter-cardiac interval (IBI) sequences through reliable peak position detection.

[0003] Among current mainstream technologies, Variational Mode Decomposition (VMD) has become a commonly used method for extracting respiratory and heart rate signals because it can adaptively decompose complex signals into multiple stationary modes. However, the decomposition effect of Variational Mode Decomposition depends on the selection of the number of modes and the penalty factor. Existing methods can determine VMD parameters using fixed parameters, empirical parameters, full traversal, or general optimization algorithms, but they usually still aim at the quality of a single decomposition or the selection of a single heart rate mode. They do not establish separate task adaptation functions for the three types of tasks: respiratory rate estimation, heart rate rate estimation, and heart rate variability analysis, nor do they use the respiratory task output as a respiratory harmonic prior for optimizing the heart rate task parameters.

[0004] Traditional parameter selection often focuses on general decomposition quality metrics such as reconstruction error and envelope entropy. However, these metrics do not necessarily guarantee that the decomposition results are suitable for heart rate estimation or heart rate variability calculation. In addition, fully traversing the candidate mode count and penalty factor requires performing VMD for each set of parameters, which is computationally expensive. Heart rate modes are also easily affected by respiratory harmonics or low-frequency boundary peaks. If modes are selected solely based on spectral peak values, second, third, or low-frequency trends in respiration may be misinterpreted as heart rate signals, leading to a decrease in the accuracy of HRV analysis. Summary of the Invention

[0005] Therefore, it is necessary to provide a non-contact heart rate variability analysis method and device based on task-oriented parameter selection that can adaptively match breathing, heart rate, and HRV tasks, suppress respiratory harmonics, and improve the accuracy of heart rate variability analysis, in order to address the above-mentioned technical problems.

[0006] A non-contact heart rate variability analysis method based on task-oriented parameter selection, the method comprising: Slow-time signals of micro-movements in the human chest and abdomen were acquired and preprocessed to obtain preprocessed slow-time signals. Based on the characteristics of the breathing task and the heart rate task, the range of modality number and penalty factor values ​​are set respectively, and the candidate parameter sets for breathing and heart rate are constructed. Using the respiratory fitness function as the objective function, mixed integer Bayesian optimization is performed on the respiratory candidate parameter set. Variational mode decomposition is then performed on the preprocessed slow-time signal to iteratively select combinations of respiratory candidate parameters and calculate the respiratory fitness score. Based on the respiratory fitness score, the optimal combination of respiratory parameters is determined and the respiratory rate is output. Using the respiratory frequency as the respiratory harmonic prior and the heart rate fitness function as the objective function, mixed integer Bayes optimization is performed on the heart rate candidate parameter set. The combination of heart rate candidate parameters is iteratively selected to perform variational mode decomposition on the preprocessed slow-time signal to calculate the heart rate fitness score. Based on the heart rate fitness score, the optimal combination of heart rate parameters is determined and the heart rate frequency is output. Using the respiratory frequency as the respiratory coupling prior and the heart rate frequency as the heart rate variability heart rate prior, the variational mode decomposition results corresponding to the optimal heart rate parameter combination are scored by the heart rate variability fitness function, the optimal mode is selected and the heart rate variability reconstruction signal is generated. Based on the heart rate frequency, peak detection, peak position refinement, and heart rate interval quality control are performed on the heart rate variability reconstruction signal to obtain an effective heart rate interval sequence; heart rate variability index is calculated and output based on the effective heart rate interval sequence.

[0007] On the other hand, a non-contact heart rate variability analysis device based on task-oriented parameter selection is also provided, comprising: The signal acquisition and preprocessing module is used to acquire slow-time signals of micro-movements in the human chest and abdomen, and obtain preprocessed slow-time signals after preprocessing. The parameter set construction module is used to construct candidate parameter sets for breathing and heart rate based on the characteristics of breathing and heart rate tasks, respectively, by setting the range of modality number and penalty factor values. The respiratory parameter optimization module is used to perform mixed integer Bayesian optimization on the respiratory candidate parameter set with the respiratory fitness function as the objective function, iteratively select the combination of respiratory candidate parameters to perform variational mode decomposition on the preprocessed slow-time signal, calculate the respiratory fitness score, determine the optimal combination of respiratory parameters based on the respiratory fitness score, and output the respiratory rate. The heart rate parameter optimization module is used to perform mixed integer Bayes optimization on the heart rate candidate parameter set with the respiratory frequency as the respiratory harmonic prior and the heart rate fitness function as the objective function, iteratively select the heart rate candidate parameter combination to perform variational mode decomposition on the preprocessed slow-time signal, calculate the heart rate fitness score, determine the optimal heart rate parameter combination based on the heart rate fitness score, and output the heart rate frequency. The heart rate variability reconstruction module is used to score the variational mode decomposition results corresponding to the optimal heart rate parameter combination using the respiratory frequency as the respiratory coupling prior and the heart rate frequency as the heart rate variability heart prior, to screen the optimal mode and generate a heart rate variability reconstruction signal. The heart rate variability index calculation module is used to perform peak detection, peak position refinement, and inter-beat interval quality control on the heart rate variability reconstruction signal based on the heart rate frequency to obtain an effective inter-beat interval sequence; and to calculate and output the heart rate variability index based on the effective inter-beat interval sequence.

[0008] Compared with existing technologies, the non-contact heart rate variability analysis method and device based on task-oriented parameter selection provided by this invention have the following advantages: 1. Abandoning the traditional fixed parameter or manual setting mode, we independently construct datasets and fitness functions for breathing and heart rate tasks, and use Bayesian optimization to iteratively search for the optimal number of modalities and penalty factors, reducing the number of repeated VMD decompositions caused by full traversal, and accurately matching the target frequency band and technical requirements of different tasks.

[0009] 2. A task-specific modal scoring mechanism is adopted to replace general indicators such as reconstruction error, which is in line with the low-frequency stability of respiratory tasks, the anti-respiratory harmonic capability of heart rate tasks, and the peak position accuracy requirements of HRV tasks, thereby improving the targeting and accuracy of modal screening.

[0010] 3. The heart rate task uses respiratory rate as a priori to identify respiratory harmonics, avoiding misinterpreting respiratory frequency doubling and low-frequency trends as heart rate signals, reducing the impact of interference on signal extraction from the source, and effectively suppressing respiratory harmonics and low-frequency interference.

[0011] 4. Based on the heart rate decomposition results, HRV-guided modal screening is performed, focusing on peak structure clarity and peak position accuracy. Combined with peak detection, peak position refinement and interval quality control, the accuracy of the cardiac interval sequence is ensured, which greatly improves the accuracy and robustness of heart rate variability analysis in non-contact scenarios. Attached Figure Description

[0012] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention, and those skilled in the art can obtain other related drawings based on these drawings without creative effort.

[0013] Figure 1 This is a flowchart illustrating a non-contact heart rate variability analysis method based on task-oriented parameter selection in one embodiment. Figure 2 This is a structural block diagram of a non-contact heart rate variability analysis device based on task-oriented parameter selection in one embodiment.

[0014] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0016] It is understood that the technical solutions of the various embodiments of the present invention can be combined with each other, but only if they are based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0017] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0018] Example 1 like Figure 1 As shown, a non-contact heart rate variability analysis method based on task-oriented parameter selection is provided, including the following steps: Step 201: Obtain the slow-time signal of human chest and abdomen micro-movements, and obtain the preprocessed slow-time signal after preprocessing.

[0019] Step 202: Based on the characteristics of the breathing task and the heart rate task, set the range of modality number and penalty factor values ​​respectively, and construct the breathing candidate parameter set and the heart rate candidate parameter set.

[0020] Step 203: Using the respiratory fitness function as the objective function, perform mixed integer Bayesian optimization on the respiratory candidate parameter set, iteratively select respiratory candidate parameter combinations, perform variational mode decomposition on the preprocessed slow-time signal, and calculate the respiratory fitness score; determine the optimal respiratory parameter combination based on the respiratory fitness score, and output the respiratory rate.

[0021] Step 204: Using respiratory rate as the respiratory harmonic prior and heart rate fitness function as the objective function, perform mixed integer Bayesian optimization on the heart rate candidate parameter set, iteratively select the heart rate candidate parameter combination, perform variational mode decomposition on the preprocessed slow-time signal, and calculate the heart rate fitness score; determine the optimal heart rate parameter combination based on the heart rate fitness score, and output the heart rate frequency.

[0022] Step 205: Using respiratory rate as the respiratory coupling prior and heart rate rate as the heart rate variability prior, the variational mode decomposition results corresponding to the optimal heart rate parameter combination are scored by the heart rate variability fitness function, the optimal mode is selected, and the heart rate variability reconstruction signal is generated.

[0023] Step 206: Based on heart rate frequency, perform peak detection, peak position refinement, and heart rate interval quality control on the heart rate variability reconstruction signal to obtain an effective heart rate interval sequence; calculate and output the heart rate variability index based on the effective heart rate interval sequence.

[0024] The aforementioned non-contact heart rate variability analysis method based on task-specific parameter selection avoids the band mismatch and performance degradation problems caused by traditional methods that share the same set of variational mode decomposition parameters for respiratory, heart rate, and heart rate variability analysis by constructing independent candidate parameter sets for respiratory and heart rate tasks respectively, and using a task-specific fitness function as the objective function for mixed-integer Bayesian optimization. Furthermore, by using respiratory frequency as the respiratory harmonic prior for the heart rate task and introducing a respiratory harmonic penalty factor into the heart rate modality scoring, the method effectively suppresses the risk of respiratory octaves being misclassified as heart rate components, thus improving the accuracy of heart rate variability analysis. The accuracy of heart rate frequency estimation is improved. Furthermore, using respiratory rate and heart rate rate as dual priors, a heart rate variability adaptation function containing positive reward and negative penalty is constructed, focusing on peak structure integrity, rhythm clarity, and respiratory coupling inhibition. The optimal mode most suitable for beat-by-beat peak detection is screened out, significantly improving the quality of the heart rate variability reconstruction signal. Finally, by combining peak position refinement and multi-dimensional interbeat interval quality control, the interference of respiratory harmonics, low-frequency trends, and mode misselection on interbeat interval calculation in non-contact scenarios is effectively reduced, thereby greatly improving the accuracy and robustness of heart rate variability indicators.

[0025] In step 201, the slow-time signal of human chest and abdomen micro-motion refers to the original physiological signal obtained by the millimeter-wave radar transmitting and receiving the reflected echo of human chest and abdomen micro-motion, and sampling it in the slow-time dimension, which includes micro-motion phase change information caused by respiration and heart rate; preprocessing refers to a series of operations to standardize, denoise and correct the trend of the original slow-time signal, with the aim of removing non-physiological interference and obtaining a clean basic signal.

[0026] In the specific implementation of step 201, slow-time signals of human chest and abdomen micro-motion are acquired using millimeter-wave radar, and preprocessing operations are performed sequentially. Specifically, firstly, mean removal is performed to eliminate the DC component of the signal and avoid baseline shift affecting subsequent analysis; then, a sliding window detrending algorithm is used to remove non-physiological trend terms such as low-frequency drift and motion artifacts; subsequently, outlier repair is performed to identify and replace abrupt abnormal points in the signal and suppress random spike interference; finally, segmented bandpass filtering is performed to filter out high-frequency noise and extremely low-frequency interference, ultimately obtaining the preprocessed slow-time signal.

[0027] This step, through multi-dimensional preprocessing, effectively removes non-physiological trends, outliers, and irrelevant noise from the original signal, reducing the negative impact of interference on subsequent variational mode decomposition, feature extraction, and parameter calculation, thus laying a reliable foundation for subsequent accurate signal processing.

[0028] In step 202, the respiratory candidate parameter set and the heart rate candidate parameter set refer to the sets of variational mode decomposition parameter combinations independently set for the respiratory task and the heart rate task, respectively. Each parameter combination consists of the number of modes. With penalty factor Pairing constitutes; where, the number of modes For integer variables, the penalty factor As continuous variables, the candidate sets of respiratory parameters and heart rate parameters constitute the mixed integer search spaces for the respiratory task and heart rate task, respectively.

[0029] In the specific implementation of step 202, based on the low-frequency and stable characteristics of respiratory signals, the range of modal numbers for the respiratory task is set to [value range missing]. The penalty factor ranges from 100 to 100. ,in, This represents the set of integers. Based on the characteristics of heart rate signals—high frequency, narrow bandwidth, and susceptibility to harmonic interference—the range of modal numbers for the heart rate task is set to [value range missing]. The penalty factor ranges from 100 to 100. .

[0030] Then, iterate through the integer modal number values ​​and continuous penalty factor values ​​corresponding to each task to generate all possible parameter combinations, which constitute the candidate parameter set for the breathing task. Heart rate task candidate parameter set The two sets of candidate parameters are independent of each other, and different combinations of parameters can be selected as needed, without any restrictions on sharing.

[0031] This step constructs a mixed integer search space that conforms to the physiological frequency band characteristics of each task by independently setting the range of modality number and penalty factor for the breathing task and the heart rate task, and clarifying the integer constraint of modality number and the continuous characteristic of penalty factor. This provides a precise search boundary for subsequent Bayesian optimization, avoids the computational redundancy caused by full traversal, and ensures the physical rationality of parameter search.

[0032] Before implementing steps 203 to 206, the unified common framework for parameter search and modal scoring of the breathing task, heart rate task, and heart rate variability task will be explained. When each task is executed subsequently, only the target frequency band, evaluation weight, and penalty rules need to be adjusted according to the corresponding task requirements, and the core logic of the common framework can be reused.

[0033] First, the mixed-integer Bayesian optimization process is explained. Mixed-integer Bayesian optimization is used to efficiently find the optimal number of modalities within the mixed-integer search space of breathing and heart rate tasks. With penalty factor α Combinatorial optimization. Unlike full traversal, Bayesian optimization predicts the performance of unevaluated parameter combinations by establishing surrogate models and uses acquisition functions to balance exploration and exploitation, thereby obtaining the optimal solution with fewer variational mode decompositions.

[0034] Specifically, regarding the task Define the vector of parameter combinations to be optimized as follows: (1) In the formula, Indicates breathing task; Indicates heart rate task; For the task The corresponding variational mode decomposition mode number is an integer variable; For the task The corresponding penalty factor is a continuous variable.

[0035] Task The mixed integer search space is: (2) The goal of Bayesian optimization is to Inner Maximization Task Fit Function ,Right now: (3) In the formula, Indicates task The optimal combination of parameters; For the task The task adaptation function.

[0036] The iterative process is as follows: First, randomly select within the search space The parameters are combined, and variational mode decomposition is performed on each group to calculate the task fitness score, forming a historical observation set: (4) In the formula, Indicates the first The parameter combination for this experiment; This indicates the task fitness score corresponding to this parameter combination.

[0037] Based on historical observation sets, a Gaussian process is used as a surrogate model to predict arbitrary combinations of unevaluated parameters. mean fitness score and uncertainty : (5) In the formula, Represents the task adaptation function The proxy model. Among them, The larger the value, the higher the score that the parameter combination is likely to achieve; The larger the value, the more it indicates that the area around this parameter combination has not been fully explored and is worth continuing to experiment with.

[0038] Expected Improvement (EI) is used as the acquisition function to balance utilizing the current optimal region with exploring undersampled regions. The acquisition function is defined as follows: (6) in, This is a data acquisition function used to measure parameter combinations. Does it deserve to be evaluated again? It is the highest fitness score in the current historical observation set; To explore the adjustment coefficient. When When the value is large, the algorithm tends to explore the region of parameters that are not yet known; when When the parameters are smaller, the algorithm tends to utilize the parameter range that currently appears to be optimal.

[0039] Under the Gaussian process assumption, the expected lift has an analytical expression: (7) (8) in, and These are the cumulative distribution function and probability density function of the standard normal distribution, respectively. To avoid dividing by zero for extremely small positive numbers.

[0040] Then, the acquisition function is maximized to obtain the next set of parameter combinations to be optimized: (9) Based on selection For preprocessing slow-time signals Perform variational mode decomposition and calculate the corresponding task fitness score. And add it to the historical observation set.

[0041] Furthermore, for mixed integer search spaces, this invention can be applied to integer variables. Enumerate them one by one, and in each given For continuous variables The acquisition function is maximized to obtain the next set of parameters to be evaluated: (10) Repeat the above process until the preset maximum number of iterations is reached. or continuous The improvement in the optimal fitness score in each iteration is less than the stopping threshold. Ultimately, the parameter combination with the highest fitness score was selected from the historical observation set as the task. Optimal parameter combination .

[0042] In this iterative process, based on each obtained combination of parameters to be optimized, the preprocessed slow-time signal is... Perform variational mode decomposition, the decomposition formula is: (11) In the formula, The modal number; For the first One mode; This represents the sequence number of the slow-time sampling point. The output modal matrix after decomposition is... .

[0043] For each mode in the mode matrix The fundamental characteristics are calculated. These fundamental characteristics include the main peak frequency, signal-to-noise ratio score, bandwidth purity score, frequency stability score, spectral peak width score, autocorrelation intensity score, frequency domain-time domain consistency score, and ideal bandwidth soft score.

[0044] According to the task The candidate modal scoring function calculates the candidate modal score. This score represents the modality. For the task The degree of suitability. For example, in a breathing task, a high candidate modality score indicates that the modality is more like breathing; in a heart rate task, a candidate modality score indicates that the modality is more like heart rate.

[0045] For the current combination of parameters to be optimized selected in the Bayesian optimization iteration, the highest score of all modes in its decomposition result is taken as the target mode score of the parameter combination to be optimized: (12) In the formula, The target mode score is defined as follows. This step means that for a set of VMD parameters, the combination of parameters is considered valuable as long as there is at least one high-quality target mode in its decomposition results.

[0046] The reconstruction quality score was then calculated: (13) In the formula, For extremely small positive numbers, avoid division by zero; This is the amplitude limiting function. This score is used to measure whether the sum of all modes can reconstruct the original signal well. The higher the score, the closer the sum of all modes is to the original signal.

[0047] Calculate the modal separation score: (14) In the formula, For the first One mode; For the first The score measures the independence between different modalities. If multiple modalities are highly correlated, it indicates redundancy in the decomposition, and the separation score decreases. This metric helps prevent VMD from breaking down the same physiological component into multiple highly similar modalities.

[0048] Computational complexity penalty Modal number The larger the value, the easier it is to over-decompose; therefore, excessively large values ​​need to be controlled. Apply a mild punishment.

[0049] Calculate the penalty for no ideal candidate If the current decomposition results do not contain any high-quality candidate modes within the ideal frequency band for the task, it indicates that although the parameter combination may have a small reconstruction error, it is not suitable for the task and therefore needs to be reduced in score.

[0050] Subsequently, the combined parameter score is obtained through fusion: (15) In the formula, , , The fusion weights are calculated for the overall score; This is the amplitude limiting function.

[0051] Finally, among the evaluated parameter combinations in Bayesian optimization, the parameter combination with the highest task fitness function value is selected as the optimal parameter combination for the current task: (16) In the formula, This represents the optimal parameter combination. Due to the use of Bayesian optimization, The search results, guided by the evaluated parameter combinations and the surrogate model, are used to determine the optimal combination, eliminating the need for an exhaustive search of all candidate parameter combinations.

[0052] The key to the above framework is that VMD parameters are no longer evaluated by a single decomposition quality, but by whether the combination of parameters can produce a target modality suitable for the current task.

[0053] Furthermore, the calculation method for the basic characteristics is explained. The calculation process for the main peak frequency includes: First, the modality To remove the mean: (17) In the formula, The mode after removing the mean; This is the signal length.

[0054] Then, the power spectral density is obtained using the power spectral estimation method. In the mission frequency band Find the largest spectral peak inside: (18) (19) In the formula, This indicates the most dominant frequency component of the mode within the mission frequency band. This indicates the power corresponding to that frequency. For a breathing task, this frequency represents the candidate respiratory rate; for a heart rate task, this frequency represents the candidate heart rate rate.

[0055] To improve the accuracy of frequency estimation, three-point parabolic interpolation can be performed on the spectral peak. Let the frequency point where the spectral peak is located be... The left and right adjacent spectral values ​​are respectively , , Then the peak position offset is: (20) The interpolated frequencies are: (twenty one) In the formula, This is the interpolated main peak frequency; For the frequency of the spectral peak; This refers to the frequency resolution.

[0056] Signal-to-noise ratio (SNR) is used to measure whether a mode has a prominent main peak within the target frequency band. If the main peak power is much higher than the background power within the frequency band, it indicates that the mode has a significant periodic component and is more likely to be a valid physiological signal.

[0057] The calculation process for the signal-to-noise ratio score is as follows: Define the modal signal-to-noise ratio: (twenty two) In this formula, the numerator is the peak power, and the denominator is the median of the power spectrum within the target frequency band. The median is used instead of the mean to reduce the impact of anomalous peaks on background power estimation.

[0058] Map the modal signal-to-noise ratio to Fraction: (twenty three) In the formula, This represents the signal-to-noise ratio (SNR) score. The formula uses logarithmic compression because the SNR can span a wide range. If the raw SNR is used directly, a few maxima can dominate the score; logarithmic mapping makes the score more stable.

[0059] Band purity is used to measure whether modal energy is concentrated in a narrow frequency region. If the modal energy is concentrated near the main peak, it indicates that it is more like a single physiological rhythm; if the energy is dispersed within the band, it indicates that the mode may contain multiple rhythms or noise.

[0060] The calculation process for the bandwidth purity fraction is as follows: Define the proportion of energy near the main peak to the total energy of the mission frequency band: (twenty four) In the formula, For frequency band purity; The bandwidth is the main peak neighborhood. For breathing tasks, the main peak neighborhood can be narrower because the respiratory rate changes slowly; for heart rate tasks, it can be slightly wider because heart rate signals are more susceptible to noise and morphological changes in non-contact scenarios.

[0061] Normalize the bandwidth purity: (25) In the formula, The frequency band purity fraction, with values ​​ranging from 1 to 10. ; This is a task-related normalization constant. If Reaching or exceeding If the purity fraction is close to 1, then the purity fraction is close to 1.

[0062] Frequency stability measures whether a mode maintains a similar dominant frequency over different time periods. Real breathing and heart rate typically do not fluctuate drastically over short periods, while noise or spurious peaks may exhibit significant frequency variations across different time windows.

[0063] The calculation process for the frequency stability score is as follows: The modes are divided into The first time window, the first The main frequency within each time window is Define frequency fluctuation: (26) The frequency stability score is then: (27) In the formula, The standard deviation of the segmented main frequency; The frequency stability score is given by the following values: ; These are task-related scale parameters. The formula states that if the dominant frequency remains almost constant across different time windows, then... A small value results in a stability score close to 1; a large fluctuation in the dominant frequency leads to a lower stability score. This metric is used to suppress accidental noise peaks or short-term spurious peaks.

[0064] Peak width is used to measure whether a mode is narrow enough. An ideal VMD mode should form a narrow main peak in the spectrum. If the peak width is too wide, it indicates that the mode may contain multiple frequency components or be affected by strong noise.

[0065] The calculation process for the peak width fraction is as follows: Let the peak width be The task bandwidth is: (28) Define relative peak width: (29) The peak width fraction is then: (30) In the formula, The peak width fraction, with values ​​ranging from 0 to 1. . and These are the low and high thresholds for relative peak width, respectively, with default values ​​of 0.08 and 0.35. The peak width fraction means: relative peak width less than... The band width was considered very narrow, so it received full marks; the relative peak width was greater than... The spectrum was deemed too wide, so 0 points were given; the middle region showed a linear decrease.

[0066] Spectral peaks indicate the presence of periodic components in the frequency domain, but relying solely on the spectrum can be misleading due to noise spikes. Therefore, this invention also uses autocorrelation analysis to verify the actual existence of the rhythm in the time domain.

[0067] The calculation process for the autocorrelation strength score is as follows: Autocorrelation analysis was performed on the modes to obtain the autocorrelation principal periods. Its corresponding frequency is: (31) If the autocorrelation peak intensity is Then the autocorrelation strength score is: (32) In the formula, The autocorrelation strength score has a value of [value missing]. ; and These are the low and high thresholds for the autocorrelation strength, respectively. This score indicates whether the mode exhibits a stable periodicity in the time domain. The larger the value, the stronger the similarity between the signal and itself after a one-cycle delay, and the more obvious the periodicity.

[0068] Frequency-time consistency is used to measure whether the frequency estimated by the spectrum and the frequency estimated by the autocorrelation are consistent. First, frequency-time consistency is calculated: (33) In the formula, For task-related tolerances; To estimate the frequency in the frequency domain; This is the frequency estimate for autocorrelation. When... and When approaching, The closer to 1; the greater the difference between the two, The smaller the value, the more reliable the mode is if both the frequency and time domains point to the same period.

[0069] The frequency-time domain consistency score is then: (34) In the formula, The frequency-time domain consistency score, with values ​​ranging from 1 to 10. ; and These are the low and high thresholds for the frequency-time domain consistency score, respectively.

[0070] The task band is the wide range that can be searched, while the ideal band is the range that better matches the expected physiological state. For example, a heart rate task may allow a heart rate search range of 0.8–2.0 Hz, but a resting state prefers 60–75 bpm. Therefore, it is necessary to calculate the soft fraction of the ideal band, the process of which is as follows: For the task Set the ideal frequency band: .

[0071] If the dominant modal peak is within the ideal frequency band, the ideal frequency band fraction is 1; if it deviates from the ideal frequency band, Gaussian soft attenuation is applied based on the distance. (35) In the formula, The width of the Gaussian decay; For the ideal frequency band soft fraction, take the value of .

[0072] Furthermore, after searching for the optimal parameter combination and obtaining the respiratory mode or heart rate mode based on the variational mode decomposition results of the optimal respiratory parameter combination, the final frequency is obtained by fusing frequency domain estimation and autocorrelation estimation, instead of simply using the maximum spectral peak.

[0073] Specifically, bandpass filtering is applied to the respiratory or heart rate modes: (36) In the formula, This is the filtered signal; For input signals; The target frequency band for the mission.

[0074] Frequency domain main peak interpolation estimation and autocorrelation period estimation are performed on the filtered respiratory signal to obtain the frequency domain estimated frequency. With autocorrelation estimation frequency .

[0075] Calculate the consistency coefficient between the frequency domain estimated frequency and the autocorrelation estimated frequency: (37) In the formula, The consistency coefficient; For task-related frequency tolerance.

[0076] The fusion weights are dynamically determined based on the consistency coefficient, and then the frequency domain estimated frequency and the autocorrelation estimated frequency are adaptively weighted and fused. The final frequency is: (38) In the formula, Estimating weights in the frequency domain; The autocorrelation estimation weights are used. It's worth noting that in the breathing task, autocorrelation is significant for periodic stability, thus increasing the ACF weight is appropriate; in the heart rate task, heart rate spectral peaks are typically more direct, therefore the frequency domain weight is higher.

[0077] Frequency corresponding to the number of times per minute for: (39) when hour, Respiratory rate; when hour, This represents the heart rate value.

[0078] In step 203, the breathing task refers to the process of extracting the breathing modality from the slow-time signal and estimating the breathing rate; the breathing candidate modality score refers to the index that is constructed and quantified based on the breathing characteristics to determine the degree of modality adaptation to the breathing signal; the optimal breathing parameter combination is the VMD parameter with the highest total score of the breathing task; and the breathing rate refers to the number of breaths per unit time estimated by the optimal breathing modality.

[0079] In the specific implementation of step 203, the goal of the breathing task is to find a stable, low-frequency, narrow-band, and clearly periodic breathing mode from the slow-time signal. The breathing mode is used to estimate the respiratory rate and provide a respiratory rate prior for the subsequent heart rate task, enabling the heart rate branch to identify and suppress respiratory harmonics.

[0080] Let the target frequency band for the breathing task be: Hz, this frequency band corresponds to a resting breathing range of 10–20 bpm; the ideal frequency band is .

[0081] Bayesian optimization on the respiratory candidate parameter set The current candidate parameter combination for respiration is selected in the middle iteration. , The respiratory mode matrix is ​​obtained by performing variational mode decomposition on the preprocessed slow-time signal using formula (11).

[0082] Subsequently, the basic features of each respiratory mode in the respiratory modality matrix are calculated using formulas (22) to (35). After determining the weights of the basic features, the basic scores of the respiratory candidate modes are calculated based on the basic features: (40) In the formula, Baseline scores for respiratory candidate modalities; Baseline scores for respiratory candidate modalities; The signal-to-noise ratio score is used to measure whether there is a prominent main peak within the breathing frequency band. Frequency purity fraction is used to measure whether respiratory energy is concentrated in a narrow frequency range; The frequency stability score measures whether the respiratory rate is stable across different time windows. The autocorrelation intensity score is used to measure whether a mode has obvious periodicity in the time domain; The frequency-time domain consistency score is used to measure whether the spectrum estimate and the autocorrelation estimate are consistent. This is the peak width fraction, used to penalize excessively wide peaks and avoid selecting mixed modes; The ideal frequency band soft fraction is used to encourage modes to fall within the ideal frequency band of resting breathing; , , , , , , Assign weights to the basic features in the respiratory candidate modalities.

[0083] Furthermore, a soft penalty for the ideal band is calculated to balance the weight of the ideal frequency band and avoid over-reliance on the ideal range. The calculation expression is as follows: (41) In the formula, The ideal score comes with a soft penalty. This represents the intensity of the influence of the ideal frequency band on the total score. If the mode is located within the ideal breathing band, =1, the factor is close to 1; if the mode deviates from the ideal band, the factor decreases.

[0084] The final respiratory candidate modality score is obtained by combining the baseline score of the respiratory candidate modality with the soft penalty of the ideal band: (42) For any set of candidate respiratory parameters ( , The highest candidate respiratory modality score is selected as the target respiratory modality score. .

[0085] If a candidate mode exists that is within the ideal breathing frequency band and has a score exceeding the threshold, then the parameter combination is considered to have a high-quality ideal breathing candidate. (43) In the formula, Candidate for ideal breathing; The threshold for judging ideal breathing candidates; Ideal frequency band for respiratory tasks.

[0086] If none of the modalities meet this condition, it means that although the current parameter combination may have a good reconstruction effect, it cannot generate a good breathing modality. Therefore, a penalty term is applied, including a penalty for no ideal candidate mode in the breathing task and a penalty for no ideal candidate mode in the breathing task. The breathing task complexity penalty is as follows: (44) The penalty for not having an ideal candidate for the breathing task is: (45) The weights of each coefficient are determined, and the respiratory fitness score is calculated by combining the reconstruction quality, modal separation degree, and preset penalty term. The expression is as follows: (46) (47) In the formula, Assess respiratory fitness scores; For respiratory fitness function; The target respiratory modality score; To reconstruct the quality score; Modal separation score; Penalty for the complexity of the breathing task; There are no ideal candidate penalties for the breathing task; , , The formula represents the fusion weights. It explains that the breathing task prioritizes generating high-quality breathing modalities, followed by reconstruction quality and modality separation, while avoiding excessive modalities and parameter combinations without ideal candidates.

[0087] By comparing the respiratory fitness scores corresponding to the evaluated candidate combinations of respiratory parameters using Bayesian optimization, the combination with the highest respiratory fitness score is selected as the optimal combination of respiratory parameters. .

[0088] After the parameter search is completed, a formal respiratory mode selection is performed based on the optimal respiratory decomposition results. The formal selection is more refined than the parameter search because it also considers dynamic priors.

[0089] Specifically, variational mode decomposition is re-executed based on the optimal combination of respiratory parameters to obtain the optimal respiratory mode matrix.

[0090] Let the static respiratory prior center frequency be: Hz.

[0091] For each optimal breathing mode in the optimal breathing mode matrix, calculate the static breathing prior score: (48) In the formula, The static respiration prior fraction; For the first The main peak frequency of each mode; It represents the static prior standard deviation.

[0092] If the dynamic breathing prior is obtained from the first round of rough estimation, then the dynamic breathing prior score is: (49) In the formula, For dynamic respiratory prior fractions; For dynamic respiratory prior frequency; The prior standard deviation of dynamic respiration.

[0093] The static and dynamic respiratory prior scores are weighted and mixed to obtain the mixed respiratory prior: (50) In the formula, For respiratory mixture prior fractions; For respiratory dynamics, use mixed weights.

[0094] The use of a hybrid prior, rather than a fully dynamic prior, is to avoid locking in subsequent choices if the first round of estimation is incorrect. Static priors provide physiological common sense, while dynamic priors provide individual adaptability.

[0095] The formal selection score for breathing is calculated by weighted fusion based on the signal-to-noise ratio score, bandwidth purity score, frequency stability score, and mixed prior of the optimal breathing mode. (51) In the formula, Formal selection of scores for breathing; Signal-to-noise ratio score; This represents the frequency band purity fraction. Frequency stability score; , , , To assign weights.

[0096] If the candidate modality has a low signal-to-noise ratio or purity, it is soft-weighted instead of being directly deleted. (52) The optimal breathing mode with the highest formal selection score is selected as the final breathing mode: (53) In the formula, For assignment operation; This is the optimal respiratory modality index.

[0097] Subsequently, a respiratory task bandpass filter is performed on the final respiratory modality to obtain the filtered respiratory signal; frequency domain main peak interpolation estimation and autocorrelation period estimation are performed on the filtered respiratory signal to obtain the frequency domain estimated frequency and autocorrelation estimated frequency.

[0098] The consistency coefficient between the frequency domain estimated frequency and the autocorrelation estimated frequency is calculated. The fusion weight is dynamically determined based on the consistency coefficient. Then, the frequency domain estimated frequency and the autocorrelation estimated frequency are adaptively weighted and fused to obtain the breathing frequency. For specific calculation formulas, please refer to formulas (36) to (38).

[0099] This step significantly improves the stability and accuracy of respiratory rate estimation by independently searching for optimal VMD parameters for the breathing task and combining multidimensional modal quality assessment with hybrid prior-guided modal selection, providing a reliable respiratory harmonic prior for subsequent heart rate tasks.

[0100] In step 204, the heart rate task refers to the process of separating heart rate modes from slow-time signals, suppressing respiratory harmonic interference, and accurately estimating heart rate frequency; the heart rate candidate mode score is constructed based on the physiological characteristics of heart rate and incorporates a respiratory harmonic suppression mechanism, and is used to quantify the modality-fit heart rate signal and distinguish between effective heart rate and respiratory interference; the optimal heart rate parameter combination is the VMD parameter with the highest heart rate task fitness function value; heart rate frequency refers to the number of heart rate counts per unit time estimated by the optimal heart rate mode; respiratory harmonic prior refers to the suppression mechanism that uses the estimated respiratory frequency to determine whether a candidate mode is likely a respiratory harmonic; the respiratory harmonic penalty factor is a scoring factor that reduces the weight of suspected respiratory harmonic modes.

[0101] In the specific implementation of step 204, the goal of the heart rate task is to extract heart rate modes from the slow-time signal to estimate the heart rate and provide a base signal for the HRV task. The heart rate task is more complex than the breathing task because the heart rate signal amplitude is weaker and easily approximates respiratory harmonics. Therefore, the heart rate task not only needs to find modes within the heart rate frequency band but also needs to determine whether the mode might be a respiratory harmonic.

[0102] Let the target frequency band for the heart rate task be: Hz; ideal frequency band is This ideal frequency band corresponds to 60–75 bpm.

[0103] Bayesian optimization on the candidate parameter set for the heart rate task Iterative selection of candidate parameter combinations for the current heart rate ( , The heart rate mode matrix is ​​obtained by performing variational mode decomposition on the preprocessed slow-time signal using formula (10).

[0104] Subsequently, the basic characteristics of each heart rate mode in the heart rate mode matrix are calculated using formulas (22) to (35), and the candidate heart rate mode scores are calculated based on the basic characteristics combined with the respiratory harmonic penalty factor. It can be understood that one of the most important disturbances in the heart rate task is respiratory harmonics. If the respiratory rate is... If the second and third harmonics of respiratory harmonics are not properly quantified, they may fall into the heart rate frequency band. Therefore, it is necessary to introduce quantitative indicators for respiratory harmonic interference to achieve respiratory harmonic suppression.

[0105] Specifically, the selection process for the respiratory harmonic penalty factor is as follows: For each heart rate mode, the main frequency Calculate the minimum distance between the modal frequency and the breathing harmonic: (54) In the formula, The order of the breathing harmonics; The range of respiratory harmonic orders is determined by the upper and lower limits of the heart rate frequency band.

[0106] Based on the nearest harmonic distance, a combination of hard and soft penalties is used to determine the respiratory harmonic penalty factor. Specifically, when the nearest harmonic distance... Less than or equal to the preset hard penalty threshold At that time, that is If the candidate is deemed to have a high harmonic risk, a hard penalty is applied, and the preset weighting coefficient is reduced. As a respiratory harmonic penalty factor This embodiment is... The default value is 0.55.

[0107] When the nearest harmonic distance Greater than the preset hard penalty threshold At that time, that is In this case, a soft penalty method is adopted. First, the harmonic risk value is calculated, and then the respiratory harmonic penalty factor is determined based on the harmonic risk value. The calculation expression is as follows: (55) (56) In the formula, Harmonic risk value; It is a respiratory harmonic penalty factor; This refers to the breathing harmonic weighting coefficient; This is a soft penalty scale for respiratory harmonics. The closer the candidate frequency is to the respiratory harmonic, the better. The closer to 1, the stronger the penalty; when the candidate frequency is far from the breathing harmonic, As the value approaches zero, the penalty weakens. The quantitative index for respiratory harmonic interference applies a fixed hard penalty to strong harmonic interference at close range and a Gaussian soft penalty to weak interference at long range, effectively preventing respiratory harmonics from being misidentified as heart rate modes.

[0108] Calculate the ideal heart rate band with a soft penalty, balance the weighting of the ideal heart rate band, and avoid over-reliance on the ideal range: (57) In the formula, A score with soft penalty is assigned to the ideal heart rate. The intensity of the influence of the ideal frequency band of heart rate; The ideal frequency band soft fraction.

[0109] Determine the weights of the basic features, and calculate the baseline score of the candidate heart rate modality based on the basic features: (58) In the formula, For baseline scores of candidate heart rate modalities; , , , , , , Weights are assigned to the basic features in the heart rate candidate modes.

[0110] By integrating the baseline heart rate candidate modality score, respiratory harmonic penalty, and ideal band soft penalty, the final heart rate candidate modality score is obtained: (59) For any set of candidate heart rate parameters ( , The highest candidate heart rate modality score is selected as the target heart rate modality score. .

[0111] If a candidate mode exists that is within the ideal heart rate frequency band and has a score not lower than the threshold, then the current decomposition result is considered to have high-quality heart rate candidates. (60) In the formula, Candidates for ideal heart rate; The threshold for judging ideal heart rate candidates; Ideal frequency band for heart rate tasks.

[0112] If none of the modalities meet this condition, it means that although the current parameter combination may have a good reconstruction effect, it cannot generate a good heart rate modality. Therefore, a penalty term is applied, including a penalty for no ideal candidate for the heart rate task and a penalty for the complexity of the heart rate task. The penalty for no ideal candidate for the heart rate task is as follows: (61) The penalty for the complexity of heart rate tasks is: (62) The weights of each coefficient are determined, and the heart rate fitness score is calculated by combining the reconstruction quality, modal separation degree, and preset penalty term. The expression is as follows: (63) (64) In the formula, Score heart rate fitness; This is the heart rate adaptation function; The respiratory frequency serves as a priori for respiratory harmonics; The target heart rate modal score; To reconstruct the quality score; Modal separation score; Penalize the complexity of heart rate tasks; There are no ideal candidate penalties for heart rate tasks; , , For weight fusion.

[0113] By comparing the heart rate fitness scores corresponding to the evaluated candidate combinations of respiratory parameters optimized by Bayesian methods, the combination with the highest heart rate fitness score is selected as the optimal heart rate parameter combination. .

[0114] After the parameter search is completed, the formal heart rate mode selection is performed from the optimal heart rate parameter set. During the formal heart rate selection, dynamic priors, low-quality gating, low-frequency boundary penalties, and strong evidence harmonic preservation mechanisms are further introduced.

[0115] Specifically, variational mode decomposition is performed based on the optimal heart rate parameter combination to obtain the optimal heart rate mode matrix.

[0116] For each optimal heart rate mode in the optimal heart rate mode matrix, the static prior is derived from the ideal heart rate frequency band: (65) If a dynamic mental prior exists, then the dynamic mental prior score is: (66) In the formula, The static cardiac a priori score; For dynamic cardiac a priori scores; For dynamic heart-first a priori frequency; The first a priori standard deviation is used for dynamic heart rate.

[0117] By weighted mixing of static and dynamic heart rate prior scores, a mixed heart rate prior is obtained. (67) In the formula, Mixed prior fractions for heart rate; It is a dynamic weighted mixture of heart rate.

[0118] Based on the signal-to-noise ratio score, band purity score, frequency stability score, heart rate mixture prior, and harmonic risk value of the optimal heart rate mode, a weighted fusion-based baseline score for formal heart rate selection is calculated. (68) In the formula, A baseline score is officially selected for heart rate; , , , , To assign weights. Wherein, Indicates harmonic rejection capability. The further a candidate is from breathing harmonics, the better. The smaller, The larger.

[0119] If a candidate mode simultaneously satisfies low signal-to-noise ratio, low purity, and low stability, it will not be included in the candidate pool. (69) In the formula, For the first Signal-to-noise ratio of each mode; The signal-to-noise ratio threshold; For the first The frequency band purity of each mode; This is the purity threshold. For the first Frequency stability score of each mode; This is the stability threshold.

[0120] Furthermore, low-frequency peaks near the lower limit of the heart rate band are prone to originating from trend terms, respiratory residues, or body movement, rather than the true heart rate. Therefore, a low-frequency boundary penalty is introduced to suppress interference from spurious low-frequency peaks. Specifically, if the candidate frequency is below the upper limit of the low-frequency protection: Then calculate the normalized low-frequency offset: (70) Calculate the low-frequency penalty factor: (71) In the formula, Protect the low-frequency heart rate boundary; This is the normalized low-frequency offset; To restore the width of the low-frequency boundary; It is a low-frequency penalty factor; This is the minimum penalty factor. The formula states that when the candidate frequency is lower than... When the penalty factor is close to As the candidate frequency moves further away from the low-frequency boundary, the penalty gradually weakens.

[0121] The heart rate score obtained through fusion is the official selection score: (72) The optimal heart rate mode with the highest formal selection score is selected as the final heart rate mode: .

[0122] Subsequently, bandpass filtering of the final heart rate mode is performed to obtain the filtered heart rate signal; frequency domain main peak interpolation estimation and autocorrelation period estimation are performed on the filtered heart rate signal to obtain the frequency domain estimated frequency and autocorrelation estimated frequency.

[0123] The consistency coefficient between the frequency domain estimated frequency and the autocorrelation estimated frequency is calculated. The fusion weight is dynamically determined based on the consistency coefficient. Then, the frequency domain estimated frequency and the autocorrelation estimated frequency are adaptively weighted and fused to obtain the heart rate frequency. For specific calculation formulas, please refer to formulas (36) to (38).

[0124] This step constructs a harmonic penalty mechanism based on respiratory rate as a priori, and combines low-frequency boundary penalty and dynamic priori guidance to effectively suppress the risk of respiratory harmonics and low-frequency trends being mistakenly selected as heart rate signals, thereby improving the accuracy and anti-interference ability of heart rate estimation.

[0125] In step 205, the heart rate variability-guided modal score is an evaluation index specifically designed for heart rate variability analysis, focusing on the integrity of the heart rate peak structure, rhythm stability, and anti-interference ability, which is different from the conventional heart rate modal score; the optimal modality is a pure heart rate modality adapted to beat-by-beat heart rate peak detection; the heart rate variability reconstruction signal is a dedicated signal that removes interference and retains the complete heart rate rhythm, used for subsequent accurate peak detection.

[0126] In the specific implementation of step 205, the optimal heart rate decomposition result obtained in step 204 is reused, with respiratory rate as the criterion. For respiratory coupling prior, heart rate frequency To provide a priori information on heart rate variability (HRV), a dedicated scoring system for HRV was constructed. This system includes positive reward indicators and negative penalty indicators to accurately select the optimal modality suitable for HRV analysis.

[0127] Specifically, the optimal heart rate mode matrix is ​​obtained based on the variational mode decomposition results of the optimal heart rate parameter combination.

[0128] For each optimal heart rate mode in the optimal heart rate mode matrix, a positive reward metric and a negative penalty metric are calculated based on respiratory rate and heart rate. The positive reward metric rewards high-quality modes that meet the requirements of HRV analysis. The negative penalty metric penalizes interfering modes that do not meet the requirements of HRV analysis.

[0129] Among them, positive reward indicators include heart rate frequency band matching degree, heart rate prior matching degree, peak structure integrity, rhythm clarity and energy matching degree.

[0130] The expression for calculating heart rate band matching degree is: (73) In the formula, This is the heart rate frequency band matching score, which is 1 when the mode is within the target heart rate frequency band, and 0 otherwise. This is the target frequency band for heart rate tasks.

[0131] The expression for calculating the initial a priori matching degree is: (74) In the formula, The heart-first a priori matching score; The first a priori standard deviation; The HRV heart prior is the score indicating how close the modal frequencies are to the heart prior estimated in the previous stages. The closer to the prior, the higher the score.

[0132] The expression for calculating peak structure integrity is: (75) In the formula, Peak structure fraction; To detect the number of heart rate peaks using modal analysis; The peak spacing variation coefficient is used to determine whether a mode has a peak structure suitable for step-by-step detection. Insufficient peak count indicates that HRV calculation is not supported; excessively chaotic peak spacing indicates an unreliable peak structure.

[0133] The expression for calculating rhythm clarity is: (76) In the formula, This is a rhythm clarity score; a higher value indicates a purer heart rate rhythm. For the first Power spectrum of each mode; It is a very small positive number.

[0134] The formula for calculating energy fitness is: (77) In the formula, Energy fraction; The modal relative energy; This is a reference threshold for HRV energy fit. This parameter is used to avoid selecting modes with excessively low energies. While modes with excessively low energies may have appropriate frequencies, peak detection is easily affected by noise.

[0135] The reverse penalty indicators include low-frequency trend penalty, breathing proximity penalty, flat trend penalty, and out-of-band penalty. The calculation expression for the low-frequency trend penalty is as follows: (78) In the formula, This is a penalty score for low-frequency trends. This item is used to penalize excessively low-frequency modes. Below [a certain value]... The modalities are more likely to be trends, body movements, or respiratory residues than the signals needed to detect heart rate peaks.

[0136] The expression for calculating the proximity penalty for breathing is: (79) In the formula, The penalty score is based on proximity to the breath. For HRV respiratory prior; It is the frequency band adjacent to respiration; This represents the standard deviation of the respiratory proximity penalty. This term measures whether a candidate modality is close to or falls within the respiratory frequency band. The closer it is to breathing, the less likely it is to be a heart rate modality, and therefore the greater the penalty.

[0137] The expression for calculating the flat trend penalty is: (80) In the formula, Penalize scores for flat trends; It is an oscillating indicator; This is the threshold for HRV oscillation indicators. If the mode is too flat or the trend is too strong, it is not conducive to the detection of heart rate peaks, and therefore a penalty is required.

[0138] The expression for calculating out-of-band penalty is: (81) In the formula, The penalty score is for points deducted from the outside of the zone. The heart rate frequency band matching score.

[0139] By integrating positive reward and negative penalty metrics, the overall score for the heart rate variability-oriented mode of each submodality is obtained: (82) In the formula, , , , , Assign weights to positive reward metrics; , , , Assign weights to the reverse penalty indicators.

[0140] The heart rate variability-guided modal total score reflects the characteristics of the HRV task: correct frequency is only the foundation, and more importantly, the modality must have a waveform structure that can be used for beat-by-beat peak detection.

[0141] Furthermore, the HRV modality should avoid strong coupling with respiratory signals as much as possible. If a respiratory reference signal is present... Candidate HRV mode The punishments related to breathing are: (83) In the formula, The penalty score is related to breathing. This is a standardized operation used to eliminate the effects of differences in signal amplitude. This is the function for calculating the correlation coefficient. The larger the penalty value, the stronger the temporal correlation between the candidate mode and the respiratory signal, the more severe the respiratory coupling interference, and the less suitable it is as a HRV signal.

[0142] If the respiratory rate has been obtained through step 203 Calculate the minimum distance between the modal frequency and each order of respiratory harmonics, and construct a respiratory harmonic proximity penalty: (84) In the formula, The penalty score is the number of times the respiratory rate is adjacent to the nearest octave. The order of the breathing harmonics; To mitigate harmonic tolerance, the decay rate of the penalty is controlled. When the candidate mode frequency approaches an arbitrary order breathing harmonic, the penalty value increases significantly, suppressing misselection of breathing harmonics.

[0143] By combining the correlation penalty and the harmonic proximity penalty, the total breathing coupling penalty is obtained: (85) In the formula, The total respiratory coupling penalty score.

[0144] Therefore, the heart rate variability task adaptation function also needs to be integrated with the total respiratory coupling penalty score to obtain the final heart rate variability task adaptation function value.

[0145] Based on the final heart rate variability task adaptation function values, the highest-scoring optimal heart rate mode is selected as the optimal mode. This optimal mode is then smoothed and baseline-corrected to generate a reconstructed heart rate variability signal for heart rate peak detection. .

[0146] This step uses a dedicated HRV-guided scoring system to eliminate all interfering components and select modalities with clear peak structures, stable rhythms, and resistance to respiratory coupling. This provides a high-quality signal foundation for subsequent heart rate peak detection and IBI calculation, significantly improving the accuracy of HRV indicators in non-contact scenarios.

[0147] In step 206, peak detection refers to locating the peak position of the waveform corresponding to each heart rate from the HRV reconstructed signal; peak position refinement refers to correcting the peak position error through subsampling interpolation to improve the accuracy of the heartbeat interval; heartbeat interval quality control refers to screening and correcting adjacent heart rate intervals from multiple dimensions such as physiological rationality, statistical robustness, and abrupt jumps; heart rate variability indicators are core parameters reflecting the autonomic nervous system regulation function of the heart, including mean interval, interval standard deviation, and root mean square of short time difference.

[0148] In the specific implementation of step 206, the heart rate variability reconstruction signal obtained in step 205 is used. The heart rate frequency obtained in step 204 The heart rate peak detection, peak position refinement, interval quality control, and HRV index calculation are performed sequentially.

[0149] First, to improve the detectability of heart rate peaks, the heart rate variability reconstruction signal is first subjected to multi-dimensional enhancement processing, including bandpass filtering, Hilbert envelope fusion, and local peak normalization.

[0150] Specifically, the reconstructed signal of heart rate variability Perform heart rate bandpass filtering to suppress out-of-band noise and low-frequency trends: (86) In the formula, The signal is after bandpass filtering; For heart rate task target frequency band; Subsequently, the Hilbert envelope is calculated on the bandpass filtered signal to enhance the weaker heart rate ripple components: (87) In the formula, For Hilbert envelope; This is the Hilbert transform operator.

[0151] The bandpass filtered signal is combined with the Hilbert envelope to obtain the signal used for peak detection: (88) In the formula, , For weight fusion.

[0152] This step utilizes both the raw oscillation information and the envelope information of the heart rate waveform. The raw bandpass signal retains peak information, while the envelope can amplify weaker heart rate fluctuations.

[0153] To eliminate the influence of the slow change trend in signal amplitude, local peak normalization is performed on the enhanced signal: (89) In the formula, This is the peak detection signal after local normalization; This is a sliding median filter operator used to estimate the local baseline; This is the sliding median absolute deviation operator, used to estimate local noise levels; The value should be a very small positive number to avoid a denominator of zero. Local normalization aims to eliminate signal amplitude drift, making heart rate peaks at different time periods comparable at similar scales, thus improving the consistency of peak detection.

[0154] Subsequently, candidate peak detection was conducted. Specifically, based on the initial detection... Calculate the prior heartbeat interval : (90) To balance high recall and high reliability in heartbeat peak detection, this embodiment employs a multi-strategy candidate peak detection scheme, while also introducing derivative peak detection to supplement weak heartbeat events.

[0155] The multi-strategy candidate peak detection scheme includes high-recall candidate peaks and high-quality candidate peaks. High-recall candidate peaks employ a lower peak prominence threshold and a smaller minimum peak distance, aiming to minimize the omission of true heartbeat peaks and construct a candidate set. High-quality candidate peaks employ a higher peak prominence threshold and a minimum peak distance closer to the physiological heartbeat interval, aiming to obtain reliable anchor peaks and construct a candidate set. ; The introduced derivative peak candidate is constructed by taking the derivative of the signal and using the rising edge information to capture heartbeat events with insufficiently prominent peaks but obvious rising edges, thus building a candidate set. .

[0156] Finally, by merging the three candidate peak sets, a complete candidate peak set is obtained: (91) After candidate peak detection, not all candidates can be directly identified as heartbeat peaks, as they may contain noise peaks, shoulder peaks, and repeating peaks. Therefore, a quality score needs to be assigned to each candidate peak. Specifically, for the candidate peak position... In its neighborhood The local median baseline and local MAD are calculated to obtain the local... Fraction: (92) In the formula, Candidate peak The local robustness standard score is used to measure whether the peak is sufficiently prominent relative to the local background; the larger the value, the more significant the peak. Candidate peak The local neighborhood; This represents the absolute deviation of the median.

[0157] Subsequently, the multidimensional peak quality score is calculated, including: Peak prominence score: (93) In the formula, Peak prominence score; Peak prominence; Peak prominence normalization threshold The peak prominence score measures how prominent a candidate peak is relative to its neighboring valleys; the higher the score, the better the peak's distinguishability.

[0158] Local Fraction: (94) In the formula, For local Fraction; This is the threshold for local robust standard score normalization.

[0159] Peak width fraction: (95) In the formula, Peak width fraction; For peak width; Slow sampling rate; The a priori cardiac interval; This is the scale ratio for the peak width score. The peak width score is used to measure whether the peak width conforms to the reasonable width of a heartbeat peak. A peak that is too narrow may be a noise spike, while a peak that is too wide may be a respiration or trend residue.

[0160] Fractions with left-right symmetry: (96) In the formula, Fractions that are symmetrical about left and right; and These represent the decrease in amplitude on the left and right sides of the candidate peak, respectively. If the decrease amplitudes on the left and right sides are similar, the peak shape is more symmetrical and the quality is higher.

[0161] By combining the above four scores, the final peak quality score is obtained: (97) In the formula, , , , For weight fusion.

[0162] The final peak quality score comprehensively considers the peak prominence, local significance, peak width rationality, and peak shape symmetry, fully reflecting the reliability of the candidate peak.

[0163] There may be multiple very close peaks among the candidate peaks, actually corresponding to the same heartbeat. To avoid double counting, only the highest quality peak is retained during the physiological refractory period, requiring non-maximum suppression. First, calculate the overall score: (98) In the formula, For comprehensive scoring; Normalized peak amplitude; , This is used for the overall scoring weight. The overall score prioritizes peak quality and uses peak amplitude as a secondary factor to avoid retaining only the highest peaks with poor morphological quality.

[0164] Non-maximum suppression is performed based on the comprehensive score, and the candidate peak with the highest comprehensive score in each physiological refractory period is retained.

[0165] If the time interval between two adjacent peaks is significantly longer than the reasonable range of the current rhythm, peak omission may occur. In this case, the detection threshold is lowered within the long interval to search for weak peaks and perform long-interval peak compensation. If the interval between adjacent peaks satisfies: (99) In the formula, This represents the current IBI median. This is the long-interval determination coefficient. We then search for weak peaks within this interval, and candidate weak peaks must still meet the minimum peak quality requirement: (100) In the formula, A candidate weak peak; This is the threshold for weak peak quality. The purpose of long-interval peak filling is not to forcibly smooth the rhythm, but to attempt to fill in possible true weak peaks when there are obvious missing peaks.

[0166] Due to the low sampling rate of the slow time, the positions of integer sampling points may deviate from the true peak. Peak position errors directly affect IBI, therefore, the retained heartbeat peaks need to be refined through subsampling. Specifically, let the local integer peaks be... The value of its adjacent sampling points , , The subsampling offset is then: (101) Limit offset range: The refined peak heart rate time is: (102) In the formula, This refers to the peak heart rate time after refinement. This represents the slow sampling rate of the radar signal. Peak position refinement is equivalent to fitting the peak position with a local parabola, which can improve the peak time accuracy to the sub-sampling level without increasing the sampling rate, significantly reducing the error in calculating the cardiac interval.

[0167] The refined peak time was used to calculate the original IBI: (103) In the formula, The cardiac interval is the time difference between two adjacent heartbeat peaks.

[0168] IBI quality control includes physiological range filtering, robust anomaly filtering, jump filtering, and peak quality filtering.

[0169] First, calculate the median of the original IBI. Set physiological range: (104) (105) IBIs that meet the following criteria pass the physiological range test: (106) Then calculate the robustness metric: (107) Robust anomaly assessment: (108) Robust anomaly detection is used to remove anomalous intervals that deviate significantly from the overall IBI distribution.

[0170] The jump is restricted to: (109) In the formula, This is the absolute allowable threshold for jumps between adjacent heartbeats; This is the relative tolerance coefficient for jumps between adjacent heartbeats.

[0171] Jump restrictions are used to eliminate cases where the abrupt change between adjacent IBIs is too large. The abrupt change may originate from a missing peak, a false peak, or an incorrect peak position.

[0172] Furthermore, to describe the credibility of IBI in more detail, the IBI quality score is calculated: (110) The rhythm fraction is: (111) Rhythm score is used to measure whether the current heartbeat interval is close to the typical heartbeat interval of that signal segment. If An IBI close to the median indicates that the interval conforms to the overall heart rhythm, and the score is close to 1. If the interval is significantly too long or too short, the score decreases. The formula uses a Gaussian form, meaning that the greater the deviation, the lower the confidence level becomes non-linearly. It should be noted that this score is not intended to force all IBIs to be the same, but rather to identify obviously unreasonable intervals. The jump score is: (112) The jump score measures whether the change between two adjacent inter-mutation intervals (IBIs) is excessive. Normal HRV allows for IBI variation, but extreme jumps within a short period are usually more likely to originate from false peaks or missed peaks. This item is used to reduce the quality score of the mutation interval.

[0173] Peak pair mass fraction: (113) Peak pair quality score indicates whether the two adjacent peaks constituting the current IBI are reliable. If both peaks are of high quality, the IBI is more reliable; if either peak is of low quality, the reliability of the IBI decreases.

[0174] Ultimately, the effective cardiac interval will be determined through quality control (QC). Constructing an effective cardiac interval sequence It is used for calculating the HRV index.

[0175] Based on the effective heart rate interval sequence, core heart rate variability indices are calculated, including: Mean heartbeat interval: (114) Overall standard deviation of cardiac interval: (115) Root mean square of the difference between adjacent heartbeats: (116) The proportion of consecutive heartbeats with a time difference greater than 50 ms: (117) hour, .

[0176] Instantaneous heart rate: (118) The average heart rate and median heart rate are as follows: (119) (120) In the formula, The mean heartbeat interval reflects the baseline heart rate level; The standard deviation of all normal heartbeat intervals reflects the overall HRV level; It is the root mean square of the difference between adjacent heartbeats, which is more sensitive to short-term heart rate fluctuations and reflects short-term HRV; The proportion of adjacent heartbeats with a time difference greater than 50ms is commonly represented by the HRV time-domain index. To count the number of elements that meet the conditions; IBI conversion after quality control.

[0177] This step employs a signal enhancement scheme involving bandpass filtering, Hilbert envelope fusion, and local peak normalization, combined with high-precision peak position refinement and rigorous interval quality control, to maximize the sensitivity and accuracy of peak detection. Ultimately, it outputs stable and reliable heart rate variability indicators, achieving non-contact, high-precision heart rate variability analysis.

[0178] It should be understood that, although this embodiment Figure 1 The steps are shown sequentially as indicated by the arrows, but they are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order in which these steps are performed; they can be executed in other orders. Figure 1 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0179] Example 2 Based on the non-contact heart rate variability analysis method based on task-oriented parameter selection in Example 1, this embodiment discloses a non-contact heart rate variability analysis device based on task-oriented parameter selection, such as... Figure 2 As shown, the non-contact heart rate variability analysis device based on task-oriented parameter selection includes: a signal acquisition and preprocessing module 401, a parameter set construction module 402, a respiratory parameter optimization module 403, a heart rate parameter optimization module 404, a heart rate variability reconstruction module 405, and a heart rate variability index calculation module 406, wherein: The signal acquisition and preprocessing module 401 is used to acquire slow-time signals of micro-movements in the human chest and abdomen, and obtains preprocessed slow-time signals after preprocessing.

[0180] The parameter set construction module 402 is used to construct the respiratory candidate parameter set and the heart rate candidate parameter set by setting the range of modality number and penalty factor values ​​according to the characteristics of the breathing task and the heart rate task, respectively.

[0181] The respiratory parameter optimization module 403 is used to perform mixed integer Bayesian optimization on the respiratory candidate parameter set with the respiratory fitness function as the objective function, iteratively select the combination of respiratory candidate parameters, perform variational mode decomposition on the preprocessed slow-time signal, calculate the respiratory fitness score, determine the optimal combination of respiratory parameters based on the respiratory fitness score, and output the respiratory rate.

[0182] The heart rate parameter optimization module 404 is used to perform mixed integer Bayesian optimization on the heart rate candidate parameter set with respiratory frequency as the respiratory harmonic prior and heart rate fitness function as the objective function. It iteratively selects the combination of heart rate candidate parameters to perform variational mode decomposition on the preprocessed slow-time signal and calculates the heart rate fitness score. Based on the heart rate fitness score, the optimal combination of heart rate parameters is determined and the heart rate frequency is output.

[0183] The heart rate variability reconstruction module 405 is used to score the variational mode decomposition results corresponding to the optimal heart rate parameter combination using respiratory rate as the respiratory coupling prior and heart rate rate as the heart rate variability heart prior, to select the optimal mode and generate the heart rate variability reconstruction signal.

[0184] The heart rate variability index calculation module 406 is used to perform peak detection, peak position refinement and inter-beat interval quality control on the heart rate variability reconstruction signal based on heart rate frequency to obtain an effective inter-beat interval sequence; and calculate and output the heart rate variability index based on the effective inter-beat interval sequence.

[0185] In this embodiment, the specific working process and working principle of the signal acquisition and preprocessing module 401, parameter set construction module 402, respiratory parameter optimization module 403, heart rate parameter optimization module 404, heart rate variability reconstruction module 405, and heart rate variability index calculation module 406 are the same as those in Embodiment 1, and therefore will not be described again in this embodiment. Each unit module can be implemented entirely or partially through software, hardware, or a combination thereof. Each unit module can be embedded in or independent of the processor in the computer device in hardware form, or it can be stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to the above unit modules.

[0186] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0187] The embodiments described above are merely examples of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A non-contact heart rate variability analysis method based on task-oriented parameter selection, characterized in that, The method includes: Slow-time signals of micro-movements in the human chest and abdomen were acquired and preprocessed to obtain preprocessed slow-time signals. Based on the characteristics of the breathing task and the heart rate task, the range of modality number and penalty factor values ​​are set respectively, and the candidate parameter sets for breathing and heart rate are constructed. Using the respiratory fitness function as the objective function, mixed integer Bayesian optimization is performed on the respiratory candidate parameter set. Variational mode decomposition is then performed on the preprocessed slow-time signal to iteratively select combinations of respiratory candidate parameters and calculate the respiratory fitness score. Based on the respiratory fitness score, the optimal combination of respiratory parameters is determined and the respiratory rate is output. Using the respiratory frequency as the respiratory harmonic prior and the heart rate fitness function as the objective function, mixed integer Bayes optimization is performed on the heart rate candidate parameter set. The combination of heart rate candidate parameters is iteratively selected to perform variational mode decomposition on the preprocessed slow-time signal to calculate the heart rate fitness score. Based on the heart rate fitness score, the optimal combination of heart rate parameters is determined and the heart rate frequency is output. Using the respiratory frequency as the respiratory coupling prior and the heart rate frequency as the heart rate variability heart rate prior, the variational mode decomposition results corresponding to the optimal heart rate parameter combination are scored by the heart rate variability fitness function, the optimal mode is selected and the heart rate variability reconstruction signal is generated. Based on the heart rate frequency, peak detection, peak position refinement, and heart rate interval quality control are performed on the heart rate variability reconstruction signal to obtain an effective heart rate interval sequence; heart rate variability index is calculated and output based on the effective heart rate interval sequence.

2. The non-contact heart rate variability analysis method based on task-oriented parameter selection according to claim 1, characterized in that, Mixed-integer Bayesian optimization is performed on the set of respiratory candidate parameters. Variational mode decomposition is then performed on the preprocessed slow-time signal to iteratively select combinations of respiratory candidate parameters and calculate the respiratory fitness score, including: In each iteration, a surrogate model is constructed based on the evaluated respiratory candidate parameter combinations and corresponding respiratory fitness scores. The next set of respiratory candidate parameter combinations is selected from the respiratory candidate parameter set by maximizing the expected value of the acquisition function. Variational mode decomposition is performed on the preprocessed slow-time signal to obtain the respiratory mode matrix. Calculate the basic respiratory features of each respiratory mode in the respiratory modality matrix, and calculate the respiratory candidate mode score based on the basic respiratory features; The highest candidate respiratory modality score is selected as the target respiratory modality score, and the respiratory fitness score is calculated by combining reconstruction quality, modality separation degree and preset penalty term.

3. The non-contact heart rate variability analysis method based on task-oriented parameter selection according to claim 2, characterized in that, Determining the optimal combination of respiratory parameters based on the respiratory fitness score includes: The respiratory fitness score is calculated using the respiratory fitness function, expressed as follows: ； ； In the formula, Assess respiratory fitness scores; For respiratory fitness function; The target respiratory modality score; To reconstruct the quality score; Modal separation score; Penalty for the complexity of the breathing task; There are no ideal candidate penalties for the breathing task; , , For weighting; The candidate respiratory parameter combination with the highest respiratory fitness score was selected as the optimal respiratory parameter combination.

4. The non-contact heart rate variability analysis method based on task-oriented parameter selection according to claim 1, characterized in that, The process of obtaining respiratory rate includes: Based on the optimal combination of respiratory parameters, variational mode decomposition is re-executed to obtain the optimal respiratory mode matrix; For each optimal breathing mode in the optimal breathing mode matrix, calculate the static breathing prior score and the dynamic breathing prior score; The static breathing prior score and the dynamic breathing prior score are weighted and mixed to obtain the mixed breathing prior. The breathing formal selection score is calculated by weighted fusion based on the signal-to-noise ratio score, bandwidth purity score, frequency stability score, and the hybrid prior of the optimal breathing mode. The optimal breathing mode with the highest formal selection score is selected as the final breathing mode. The final breathing modality is subjected to a breathing task bandpass filter to obtain the filtered breathing signal; the filtered breathing signal is then subjected to frequency domain main peak interpolation estimation and autocorrelation period estimation to obtain the frequency domain estimated frequency and autocorrelation estimated frequency. The consistency coefficient between the frequency domain estimated frequency and the autocorrelation estimated frequency is calculated. The fusion weight is dynamically determined based on the consistency coefficient. Then, the frequency domain estimated frequency and the autocorrelation estimated frequency are adaptively weighted and fused to obtain the breathing frequency.

5. The non-contact heart rate variability analysis method based on task-oriented parameter selection according to any one of claims 1 to 4, characterized in that, Mixed-integer Bayesian optimization is performed on the heart rate candidate parameter set, and variational mode decomposition is performed on the preprocessed slow-time signal to iteratively select heart rate candidate parameter combinations, and a heart rate fitness score is calculated, including: In each iteration, a surrogate model is constructed based on the evaluated heart rate candidate parameter combinations and corresponding heart rate fitness scores. The next heart rate candidate parameter combination is selected from the heart rate candidate parameter set by maximizing the expectation boost acquisition function. Variational mode decomposition is performed on the preprocessed slow-time signal to obtain the heart rate mode matrix. Calculate the basic heart rate characteristics of each heart rate mode in the heart rate mode matrix, and calculate the heart rate candidate mode score based on the basic heart rate characteristics and the respiratory harmonic penalty factor. The highest heart rate candidate modal score is selected as the target heart rate modal score, and the heart rate fitness score is calculated by combining reconstruction quality, modal separation degree and preset penalty term.

6. The non-contact heart rate variability analysis method based on task-oriented parameter selection according to claim 5, characterized in that, The selection of the respiratory harmonic penalty factor includes: Using the respiratory frequency as a priori for respiratory harmonics, the nearest harmonic distance between the dominant frequency of each heart rate mode in the heart rate mode matrix and the integer multiples of the respiratory frequency is calculated. Based on the nearest harmonic distance, the breathing harmonic penalty factor is determined by combining hard and soft penalties; Wherein, when the nearest harmonic distance is less than or equal to a preset hard penalty threshold, a hard penalty method is adopted, and the preset strong weighting coefficient is used as the breathing harmonic penalty factor; When the nearest harmonic distance is greater than the preset hard penalty threshold, a soft penalty method is used to calculate the harmonic risk value, and a respiratory harmonic penalty factor is determined based on the harmonic risk value.

7. The non-contact heart rate variability analysis method based on task-oriented parameter selection according to claim 6, characterized in that, Determining the optimal combination of heart rate parameters based on the heart rate fitness score includes: The heart rate fitness score is calculated using the heart rate fitness function, expressed as follows: ； ； In the formula, Score heart rate fitness; This is the heart rate adaptation function; The respiratory frequency serves as a priori for respiratory harmonics; The target heart rate modal score; To reconstruct the quality score; Modal separation score; Penalize the complexity of heart rate tasks; There are no ideal candidate penalties for heart rate tasks; , , For weighting; The respiratory candidate parameter combination with the highest heart rate fitness score was selected as the optimal heart rate parameter combination.

8. The non-contact heart rate variability analysis method based on task-oriented parameter selection according to claim 7, characterized in that, The process of obtaining heart rate includes: Variational mode decomposition is performed based on the optimal heart rate parameter combination to obtain the optimal heart rate mode matrix; For each optimal heart rate mode in the optimal heart rate mode matrix, calculate the static heart rate advance score and the dynamic heart rate advance score; The static heart rate prior score and the dynamic heart rate prior score are weighted and mixed to obtain the mixed heart rate prior. Based on the signal-to-noise ratio score, band purity score, frequency stability score, heart rate mixing prior, and harmonic risk value of the optimal heart rate mode, the formal selection score of heart rate is calculated by weighted fusion. The optimal heart rate mode with the highest formal heart rate selection score is selected as the final heart rate mode; The final heart rate mode is subjected to bandpass filtering to obtain the filtered heart rate signal; frequency domain main peak interpolation estimation and autocorrelation period estimation are performed on the filtered heart rate signal to obtain the frequency domain estimated frequency and autocorrelation estimated frequency. The consistency coefficient between the frequency domain estimated frequency and the autocorrelation estimated frequency is calculated. The fusion weight is dynamically determined based on the consistency coefficient. Then, the frequency domain estimated frequency and the autocorrelation estimated frequency are adaptively weighted and fused to obtain the heart rate frequency.

9. The non-contact heart rate variability analysis method based on task-oriented parameter selection according to any one of claims 1 to 8, characterized in that, The variational mode decomposition results corresponding to the optimal heart rate parameter combination are scored using a heart rate variability fitness function, the optimal mode is selected, and a heart rate variability reconstruction signal is generated, including: The optimal heart rate mode matrix is ​​obtained based on the variational mode decomposition results of the optimal heart rate parameter combination. Using the respiratory rate as the respiratory coupling prior and the heart rate rate as the heart rate variability heart rate prior, positive reward indicators and negative penalty indicators are calculated for each optimal heart rate mode in the optimal heart rate mode matrix. The positive reward indicators include heart rate band matching degree, heart rate prior matching degree, peak structure integrity, rhythm clarity, and energy fit. The negative penalty indicators include low-frequency trend penalty, respiratory proximity penalty, flat trend penalty, and out-of-band penalty. By integrating the positive reward index and the negative penalty index, the heart rate variability task adaptation function value for each submodal is obtained; Based on the heart rate variability task adaptation function values, the optimal heart rate mode with the highest score is selected as the optimal mode; based on the optimal mode, a heart rate variability reconstruction signal for heart rate peak detection is generated.

10. A non-contact heart rate variability analysis device based on task-oriented parameter selection, characterized in that, The device includes: The signal acquisition and preprocessing module is used to acquire slow-time signals of micro-movements in the human chest and abdomen, and obtain preprocessed slow-time signals after preprocessing. The parameter set construction module is used to construct candidate parameter sets for breathing and heart rate based on the characteristics of breathing and heart rate tasks, respectively, by setting the range of modality number and penalty factor values. The respiratory parameter optimization module is used to perform mixed integer Bayesian optimization on the respiratory candidate parameter set with the respiratory fitness function as the objective function, iteratively select the combination of respiratory candidate parameters to perform variational mode decomposition on the preprocessed slow-time signal, calculate the respiratory fitness score, determine the optimal combination of respiratory parameters based on the respiratory fitness score, and output the respiratory rate. The heart rate parameter optimization module is used to perform mixed integer Bayes optimization on the heart rate candidate parameter set with the respiratory frequency as the respiratory harmonic prior and the heart rate fitness function as the objective function, iteratively select the heart rate candidate parameter combination to perform variational mode decomposition on the preprocessed slow-time signal, calculate the heart rate fitness score, determine the optimal heart rate parameter combination based on the heart rate fitness score, and output the heart rate frequency. The heart rate variability reconstruction module is used to score the variational mode decomposition results corresponding to the optimal heart rate parameter combination using the respiratory frequency as the respiratory coupling prior and the heart rate frequency as the heart rate variability heart prior, to screen the optimal mode and generate a heart rate variability reconstruction signal. The heart rate variability index calculation module is used to perform peak detection, peak position refinement, and inter-beat interval quality control on the heart rate variability reconstruction signal based on the heart rate frequency to obtain an effective inter-beat interval sequence; and to calculate and output the heart rate variability index based on the effective inter-beat interval sequence.

Images