Respiration rate continuous estimation method based on pulse wave time-frequency domain feature fusion

Abstract

This invention discloses a respiratory rate extraction method based on multimodal pulse wave feature fusion, comprising: 1. filtering and windowing the pulse wave signal; 2. extracting four types of non-uniform sampling feature sequences from each signal window: respiratory induction amplitude variation, respiratory induction intensity variation, respiratory induction frequency variation, and respiratory induction continuous contraction difference variation; 3. converting each sequence to the frequency domain using a Lomb-Scargle periodogram, extracting the corresponding respiratory rate estimate, and calculating signal quality indices including main peak significance, in-band signal-to-noise ratio, and harmonic energy ratio; 4. constructing a four-dimensional feature vector based on the respiratory rate and quality indices corresponding to each feature, obtaining a preliminary frequency estimate through comprehensive confidence weighting, consensus cluster fusion, and harmonic correction, and then outputting the final respiratory rate after time-series smoothing. This invention can achieve complementary fusion of multiple features, thereby effectively improving the robustness and accuracy of respiratory rate estimation.

Description

Technical Field

[0001] This invention relates to the field of vital sign monitoring based on biosignal feature extraction and multimodal information fusion, and particularly to a non-uniform sampling respiratory frequency estimation method and system based on multidimensional respiratory-induced variation features of pulse waves. Background Technology

[0002] Respiratory rate, as a key physiological indicator reflecting the state of human respiratory function and circulatory system, plays an irreplaceable role in accurate and real-time monitoring in scenarios such as clinical intensive care, respiratory disease screening, sleep disorder diagnosis, and daily health management. Traditional respiratory monitoring methods, such as chest and abdominal strain gauges and nasal cannula airflow sensors, can directly acquire respiratory signals, but they have limitations such as high invasiveness, low wearing comfort, and susceptibility to interference from limb movements, making them unsuitable for long-term dynamic monitoring or wearable device applications. In contrast, pulse wave-based respiratory signal extraction technology, with its advantages of being non-invasive, portable, and capable of being integrated with heart rate monitoring, has become an important research direction in the field of non-invasive physiological monitoring.

[0003] Patent CN106073783A describes a method that simultaneously acquires pulse waves and reference respiratory signals, extracts pulse wave period and diastolic amplitude features, performs Fourier transform, and constructs a linear regression model to calculate respiratory rate. While this method is logically simple, it relies heavily on a large number of reference signals to train the model, and the linear assumption makes it difficult to adapt to individual physiological differences, resulting in significant errors in nonlinear scenarios.

[0004] Patent CN112998690A uses an array sensor to collect multi-channel pulse waves. After FIR filtering and channel selection and fusion, features such as peak envelopes are extracted and a reliable metric is constructed. The respiratory rate is then output by combining an attention network and a VGG regression model. Multi-feature fusion improves robustness, but it is computationally resource-intensive. The aspect ratio threshold for sub-region selection is empirical, resulting in insufficient adaptability.

[0005] In patent CN116172512A, the pulse wave correlation entropy spectral density is first calculated, the target frequency bands for respiration and pulse waves are divided and the boundary frequencies are determined, and an empirical wavelet set is constructed to decompose the signal and extract the respiration signal. It can adaptively suppress noise, but the boundary frequencies are easily affected by weak respiratory modulation and may shift, resulting in the mixing of respiration and pulse wave components.

[0006] In patent CN116327171A, the pulse wave is first filtered and denoised, the peak and trough positions are detected, and three types of signal quality indices are calculated. Two types of respiratory wave signals are extracted through spline fitting, and after correction and alignment, a signal template is established. The final respiratory wave is obtained by combining the correlation coefficient. This method improves anti-interference capability through peak-trough dual-path fusion.

[0007] In Huang Yiping's academic paper "A Method for Calculating Respiratory Rate Based on Pulse Waves," pulse wave and respiratory wave signals are obtained from the MIMIC database. Empirical Mode Decomposition (EMD) is used to extract intrinsic mode functions, and the respiratory wave is reconstructed to calculate the respiratory rate. This method achieves a relative coherence coefficient exceeding 0.6 and an accuracy of over 0.9. However, EMD is prone to mode aliasing and lacks sufficient accuracy in reconstructing respiratory signals from low signal-to-noise ratio pulse waves.

[0008] In their paper titled "Ensemble Empirical Mode Decomposition With Principal Component Analysis: A Novel Approach for Extracting Respiratory Rate and Heart Rate From Photoplethysmographic Signal" published in the IEEE Journal of Biomedical and Health Informatics, Mohammod Abdul Motin's team used EEMD to decompose the pulse wave into multiple IMF components. After removing IMFs containing artifacts, they grouped the components by frequency and performed PCA on each group of IMFs to reconstruct the respiratory signal from the first principal component. This method alleviates the mode aliasing problem of traditional EMD. However, EEMD requires repeated addition of Gaussian white noise for decomposition, which is computationally complex and has poor real-time performance in long-term monitoring scenarios.

[0009] In their paper titled "An efficient model for extracting respiratory and blood oxygensaturation data from photoplethysmogram signals by removing motion artifacts using a heuristic-aided ensemble learning model," published in *Computers in Biology and Medicine*, Venumaheswar Rao Bondala's team first removes motion artifacts from pulse waves and performs filtering preprocessing. They then extract spectral features and the variance of respiratory peaks, optimize feature selection and weights using AGTBO, and finally input the data into an ensemble network of MLP, AdaBoost, and attention LSTM to calculate the respiratory rate. This method is highly robust to interference, but its complex model structure requires extensive training with numerous samples, and the process is difficult to trace.

[0010] In their paper titled "Deriving respiration from photoplethysmographic pulse width" published in *Medical & Biological Engineering & Computing*, Jesús Lázaro's team used low-pass derivative signals to locate the points of maximum slope on the rising and falling edges of the pulse wave, determining the start and end points of the pulse wave and calculating its width. They then used the periodic changes in pulse width to infer the respiratory rate. While this method is simple to compute in the time domain, pulse wave width detection is susceptible to baseline drift and waveform distortion. Furthermore, its accuracy significantly decreases when poor peripheral circulation leads to weak signals.

[0011] In summary, researchers have conducted extensive studies on pulse wave respiratory signal extraction and respiratory rate calculation. These studies have improved extraction accuracy and anti-interference capabilities through signal decomposition, feature fusion, modality screening, and quality index optimization, combined with time-frequency analysis and intelligent algorithms. However, the following shortcomings still exist: 1. Existing pulse wave-based respiratory rate extraction methods generally rely excessively on single respiratory-induced variation features. Due to the lack of information dimensions, such strategies exhibit poor robustness when faced with non-steady-state fluctuations in signals caused by physiological variations or motion artifacts. Although some studies have attempted to improve performance by fusing multiple features, they have failed to incorporate the key parameter of respiratory-induced continuous systolic phase variation into their fusion system. Because it has unique robustness to signal defects such as leaky pulses and can provide complementary respiratory information, feature combination systems lacking this feature have significant bottlenecks in accuracy and reliability.

[0012] 2. Traditional methods based on simple differential or fixed threshold peak detection cannot adapt to dynamic fluctuations in signal amplitude when extracting respiratory-induced amplitude changes from pulse wave signals. This leads to systematic biases in the identification of effective pulses under low signal-to-noise ratio conditions. On the other hand, differential strategies that rely on extreme point identification will mislocate the start and end points of the pulse period when faced with baseline drift, causing fundamental errors in amplitude calculation. These inherent defects at the algorithm level result in a large number of non-physiological perturbations mixed in the extracted respiratory-induced amplitude change sequence, significantly degrading the signal-to-noise ratio.

[0013] 3. Methods for extracting respiratory induction intensity change signals from pulse waves based on empirical mode decomposition and its variants rely on cubic spline interpolation to fit the envelope of the signal extrema. This makes the algorithm highly sensitive to the distribution of extrema and noise interference, and prone to mode mixing and endpoint effects, affecting the stability and repeatability of the decomposition results. In addition, the setting of key parameters in the algorithm, such as the filtering scale and the screening threshold, lacks physiological basis. Using fixed values ​​often results in insufficient algorithm adaptability, while adopting a dynamic adjustment strategy inevitably introduces uncertainty and increases computational complexity.

[0014] 4. Frequency domain analysis methods for extracting respiratory frequency based on pulse wave signals usually rely on Fast Fourier Transform (FFT). However, since the time series composed of peaks, troughs, and other feature points extracted from the signal is inherently non-uniformly sampled, data interpolation and resampling must be performed before direct FFT analysis to make it uniform. This preprocessing process inevitably introduces low-pass filtering effects and spectral distortion, and amplifies the interference of signal quality problems such as pulse omission on the spectral structure, ultimately leading to a significant reduction in the estimation accuracy and robustness of respiratory frequency.

[0015] 5. Most solutions for fusing multiple features generally employ static fusion strategies with fixed weights or overly rely on a specific feature model. These methods cannot adaptively adjust to real-time changes in signal quality. When some features fail due to interference, the error will continuously contaminate the fusion result at a fixed proportion, resulting in severely insufficient system fault tolerance and generalization ability. To improve performance, some research has turned to end-to-end deep learning models, but their internal decision-making process is like a "black box," and users cannot trace the correlation between the final result and the contribution of each underlying physiological feature, leading to low credibility and clinical interpretability of the output results. Summary of the Invention

[0016] This invention aims to address the problems of insufficient accuracy, weak anti-interference ability, and poor robustness in existing pulse wave respiratory signal extraction and respiratory rate calculation technologies. It proposes a multi-feature, multi-dimensional collaborative optimization respiratory rate estimation method, which aims to improve the accuracy and anti-interference ability of respiratory rate estimation by integrating multi-dimensional respiratory induction features, adopting precise spectrum analysis methods and dynamic weight allocation mechanisms. This will enhance the interpretability and clinical applicability of the results, thereby providing more reliable technical support for non-invasive respiratory monitoring.

[0017] To achieve the above-mentioned objectives, the present invention adopts the following technical solution: The present invention provides a method for continuous estimation of respiratory rate based on pulse wave time-frequency domain feature fusion, characterized by the following steps: Step 1: Acquire the raw pulse wave signal and perform bandpass filtering and windowing to obtain a series of signal windows. The pulse wave signal sequence below ,in, Indicates the first n One signal window The i-th pulse wave signal; Indicates the total number of signal windows; This represents the total amount of pulse wave signal within a signal window; Step 2: Calculate the first... n One signal window The sequences of changes in respiratory induction intensity, respiratory induction amplitude, respiratory induction continuous contraction difference, and respiratory induction frequency were analyzed. Step 3: Based on the respiratory induction intensity change sequence, respiratory induction amplitude change sequence, respiratory induction continuous contraction difference change sequence, and respiratory induction frequency change sequence, calculate their respective respiratory rates, main peak significance, in-band signal-to-noise ratio, and harmonic energy ratios to construct the first... n One signal window The k-th four-dimensional vector ,in, express The kth respiratory rate, express The significance of the k-th main peak below, express The signal-to-noise ratio of the k-th band is below. express The energy ratio of the kth harmonic under the following conditions ; Step 4: Based on the first n One signal window The k-th four-dimensional vector Calculate the first n One signal window The k-th comprehensive confidence level and to The following is a comprehensive confidence sequence { | Perform a weighted average to obtain the first... n One signal window The initial fusion frequency And then After harmonic error correction, the corrected fusion frequency is time-smoothed and its rationality is checked, thus outputting the first... n One signal window The final frequency below And convert to the first n One signal window Breathing rate per minute .

[0018] The characteristic of the continuous respiratory rate estimation method based on pulse wave time-frequency domain feature fusion described in this invention is that step 2 includes the following steps: Step 2.1: Apply Gaussian kernel function to... Below Perform weighted average filtering to obtain the first... n One signal window The sequence of respiratory induction intensity changes and its corresponding time series ;in, Indicates the first n One signal window The next One respiratory induction intensity change value, express timestamp; Indicates the first n One signal window The total value of the change in respiratory induction intensity was extracted from the signal. Step 2.2: For Below Perform dynamic adaptive threshold and local slope analysis to obtain the first... n One signal window The sequence of respiratory induction amplitude changes and its corresponding non-uniform time series ;in, Indicates the first n One signal window The next The change in the amplitude of each respiratory induction, express timestamp; Indicates the first n One signal window The total value of the change in respiratory induction amplitude extracted from the signal; Step 2.3: Calculation Down Differential changes in respiratory-induced continuous contraction sequence and its corresponding non-uniform time series ;in, Indicates the first n One signal window The next The value of the difference in continuous contraction induced by respiration. express timestamp; Indicates the first n One signal window The total value of differential changes in respiratory-induced continuous contraction was extracted from the signal. Step 2.4: Calculation Down Respiratory induction frequency change sequence and its corresponding non-uniform sampling time series ;in, Indicates the first n One signal window The next Changes in respiratory induction rate, express timestamp; Indicates the first n One signal window The total value of the change in respiratory induction frequency is extracted from the signal.

[0019] Furthermore, step 2.1 includes the following steps: Step 2.1.1: Calculate the half-length of the Gaussian filter Thus, the filter length is obtained. ;in, This indicates the sampling frequency of the preprocessed signal. This represents the target analog cutoff frequency of the Gaussian low-pass filter. This represents the round-up operator; Step 2.1.2: Generate Gaussian window coefficients ,in, This represents the m-th filter coefficient. Indicates shape parameters; Step 2.1.3: Normalize the Gaussian window coefficients to obtain the normalized filter coefficients. ; This represents the m-th normalized filter coefficient; Step 2.1.4: Calculate the nth signal window mean pulse wave signal ; Step 2.1.5: Use equation (1) to... Perform a bisymmetric reflection boundary extension to obtain the first... n One signal window The next An expanded pulse wave signal Thus, the first n One signal window The extended pulse wave signal sequence below : (1) In equation (1), Indicates the first n One signal window The next A pulse wave signal, Indicates the first n One signal window The next One pulse wave signal; Step 2.1.6: For Perform a weighted moving average to obtain the first... n One signal window The sequence of respiratory induction intensity changes ,in, Indicates the first n One signal window The next An expanded pulse wave signal.

[0020] Furthermore, the separation of respiratory induction intensity changes in step 2.2 includes the following steps: Step 2.2.1: Use the IMS algorithm to identify the first... n One signal window Below Candidate pulse set And obtain the corresponding amplitude sequence. ,in, Indicates the first n One signal window The signal identified below is the first The amplitude of each candidate pulse, Indicates the first n One signal window The signal identified below is the first Peak values ​​of candidate pulse waves Indicates the first n One signal window The signal identified below is the first Candidate pulse wave trough values Indicates the first n One signal window The total number of candidate pulses identified from the signal below; Step 2.2.2: Use equation (2) to obtain the first... n One signal window Next The lower limit of the amplitude threshold and the upper limit of the amplitude threshold : (2) when When, then retain the first n One signal window The next A set of candidate pulses is obtained by filtering pulses based on amplitude thresholds. ,in, Indicates the first n One signal window The candidate pulse set after amplitude threshold filtering, Indicates the first The lower limit of an amplitude threshold, Indicates the first The upper limit of the amplitude threshold, and These represent the lower limit proportional coefficient and the upper limit proportional coefficient, respectively. This indicates the total number of candidate pulses after amplitude threshold filtering; Step 2.2.3: Calculate the first... n One signal window Next The median amplitude of the rising limb of the candidate pulses and the Symmetry ratio of each effective pulse ,in, Indicates the first n One signal window Next The absolute time of the start of the rising phase of each effective pulse. Indicates the first n One signal window Next The absolute time of the end of the rising phase of each effective pulse. Indicates the first n One signal window Next The amplitude of each effective pulse signal reaches Absolute time; Step 2.2.4: When When, then retain the first n One signal window Next A set of candidate pulses is obtained by analyzing several pulses. ,in, Indicates the first n One signal window The first after symmetry ratio screening One candidate pulse, This indicates that a minimum symmetry ratio has been preset. This indicates that the maximum value of the symmetry ratio is preset. Step 2.2.5: Calculate the first... n One signal window After filtering Respiratory induction amplitude change sequence of candidate pulses and its corresponding non-uniform time series ,in, Indicates the firstn One signal window The signal filtered from the following One effective pulse wave peak value, Indicates the first n One signal window The signal filtered from the following One effective pulse wave trough value, Indicates the first n One signal window Next The absolute time of the start of the rising phase of each effective pulse. Indicates the first n One signal window Next The absolute time of the end of the rising phase of each effective pulse.

[0021] Furthermore, step 3 includes the following steps: Step 3.1: [The text appears to be incomplete and contains several grammatical errors. A more accurate translation would require , , , Convert to the first n One signal window The respiratory rate corresponding to the intensity characteristics below Respiratory rate corresponding to amplitude characteristics Respiratory rate corresponding to continuous contraction differences respiratory rate corresponding to frequency characteristics ;Will , , , Any respiratory rate in the interval is denoted as the first. n One signal window The kth respiratory rate , ; Step 3.2: Based on power spectrum, power spectrum, power spectrum, The power spectrum of the first power spectrum is calculated. n One signal window The significance of the main peak in the power spectrum includes: the significance of the main peak corresponding to the intensity characteristics. The significance of the main peak corresponding to the amplitude characteristics The significance of the main peak corresponding to the characteristics of continuous contraction differences The significance of the main peak corresponding to the frequency characteristics ;Will , , , The significance of any main peak in the spectrum is denoted as the first peak. n One signal window Significance of the k-th main peak ; Step 3.3: Based on power spectrum, power spectrum, power spectrum, The power spectrum of the first power spectrum is calculated. n One signal window The in-band signal-to-noise ratio includes: the in-band signal-to-noise ratio corresponding to the intensity feature. In-band signal-to-noise ratio corresponding to amplitude characteristics Continuous contraction difference characteristics corresponding to in-band signal-to-noise ratio In-band signal-to-noise ratio corresponding to frequency characteristics ;Will , , , The significance of any main peak in the spectrum is denoted as the first peak. n One signal window The signal-to-noise ratio of the k-th band ; Step 3.4: Based on power spectrum, power spectrum, power spectrum, The power spectrum of the first power spectrum is calculated. n One signal window The harmonic energy ratio below includes: the harmonic energy ratio corresponding to the intensity characteristic. The harmonic energy ratio corresponding to the amplitude characteristics The characteristic of continuous contraction difference corresponds to the harmonic energy ratio The harmonic energy ratio corresponding to the frequency characteristics ;Will , , , The energy ratio of any harmonic in the denoted is denoted as the th harmonic. n One signal window The energy ratio of the kth harmonic under .

[0022] Furthermore, step 4 includes the following steps: Step 4.1: [The text appears to be incomplete and contains several grammatical errors. A more accurate translation would require , , and Any two respiratory rates in the sequence are denoted as the first. respiratory rate and the respiratory rate , , ; Calculate using equation (3) and the The closeness between : (3) In equation (3), Indicates the frequency consistency threshold; Construct a consistent adjacency matrix of respiratory rate using equation (4). : (4) In equation (4), Representing the adjacency matrix The Middle Line number Column elements; Step 4.2: Perform an adjacency analysis on the adjacency matrix. Perform a depth-first search to obtain the set of connected components, and denote the largest set of connected components in the set as . ,Will The number of respiratory rates contained therein is denoted as ; Step 4.3: Calculate the first... n One signal window The significance eigenvalue of the k-th main peak , Calculated in the n One signal window The k-th in-band signal-to-noise ratio eigenvalue , Calculate the first n One signal window Below , Calculate the first n One signal window The k-th comprehensive confidence level ,in, , , This represents three weight parameters; Step 4.4: When Furthermore, the combined confidence level of the two respiratory rates included in C is greater than the high confidence threshold. or When, calculate the first n One signal window Preliminary fusion result frequency ;in, Indicates the first n One signal window The b-th respiratory rate in C below; express The overall confidence level; when And the overall confidence level of at least one respiratory rate in C is less than or equal to or When, then take { | The respiratory rate with the highest overall confidence level was used as the first... n One signal window Preliminary fusion result frequency ; Step 4.5: If it exists , making Then let Otherwise, let ;in, Indicates the first n One signal window The respiratory rate value obtained after lower harmonic error correction; Step 4.6: Calculate the nth signal window Smoothing result frequency ,in, Denotes the smoothing factor; where, Indicates the (n-1)th signal window The smoothing result frequency; when season = ; Step 4.7: Calculate the first step using equation (5). n One signal window effective frequency : (5) In equation (5), Indicates the first n- 1 signal window The effective frequency below, and These represent the preset lower limit and upper limit of respiratory rate, respectively; when season = .

[0023] The present invention provides an electronic device, including a memory and a processor, characterized in that the memory is used to store a program supporting the processor in performing the method described therein, and the processor is configured to execute the program stored in the memory.

[0024] The present invention discloses a computer-readable storage medium storing a computer program, characterized in that the computer program is executed by a processor to perform the steps of the method described thereon.

[0025] Compared with the prior art, the beneficial effects of the present invention are reflected in the following aspects: 1. This invention constructs a multi-dimensional respiratory information fusion method by synergistically utilizing four dimensions of features in pulse wave signals: respiratory-induced amplitude variation, intensity variation, frequency variation, and continuous systolic phase variation. This method fundamentally overcomes the shortcomings of single-feature information dimensions being incomplete and having weak anti-interference capabilities, while also compensating for the key information gaps in existing feature combination strategies due to neglecting continuous systolic phase variation. When signal quality is ideal, multi-dimensional information cross-validation significantly improves estimation accuracy; it also exhibits strong robustness in the event of pulse omissions, baseline drift, or transient interference.

[0026] 2. This invention overcomes the systematic bias in identifying effective pulses under low signal-to-noise ratio environments by using a dynamic adaptive threshold mechanism; the pulse localization strategy based on local slope changes in waveforms eliminates the dependence on global extreme points and can accurately lock the pulse start and end points even under baseline drift, avoiding fundamental errors in amplitude calculation; a pulse waveform morphology verification step is introduced, which effectively identifies and eliminates non-physiological pulses that meet the amplitude-period conditions but have abnormal morphology by quantitatively analyzing the symmetry characteristics of the pulse rising branch, and finally outputs a high signal-to-noise ratio RIAV sequence.

[0027] 3. This invention replaces cubic spline interpolation with Gaussian filtering, which has optimal time-frequency aggregation, to obtain the instantaneous mean of the signal. This avoids the sensitivity of the envelope fitting process to extreme points and noise, significantly enhancing the robustness of the algorithm in principle. Simultaneously, this invention employs a fixed filtering scale determination method based on physiological priors about respiratory rate. By directly linking the filtering scale to explicit physiological parameters, it overcomes the uncertainty and computational complexity caused by dynamic adjustments in traditional methods. This design ensures that the algorithm's output results exhibit high stability and repeatability under different signal-to-noise ratio conditions and signal characteristics.

[0028] 4. This invention employs Lomb-Scargle periodograms to perform spectral analysis on the non-uniform sequence of pulse wave feature points. It directly estimates the power spectrum through least-squares fitting, avoiding the spectral distortion caused by forced resampling in traditional fast Fourier transforms, thus ensuring the accuracy of respiratory frequency component extraction from a mechanistic perspective. This method is naturally adaptable to irregular data point distributions and local missing data, maintaining spectral stability even when pulse waves are missed or signals are transiently distorted, significantly improving the accuracy and robustness of respiratory frequency estimation.

[0029] 5. This invention provides a respiratory rate estimation method based on multi-feature voting and dynamic attention fusion, effectively overcoming the limitations of existing technologies. The channel attention mechanism assigns appropriate weights to each physiological feature, achieving adaptive adjustment to signal quality fluctuations. When some features fail due to interference, it can automatically suppress their negative impact, thereby significantly improving the system's fault tolerance and generalization ability in complex scenarios. Furthermore, by retaining the independent estimation results of each feature and establishing a traceable voting decision mechanism, the final respiratory rate can be clearly traced back to the contribution weights of each underlying feature, greatly improving the interpretability and clinical credibility of the results while ensuring high accuracy. Attached Figure Description

[0030] Figure 1 This is a flowchart of the process framework of the present invention; Figure 2 This is a flowchart illustrating the RIAV extraction process of the present invention. Figure 3 This is a block diagram of the respiratory feature extraction network of the present invention. Detailed Implementation

[0031] In this embodiment, a continuous respiratory rate estimation method based on pulse wave time-frequency domain feature fusion has the characteristics of high accuracy and strong robustness, and can complete the respiratory rate estimation work based on pulse wave. Its main steps include: 1. Acquiring pulse wave signals and performing preprocessing; 2. Separating four non-uniform time series corresponding to different features from the filtered signal; 3. Converting the non-uniform time domain signal into frequency domain power spectral density to extract the respiratory rate; 4. Obtaining the final respiratory rate by confidence-weighted averaging. Specifically, as follows... Figure 1 As shown, the method is performed according to the following steps: Step 1: Acquire the raw pulse wave signal and perform bandpass filtering and windowing to obtain a series of signal windows. The pulse wave signal sequence below ,in, Indicates the first n One signal window The i-th pulse wave signal; Indicates the total number of signal windows; This represents the total amount of pulse wave signal within a signal window.

[0032] like Figure 3The algorithm for respiratory rate estimation based on pulse wave signals is fully presented. The core consists of three functional modules: First, the raw signal is analyzed in the time domain to extract four types of respiratory-induced time-domain features: pulse wave amplitude variation, intensity variation, frequency variation, and continuous contraction difference variation. Then, these four types of features are analyzed in the frequency domain and processed by spectrum to obtain the respiratory rate and signal quality index combination of the corresponding features. Finally, based on the signal quality index, weights are assigned to the results of each feature to calculate the initial respiratory rate. After time smoothing, the final number of breaths per minute is output. The entire process achieves accurate estimation of respiratory rate through parallel extraction of multiple features and quality-weighted fusion.

[0033] Step 2: Calculate the first... n One signal window The sequences of changes in respiratory induction intensity, respiratory induction amplitude, respiratory induction continuous contraction difference, and respiratory induction frequency were analyzed. Step 2.1: Apply Gaussian kernel function to... Below Perform weighted average filtering to obtain the first... n One signal window The sequence of respiratory induction intensity changes and its corresponding time series ;in, Indicates the first n One signal window The next One respiratory induction intensity change value, express timestamp; Indicates the first n One signal window The total value of the change in respiratory induction intensity is extracted from the signal.

[0034] Step 2.1.1: Calculate the half-length of the Gaussian filter Thus, the filter length is obtained. ;in, This indicates the sampling frequency of the preprocessed signal. This represents the target analog cutoff frequency of the Gaussian low-pass filter. This represents the round-up operator; Step 2.1.2: Generate Gaussian window coefficients ,in, This represents the m-th filter coefficient. Indicates shape parameters; Step 2.1.3: Normalize the Gaussian window coefficients to obtain the normalized filter coefficients. ; This represents the m-th normalized filter coefficient.

[0035] Step 2.1.4: Calculate the nth signal window mean pulse wave signal ; Step 2.1.5: Use equation (1) to... Perform a bisymmetric reflection boundary extension to obtain the first... n One signal window The next An expanded pulse wave signal Thus, the first n One signal window The extended pulse wave signal sequence below : (1) In equation (1), Indicates the first n One signal window The next A pulse wave signal, Indicates the first n One signal window The next A pulse wave signal.

[0036] Step 2.1.6: For Perform a weighted moving average to obtain the first... n One signal window The sequence of respiratory induction intensity changes ,in, Indicates the first n One signal window The next An expanded pulse wave signal.

[0037] Step 2.2: As Figure 2 As shown, for Below Perform dynamic adaptive threshold and local slope analysis to obtain the first... n One signal window The sequence of respiratory induction amplitude changes and its corresponding non-uniform time series ;in, Indicates the first n One signal window The next The change in the amplitude of each respiratory induction, express timestamp; Indicates the first n One signal window The total value of the change in respiratory induction amplitude is extracted from the signal.

[0038] Step 2.2.1: Use the IMS algorithm to identify the first... n One signal window Below Candidate pulse set And obtain the corresponding amplitude sequence. ,in, Indicates the first n One signal window The signal identified below is the first The amplitude of each candidate pulse, Indicates the first n One signal window The signal identified below is the first Peak values ​​of candidate pulse waves Indicates the first n One signal window The signal identified below is the first Candidate pulse wave trough values Indicates the first n One signal window The total number of candidate pulses identified from the signal.

[0039] Step 2.2.2: Use equation (2) to obtain the first... n One signal window Next The lower limit of the amplitude threshold and the upper limit of the amplitude threshold : (2) when When, then retain the first n One signal window The next A set of candidate pulses is obtained by filtering pulses based on amplitude thresholds. ,in, Indicates the first n One signal window The candidate pulse set after amplitude threshold filtering, Indicates the first The lower limit of an amplitude threshold, Indicates the first The upper limit of the amplitude threshold, and These represent the lower limit proportional coefficient and the upper limit proportional coefficient, respectively. This indicates the total number of candidate pulses after amplitude threshold filtering; Step 2.2.3: Calculate the first... n One signal window Next The median amplitude of the rising limb of the candidate pulses and the Symmetry ratio of each effective pulse ,in, Indicates the first n One signal window Next The absolute time of the start of the rising phase of each effective pulse. Indicates the first n One signal window Next The absolute time of the end of the rising phase of each effective pulse. Indicates the first n One signal window Next The amplitude of each effective pulse signal reaches Absolute time.

[0040] Step 2.2.4: When When, then retain the first n One signal window Next A set of candidate pulses is obtained by analyzing several pulses. ,in, Indicates the first n One signal window The first after symmetry ratio screening One candidate pulse, This indicates that a minimum symmetry ratio has been preset. This indicates that the maximum value of the symmetry ratio is preset.

[0041] Step 2.2.5: Calculate the first... n One signal window After filtering Respiratory induction amplitude change sequence of candidate pulses and its corresponding non-uniform time series ,in, Indicates the first n One signal window The signal filtered from the following One effective pulse wave peak value, Indicates the first n One signal window The signal filtered from the following One effective pulse wave trough value, Indicates the first n One signal window Next The absolute time of the start of the rising phase of each effective pulse. Indicates the first nOne signal window Next The absolute time of the end of the rising phase of each effective pulse.

[0042] Step 2.3: Calculation Down Differential changes in respiratory-induced continuous contraction sequence and its corresponding non-uniform time series ;in, Indicates the first n One signal window The next The value of the difference in continuous contraction induced by respiration. express timestamp; Indicates the first n One signal window The total value of the differential change in continuous contraction induced by respiration was extracted from the signal.

[0043] Step 2.4: Calculation Down Respiratory induction frequency change sequence and its corresponding non-uniform sampling time series ;in, Indicates the first n One signal window The next Changes in respiratory induction rate, express timestamp; Indicates the first n One signal window The total value of the change in respiratory induction frequency is extracted from the signal.

[0044] Step 3: As Figure 3 As shown, based on the respiratory induction intensity change sequence, respiratory induction amplitude change sequence, respiratory induction continuous contraction difference change sequence, and respiratory induction frequency change sequence, the respiratory frequency, main peak significance, in-band signal-to-noise ratio, and harmonic energy ratio are calculated respectively, thereby constructing the first... n One signal window The k-th four-dimensional vector ,in, express The kth respiratory rate, express The significance of the k-th main peak below, express The signal-to-noise ratio of the k-th band is below. express The energy ratio of the kth harmonic under the following conditions .

[0045] Step 3.1: [The text appears to be incomplete and contains several grammatical errors. A more accurate translation would require , , , Convert to the first n One signal window The respiratory rate corresponding to the intensity characteristics below Respiratory rate corresponding to amplitude characteristics Respiratory rate corresponding to continuous contraction differences respiratory rate corresponding to frequency characteristics ;Will , , , Any respiratory rate in the interval is denoted as the first. n One signal window The kth respiratory rate .

[0046] Step 3.2: Based on power spectrum, power spectrum, power spectrum, The power spectrum of the first power spectrum is calculated. n One signal window The significance of the main peak in the power spectrum includes: the significance of the main peak corresponding to the intensity characteristics. The significance of the main peak corresponding to the amplitude characteristics The significance of the main peak corresponding to the characteristics of continuous contraction differences The significance of the main peak corresponding to the frequency characteristics ;Will , , , The significance of any main peak in the spectrum is denoted as the first peak. n One signal window Significance of the k-th main peak .

[0047] Step 3.3: Based on power spectrum, power spectrum, power spectrum, The power spectrum of the first power spectrum is calculated. n One signal window The in-band signal-to-noise ratio includes: the in-band signal-to-noise ratio corresponding to the intensity feature. In-band signal-to-noise ratio corresponding to amplitude characteristics Continuous contraction difference characteristics corresponding to in-band signal-to-noise ratio In-band signal-to-noise ratio corresponding to frequency characteristics ;Will , , , The significance of any main peak in the spectrum is denoted as the first peak. n One signal window The signal-to-noise ratio of the k-th band .

[0048] Step 3.4: Based on power spectrum, power spectrum, power spectrum, The power spectrum of the first power spectrum is calculated. n One signal window The harmonic energy ratio below includes: the harmonic energy ratio corresponding to the intensity characteristic. The harmonic energy ratio corresponding to the amplitude characteristics The characteristic of continuous contraction difference corresponds to the harmonic energy ratio The harmonic energy ratio corresponding to the frequency characteristics ;Will , , , The energy ratio of any harmonic in the denoted is denoted as the th harmonic. n One signal window The energy ratio of the kth harmonic under .

[0049] Step 4: As Figure 3 As shown, based on the first n One signal window The k-th four-dimensional vector Calculate the first n One signal window The k-th comprehensive confidence level and to The following is a comprehensive confidence sequence { | Perform a weighted average to obtain the first... n One signal window The initial fusion frequency And then After harmonic error correction, the corrected fusion frequency is time-smoothed and its rationality is checked, thus outputting the first... n One signal window The final frequency below And convert to the first n One signal window Breathing rate per minute .

[0050] Step 4.1: [The text appears to be incomplete and contains several grammatical errors. A more accurate translation would require , , and Any two respiratory rates in the sequence are denoted as the first. respiratory rate and the respiratory rate , , ; Calculate using equation (3) and the The closeness between : (3) In equation (3), This represents the frequency consistency threshold.

[0051] Construct a consistent adjacency matrix of respiratory rate using equation (4). : (4) In equation (4), Representing the adjacency matrix The Middle Line number The elements of the column.

[0052] Step 4.2: Perform an adjacency analysis on the adjacency matrix. Perform a depth-first search to obtain the set of connected components, and denote the largest set of connected components in the set as . ,Will The number of respiratory rates contained therein is denoted as .

[0053] Step 4.3: Calculate the first... n One signal window The significance eigenvalue of the k-th main peak , Calculated in the n One signal window The k-th in-band signal-to-noise ratio eigenvalue , Calculate the first n One signal window Below , Calculate the first n One signal window The k-th comprehensive confidence level ,in, , , This represents three weight parameters.

[0054] Step 4.4: When Furthermore, the combined confidence level of the two respiratory rates included in C is greater than the high confidence threshold. or When, calculate the first n One signal window Preliminary fusion result frequency ;in, Indicates the first n One signal window The b-th respiratory rate in C below; express The overall confidence level; when And the overall confidence level of at least one respiratory rate in C is less than or equal to or When, then take { | The respiratory rate with the highest overall confidence level was used as the first... n One signal window Preliminary fusion result frequency .

[0055] Step 4.5: If it exists , making Then let Otherwise, let ;in, Indicates the first n One signal window The respiratory rate value obtained after lower harmonic error correction; Step 4.6: Calculate the nth signal window Smoothing result frequency ,in, Denotes the smoothing factor; where, Indicates the (n-1)th signal window The smoothing result frequency; when season = .

[0056] Step 4.7: Calculate the first step using equation (5). n One signal window effective frequency : (5) In equation (5), Indicates the first n- 1 signal window The effective frequency below, and These represent the preset lower limit and upper limit of respiratory rate, respectively; when season = .

[0057] In this embodiment, an electronic device includes a memory and a processor. The memory stores a program that supports the processor in executing the above-described method, and the processor is configured to execute the program stored in the memory.

[0058] In this embodiment, a computer-readable storage medium stores a computer program, which is executed by a processor to perform the steps of the above method.

Claims

1. A method for continuous estimation of respiratory rate based on the fusion of time-frequency domain features of pulse waves, characterized in that, The procedure is as follows: Step 1: Acquire the raw pulse wave signal and perform bandpass filtering and windowing to obtain a series of signal windows. The pulse wave signal sequence below ,in, Indicates the first n One signal window The i-th pulse wave signal; Indicates the total number of signal windows; This represents the total amount of pulse wave signal within a signal window; Step 2: Calculate the first... n One signal window The sequences of changes in respiratory induction intensity, respiratory induction amplitude, respiratory induction continuous contraction difference, and respiratory induction frequency were analyzed. Step 3: Based on the respiratory induction intensity change sequence, respiratory induction amplitude change sequence, respiratory induction continuous contraction difference change sequence, and respiratory induction frequency change sequence, calculate their respective respiratory rates, main peak significance, in-band signal-to-noise ratio, and harmonic energy ratios to construct the first... n One signal window The k-th four-dimensional vector ,in, express The kth respiratory rate, express The significance of the k-th main peak below, express The signal-to-noise ratio of the k-th band is below. express The energy ratio of the kth harmonic under the following conditions ; Step 4: Based on the first n One signal window The k-th four-dimensional vector Calculate the first n One signal window The k-th comprehensive confidence level and to The following is a comprehensive confidence sequence { | Perform a weighted average to obtain the first... n One signal window The initial fusion frequency And then After harmonic error correction, the corrected fusion frequency is time-smoothed and its rationality is checked, thus outputting the first... n One signal window The final frequency below And convert to the first n One signal window Breathing rate per minute .

2. The method for continuous estimation of respiratory rate based on pulse wave time-frequency domain feature fusion according to claim 1, characterized in that, Step 2 includes the following steps: Step 2.1: Apply Gaussian kernel function to... Below Perform weighted average filtering to obtain the first... n One signal window The sequence of respiratory induction intensity changes and its corresponding time series ;in, Indicates the first n One signal window The next One respiratory induction intensity change value, express timestamp; Indicates the first n One signal window The total value of the change in respiratory induction intensity was extracted from the signal. Step 2.2: For Below Perform dynamic adaptive threshold and local slope analysis to obtain the first... n One signal window The sequence of respiratory induction amplitude changes and its corresponding non-uniform time series ;in, Indicates the first n One signal window The next The change in the amplitude of each respiratory induction, express timestamp; Indicates the first n One signal window The total value of the change in respiratory induction amplitude extracted from the signal; Step 2.3: Calculation Down Differential changes in respiratory-induced continuous contraction sequence and its corresponding non-uniform time series ;in, Indicates the first n One signal window The next The value of the difference in continuous contraction induced by respiration. express timestamp; Indicates the first n One signal window The total value of differential changes in respiratory-induced continuous contraction was extracted from the signal. Step 2.4: Calculation Down Respiratory induction frequency change sequence and its corresponding non-uniform sampling time series ;in, Indicates the first n One signal window The next Changes in respiratory induction rate, express timestamp; Indicates the first n One signal window The total value of the change in respiratory induction frequency is extracted from the signal.

3. The method for continuous estimation of respiratory rate based on pulse wave time-frequency domain feature fusion according to claim 2, characterized in that, Step 2.1 includes the following steps: Step 2.1.1: Calculate the half-length of the Gaussian filter Thus, the filter length is obtained. ;in, This indicates the sampling frequency of the preprocessed signal. This represents the target analog cutoff frequency of the Gaussian low-pass filter. This represents the round-up operator; Step 2.1.2: Generate Gaussian window coefficients ,in, This represents the m-th filter coefficient. Indicates shape parameters; Step 2.1.3: Normalize the Gaussian window coefficients to obtain the normalized filter coefficients. ; This represents the m-th normalized filter coefficient; Step 2.1.4: Calculate the nth signal window mean pulse wave signal ; Step 2.1.5: Use equation (1) to... Perform a bisymmetric reflection boundary extension to obtain the first... n One signal window The next An expanded pulse wave signal Thus, the first n One signal window The extended pulse wave signal sequence below : (1) In equation (1), Indicates the first n One signal window The next A pulse wave signal, Indicates the first n One signal window The next One pulse wave signal; Step 2.1.6: For Perform a weighted moving average to obtain the first... n One signal window The sequence of respiratory induction intensity changes ,in, Indicates the first n One signal window The next An expanded pulse wave signal.

4. The method for continuous estimation of respiratory rate based on pulse wave time-frequency domain feature fusion according to claim 3, characterized in that, Step 2.2, which involves isolating changes in respiratory induction intensity, includes the following steps: Step 2.2.1: Use the IMS algorithm to identify the first... n One signal window Below Candidate pulse set And obtain the corresponding amplitude sequence. ,in, Indicates the first n One signal window The signal identified below is the first The amplitude of each candidate pulse, Indicates the first n One signal window The signal identified below is the first Peak values ​​of candidate pulse waves Indicates the first n One signal window The signal identified below is the first Candidate pulse wave trough values Indicates the first n One signal window The total number of candidate pulses identified from the signal below; Step 2.2.2: Use equation (2) to obtain the first... n One signal window Next The lower limit of the amplitude threshold and the upper limit of the amplitude threshold : (2) when When, then retain the first n One signal window The next A set of candidate pulses is obtained by filtering pulses based on amplitude thresholds. ,in, Indicates the first n One signal window The candidate pulse set after amplitude threshold filtering, Indicates the first The lower limit of an amplitude threshold, Indicates the first The upper limit of the amplitude threshold, and These represent the lower limit proportional coefficient and the upper limit proportional coefficient, respectively. This indicates the total number of candidate pulses after amplitude threshold filtering; Step 2.2.3: Calculate the first... n One signal window Next The median amplitude of the rising limb of the candidate pulses and the Symmetry ratio of each effective pulse ,in, Indicates the first n One signal window Next The absolute time of the start of the rising phase of each effective pulse. Indicates the first n One signal window Next The absolute time of the end of the rising phase of each effective pulse. Indicates the first n One signal window Next The amplitude of each effective pulse signal reaches Absolute time; Step 2.2.4: When When, then retain the first n One signal window Next A set of candidate pulses is obtained by analyzing several pulses. ,in, Indicates the first n One signal window The first after symmetry ratio screening One candidate pulse, This indicates that a minimum symmetry ratio has been preset. This indicates that the maximum value of the symmetry ratio is preset. Step 2.2.5: Calculate the first... n One signal window After filtering Respiratory induction amplitude change sequence of candidate pulses and its corresponding non-uniform time series ,in, Indicates the first n One signal window The signal filtered from the following One effective pulse wave peak value, Indicates the first n One signal window The signal filtered from the following One effective pulse wave trough value, Indicates the first n One signal window Next The absolute time of the start of the rising phase of each effective pulse. Indicates the first n One signal window Next The absolute time of the end of the rising phase of each effective pulse.

5. The method for continuous estimation of respiratory rate based on pulse wave time-frequency domain feature fusion according to claim 1, characterized in that, Step 3 includes the following steps: Step 3.1: [The text appears to be incomplete and contains several grammatical errors. A more accurate translation would require , , , Convert to the first n One signal window The respiratory rate corresponding to the intensity characteristics below Respiratory rate corresponding to amplitude characteristics Respiratory rate corresponding to continuous contraction differences respiratory rate corresponding to frequency characteristics ;Will , , , Any respiratory rate in the interval is denoted as the first. n One signal window The kth respiratory rate , ; Step 3.2: Based on power spectrum, power spectrum, power spectrum, The power spectrum of the first power spectrum is calculated. n One signal window The significance of the main peak in the power spectrum includes: the significance of the main peak corresponding to the intensity characteristics. The significance of the main peak corresponding to the amplitude characteristics The significance of the main peak corresponding to the characteristics of continuous contraction differences The significance of the main peak corresponding to the frequency characteristics ;Will , , , The significance of any main peak in the spectrum is denoted as the first peak. n One signal window Significance of the k-th main peak ; Step 3.3: Based on power spectrum, power spectrum, power spectrum, The power spectrum of the first power spectrum is calculated. n One signal window The in-band signal-to-noise ratio includes: the in-band signal-to-noise ratio corresponding to the intensity feature. In-band signal-to-noise ratio corresponding to amplitude characteristics Continuous contraction difference characteristics corresponding to in-band signal-to-noise ratio In-band signal-to-noise ratio corresponding to frequency characteristics ;Will , , , The significance of any main peak in the spectrum is denoted as the first peak. n One signal window The signal-to-noise ratio of the k-th band ; Step 3.4: Based on power spectrum, power spectrum, power spectrum, The power spectrum of the first power spectrum is calculated. n One signal window The harmonic energy ratio below includes: the harmonic energy ratio corresponding to the intensity characteristic. The harmonic energy ratio corresponding to the amplitude characteristics The characteristic of continuous contraction difference corresponds to the harmonic energy ratio The harmonic energy ratio corresponding to the frequency characteristics ;Will , , , The energy ratio of any harmonic in the denoted is denoted as the th harmonic. n One signal window The energy ratio of the kth harmonic under .

6. The method for continuous estimation of respiratory rate based on pulse wave time-frequency domain feature fusion according to claim 1, characterized in that, Step 4 includes the following steps: Step 4.1: [The text appears to be incomplete and contains several grammatical errors. A more accurate translation would require , , and Any two respiratory rates in the sequence are denoted as the first. respiratory rate and the respiratory rate , , ; Calculate using equation (3) and the The closeness between : (3) In equation (3), Indicates the frequency consistency threshold; Construct a consistent adjacency matrix of respiratory rate using equation (4). : (4) In equation (4), Representing the adjacency matrix The Middle Line number Column elements; Step 4.2: Perform an adjacency analysis on the adjacency matrix. Perform a depth-first search to obtain the set of connected components, and denote the largest set of connected components in the set as . ,Will The number of respiratory rates contained therein is denoted as ; Step 4.3: Calculate the first... n One signal window The significance eigenvalue of the k-th main peak , Calculated in the n One signal window The k-th in-band signal-to-noise ratio eigenvalue , Calculate the first n One signal window Below , Calculate the first n One signal window The k-th comprehensive confidence level ,in, , , This represents three weight parameters; Step 4.4: When Furthermore, the combined confidence level of the two respiratory rates included in C is greater than the high confidence threshold. or When, calculate the first n One signal window Preliminary fusion result frequency ;in, Indicates the first n One signal window The b-th respiratory rate in C below; express The overall confidence level; when And the overall confidence level of at least one respiratory rate in C is less than or equal to or When, then take { | The respiratory rate with the highest overall confidence level was used as the first... n One signal window Preliminary fusion result frequency ; Step 4.5: If it exists , making Then let Otherwise, let ;in, Indicates the first n One signal window The respiratory rate value obtained after lower harmonic error correction; Step 4.6: Calculate the nth signal window Smoothing result frequency ,in, Denotes the smoothing factor; where, Indicates the (n-1)th signal window The smoothing result frequency; when season = ; Step 4.7: Calculate the first step using equation (5). n One signal window effective frequency : (5) In equation (5), Indicates the first n- 1 signal window The effective frequency below, and These represent the preset lower limit and upper limit of respiratory rate, respectively; when season = .

7. An electronic device, comprising a memory and a processor, characterized in that, The memory is used to store a program that supports a processor in executing the method of any one of claims 1-6, the processor being configured to execute the program stored in the memory.

8. A computer-readable storage medium storing a computer program thereon, characterized in that, The computer program is executed by a processor to perform the steps of the method according to any one of claims 1-6.

Images