Blood oxygen estimation method based on combination of successive variational mode decomposition and adaptive filtering

Abstract

The invention belongs to the technical field of physiological state estimation, and discloses a blood oxygen estimation method based on combination of successive variational mode decomposition and adaptive filtering, which comprises the following steps of: removing high-frequency noise and baseline drift from a collected pulse wave signal; performing successive variational mode decomposition and adaptive filtering on the pulse wave signal to remove motion artifacts; accurate peak values and valley values are extracted according to the denoised pulse waves; the R value is calculated according to the denoised pulse wave and the extracted peak value and valley value, the blood oxygen saturation degree is solved according to the R value, and the R value refers to the absorbance ratio of the red light and the infrared light in the blood. According to the blood oxygen estimation method based on the combination of the successive variational mode decomposition and the self-adaptive filtering, the effect of removing motion artifact noise of photoelectric volume pulse wave signals is achieved, multi-peak-value joint detection is used, peak values and valley values are accurately extracted, and therefore the R value is determined, the blood oxygen saturation degree is obtained, and the accuracy of blood oxygen estimation is improved. And the precision of estimating the blood oxygen saturation is greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of physiological state estimation, and in particular to a blood oxygen estimation method based on the combination of successive variational modal decomposition and adaptive filtering. Background Art

[0002] Blood oxygen saturation (SpO 2 ) refers to the saturation of oxygen in the blood, that is, the proportion of oxygen molecules combined with hemoglobin, usually expressed as a percentage. Under normal circumstances, the level of blood oxygen saturation should be between 95% and 100%, indicating that the hemoglobin in the blood is almost fully saturated with oxygen. SpO 2 It is a vital physiological parameter. By monitoring blood oxygen saturation, medical staff can understand the body's oxygenation status in real time and thus evaluate the function of the patient's respiratory and circulatory systems.

[0003] Photoplethysmography (PPG) is used to estimate blood oxygen saturation (SpO 2 ) is a widely used non-invasive measurement method. PPG technology is based on the principle of light absorption. By measuring the difference in absorption of light of different wavelengths in subcutaneous tissue and blood, it can effectively estimate the oxygen saturation in the blood. Compared with traditional arterial blood gas analysis, the PPG measurement process is simple, real-time and non-invasive, so it is widely used in portable devices and daily health monitoring.

[0004] Although PPG technology has significant advantages in blood oxygen saturation monitoring, it still faces some challenges in practical applications. First, PPG signals are easily interfered by motion artifacts, especially when in motion. PPG devices on fingers or wrists are prone to increased signal noise due to motion, affecting measurement accuracy. Therefore, signal processing and noise suppression techniques are usually required in algorithm design to improve measurement stability.

[0005] According to investigation and understanding, the currently available technologies do not have a special algorithm designed to remove motion artifacts that most seriously affect PPG signals. Using only adaptive filters cannot completely remove artifacts, which will cause a large amount of noise in the pulse wave, thereby affecting the accuracy of blood oxygen saturation estimation. Summary of the invention

[0006] The purpose of the present invention is to provide a blood oxygen estimation method based on the combination of successive variational modal decomposition and adaptive filtering, so as to achieve the effect of removing motion artifact noise of photoplethysmographic signals, use multi-peak joint detection and set thresholds to extract accurate peak values ​​and valley values, and then obtain more stable DC and AC components through multi-cycle signals, so as to determine the R value required for calculating blood oxygen saturation, and directly bring the R value into the Maxim equation to obtain blood oxygen saturation, thereby greatly improving the accuracy of estimating blood oxygen saturation.

[0007] To achieve the above object, the present invention provides a blood oxygen estimation method based on successive variational mode decomposition and adaptive filtering, comprising the following steps:

[0008] Step S1, removing high-frequency noise and baseline drift from the collected photoplethysmography signal PPG;

[0009] Step S2, using successive variational modal decomposition and adaptive filtering to remove motion artifacts from the PPG signal;

[0010] Step S3, extracting accurate peak values ​​and valley values ​​according to the denoised PPG signal;

[0011] Step S4, calculating the R value based on the denoised PPG signal and the extracted peak and valley values, and obtaining the blood oxygen saturation based on the R value, wherein the R value refers to the absorbance ratio of red light and infrared light in the blood.

[0012] Preferably, in step S1, the collected PPG signal is preprocessed to remove high and low frequency noise and baseline drift. The specific process is as follows:

[0013] Step S11, setting a sliding window and window overlap, processing the collected PPG signal and acceleration signal in blocks, and analyzing the time variation characteristics of the signal;

[0014] Step S12: For the segmented PPG data, use a Butterworth bandpass filter to filter to remove the noise of the high and low frequency environments, and use an autoregressive model to fit the trend of the signal, and then subtract the fitted trend from the signal to obtain a PPG signal after removing the baseline drift.

[0015] Preferably, in step S2, the successive variational mode decomposition is an improvement based on the variational mode decomposition, wherein the objective function of the variational mode decomposition is as follows:

[0016]

[0017] Where k is the number of modes to be decomposed, u k (t) is the kth mode, w kis the center frequency of the kth mode; f(t) is the original signal, and j represents the imaginary unit;

[0018] The objective function of variational mode decomposition is expressed as an optimization problem with constraints and solved. The reconstruction condition is added, that is, the sum of all modes is equal to the original signal f(t), as shown below:

[0019]

[0020] The VMD optimization problem is then expressed in Lagrangian form with constraints as follows:

[0021]

[0022] Among them, α is a weight factor used to balance the reconstruction error and bandwidth constraint, and γ is a Lagrange multiplier used to introduce the reconstruction constraint.

[0023] Preferably, the PPG signal is subjected to a method combining successive variational modal decomposition and adaptive filtering to remove motion artifacts, and the specific process is as follows:

[0024] First, based on variational mode decomposition, successive variational mode decomposition uses a step-by-step decomposition strategy to process signals. The basic idea is to gradually separate the modal components of different frequency bands from complex signals by successive decomposition until the decomposition of the entire signal is completed.

[0025] Then, the Pearson correlation coefficient between the separated harmonic signal and the triaxial acceleration is calculated as follows:

[0026]

[0027] Among them, pearson represents the Pearson correlation coefficient, X i Indicates harmonic data, Y i represents acceleration data, represents the mean of the harmonic data, represents the mean value of acceleration data;

[0028] Finally, the original signal and the motion artifact signal are passed through a normalized minimum mean square filter to obtain the PPG signal after removing the motion artifact; wherein the input is the PPG signal, the reference signal is the motion artifact, and the output is the PPG signal after removing the motion artifact.

[0029] Preferably, each decomposed harmonic will have three Pearson correlation coefficients for the three-axis acceleration, and the harmonics with Pearson correlation coefficients greater than the threshold are selected for reconstruction, that is, the signals are directly superimposed and the parts of the harmonics that are more correlated with the three-axis acceleration are retained, thereby obtaining most of the motion artifact signals in the original signal.

[0030] Preferably, in step S3, a multiple detection method combined strategy is used to perform peak and valley detection according to the denoised PPG signal;

[0031] Among them, the multiple peak detection method includes but is not limited to: local maximum method, first-order derivative method of adaptive threshold method, slope and function method of adaptive threshold method, and bandpass filtering method.

[0032] Preferably, in step S4, the R value is calculated according to the denoised PPG signal and the extracted peak and valley values, and the blood oxygen saturation is obtained according to the R value, including the following steps:

[0033] Step S41, extracting AC and DC components from the PPG signal;

[0034] Step S42, calculating the R value according to the obtained AC and DC components;

[0035] Step S43: Based on the R value, the blood oxygen saturation is calculated using an empirical formula.

[0036] Preferably, in step S41, AC and DC components are extracted, and the specific process is as follows:

[0037] Step S411: In a complete heartbeat cycle, the DC component is obtained by taking the average value of the signal, as shown below:

[0038]

[0039] Where n is the number of data points in the window, and i is the value of the PPG signal at the i-th sampling point;

[0040] Step S412: Calculate the AC component by the peak-to-valley difference of the PPG signal, as shown below:

[0041] AC = Peak-Valley;

[0042] Among them, Peak is the peak value of the signal within a heartbeat cycle; Valley is the valley value of the signal within a heartbeat cycle.

[0043] Preferably, in step S42, the R value is calculated based on the obtained AC and DC components as follows:

[0044]

[0045] Among them, the numerator is the ratio of the AC component to the DC component of the red light wavelength; the denominator is the ratio of the AC component to the DC component of the infrared light wavelength; the R value reflects the relative rate of change of the red light and infrared light signals.

[0046] Preferably, in step S43, based on the empirical relationship between blood oxygen saturation and R value, an empirical formula is used to convert the R value into blood oxygen saturation, as shown below:

[0047] SpO 2 =A*R 2 +B*R+C;

[0048] Among them, A, B, and C are constants, which are usually obtained based on a large amount of experimental data.

[0049] Therefore, the present invention adopts the above blood oxygen estimation method based on successive variational mode decomposition and adaptive filtering, and the beneficial effects are as follows:

[0050] (1) Aiming at the problem of motion artifacts in the photoplethysmography signal, an effective denoising model combining successive variational mode decomposition and adaptive filtering was established. The original photoplethysmography signal was decomposed into many harmonics through gradual variational mode decomposition. The correlation coefficient between each harmonic and the three-axis acceleration was calculated. The coefficient threshold was set to reconstruct the harmonics with correlation coefficients greater than the threshold to form harmonics containing motion artifacts. The photoplethysmography signal with high-frequency noise and baseline drift removed and the harmonics containing motion artifacts were filtered through a normalized minimum mean square filter to obtain a clean photoplethysmography signal.

[0051] (2) Use multi-peak joint detection and set thresholds to extract relatively accurate peak and valley values, calculate the DC and AC components of the photoplethysmographic pulse wave, and thus obtain the R value. Substitute the obtained R value into the Maxim equation to obtain a higher-precision blood oxygen saturation value.

[0052] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 The flowchart of the blood oxygen estimation method based on the combination of successive variational mode decomposition and adaptive filtering of the present invention;

[0054] Figure 2 A schematic diagram of data division in an embodiment of the present invention;

[0055] Figure 3 This is a noise removal flow chart in an embodiment of the present invention;

[0056] Figure 4 A schematic diagram of normalized minimum mean square filter noise removal in an embodiment of the present invention;

[0057] Figure 5 This is a diagram showing the effect of the joint peak detection algorithm in an embodiment of the present invention. DETAILED DESCRIPTION

[0058] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.

[0059] like Figure 1 As shown, the blood oxygen estimation method based on the combination of successive variational mode decomposition and adaptive filtering in the present invention includes the following steps:

[0060] Step S1: pre-process the collected photoplethysmography (PPG) signal to remove high and low frequency noise and baseline drift.

[0061] Step S11, setting a sliding window and window overlap, and processing the collected PPG signal and acceleration signal in blocks to effectively analyze the time variation characteristics of the signal and improve the efficiency of signal processing.

[0062] The PPG signal is a physiological signal that changes over time, and its characteristics (such as heart rate, blood oxygen saturation, etc.) may vary in different time periods. By dividing the signal into smaller windows and performing local analysis on the signal in each window, the amount of data required for each analysis is reduced, and the computational efficiency is improved, thereby more accurately capturing the temporal dynamic characteristics of the signal and avoiding the need for a large amount of computing resources and time when analyzing the entire PPG signal.

[0063] The purpose of setting window overlap is to avoid information loss or discontinuity at the window boundary. By allowing a certain percentage of overlap between adjacent windows, the continuity of the signal is ensured, and the impact of boundary effects on signal analysis is reduced to a certain extent. In some cases, important signal features may cross the window boundary. By setting window overlap, the probability of capturing cross-border features is increased.

[0064] Step S12: For the segmented PPG data, use a Butterworth bandpass filter to filter to remove the noise of the high and low frequency environments, and use an autoregressive model to fit the trend of the signal, and then subtract the fitted trend from the signal to obtain a PPG signal after removing the baseline drift.

[0065] During the acquisition process, PPG signals will generate various noises due to the external environment, equipment characteristics and human factors, including high and low frequency noise and baseline drift.

[0066] Step S121: High and low frequency noise will affect the clarity of the signal, making it difficult to identify the details of the PPG waveform, affecting the stability of the signal and the accuracy of feature extraction. For the segmented PPG data, a Butterworth bandpass filter is used to filter and remove the noise of the high and low frequency environment to retain the effective components of the PPG signal and improve the signal quality.

[0067] For an ideal N-order bandpass Butterworth filter, its frequency response function H(w) is as follows:

[0068]

[0069] Where w is the angular frequency of the signal; w c is the cutoff angular frequency of the filter, which means that at this frequency, the gain of the filter drops to 0.707 times the maximum value; N is the order of the filter, which determines the steepness of the filter; the higher the order, the faster the frequency response of the filter decays after the cutoff frequency.

[0070] Step S122: Baseline drift is a common low-frequency interference in PPG signals, which may be caused by factors such as breathing, body position changes, or movement of sensor positions. Removing baseline drift helps to restore the true waveform of the PPG signal and improve the accuracy of physiological parameters.

[0071] The PPG signal with high and low frequency environmental noise removed is used to fit the trend of the signal using an autoregressive model, and then the fitted trend is subtracted from the signal to obtain the PPG signal after removing the baseline drift.

[0072] The autoregressive model can be expressed as an AR(p) model as follows:

[0073] X t =c+a 1 X t-1 +…+a p X p-1 +b;

[0074] Among them, p represents the lag order, that is, using the data of the past p time points to predict the current value; X t is the value of the time series at time t; c represents the mean (or intercept) of the series; a is the parameter of the model, called the autoregressive coefficient, which represents the influence of each lag term on the current value; b is the error term or random disturbance term, which is assumed to be an independent and identically distributed normal distribution.

[0075] Step S2: removing motion artifacts from the PPG signal using a method combining successive variational modal decomposition and adaptive filtering.

[0076] Motion artifacts are interference introduced by changes in the relative position between the sensor and the skin due to the user's movements (such as walking and shaking). This interference is usually multi-directional interference in three-dimensional space, and the frequency range is usually close to the frequency range of the pulse signal. The waveform is irregular and has great dynamic changes, which will cause large fluctuations and deviations in the PPG signal. Therefore, motion artifacts are more difficult to remove than high- and low-frequency noise and baseline drift.

[0077] To address this problem, the present invention designs a method based on the combination of successive variational mode decomposition and adaptive filtering to remove motion artifacts. Sequential variational mode decomposition (SVMD) is an improvement based on variational mode decomposition (VMD), which is a signal decomposition method used to analyze non-stationary and nonlinear signals. VMD provides a more stable and effective signal decomposition method, and successive VMD further improves its adaptability to complex signals on this basis.

[0078] The objective function of VMD is as follows:

[0079]

[0080] Where k is the number of modes to be decomposed, u k (t) is the kth mode, w k is the center frequency of the kth mode; f(t) is the original signal, and j represents the imaginary unit.

[0081] In order to solve the above variational problem, it is expressed as an optimization problem with constraints. Add the reconstruction condition, that is, the sum of all modes is equal to the original signal f(t), as shown below:

[0082]

[0083] Therefore, the optimization problem of VMD is expressed in Lagrangian form with constraints as follows:

[0084]

[0085] Among them, α is a weight factor used to balance the reconstruction error and bandwidth constraint, and γ is a Lagrange multiplier used to introduce the reconstruction constraint.

[0086] Successive Variational Mode Decomposition (SVMD) uses a step-by-step decomposition strategy to process signals based on VMD. The basic idea is to gradually separate the modal components of different frequency bands from complex signals by successive decomposition until the decomposition of the entire signal is completed.

[0087] The Pearson correlation coefficient between the separated harmonic signal and the triaxial acceleration is calculated as follows:

[0088]

[0089] Among them, pearson represents the Pearson correlation coefficient, X i Indicates harmonic data, Y i represents acceleration data, represents the mean of the harmonic data, Represents the mean of the acceleration data.

[0090] Each harmonic will have three Pearson correlation coefficients for the three-axis acceleration. The harmonics with Pearson correlation coefficients greater than the threshold are selected for reconstruction, that is, the signals are directly superimposed and the parts of the harmonics that are more correlated with the three-axis acceleration are retained, thereby obtaining most of the motion artifact signals in the original signal.

[0091] The original signal and the motion artifact signal are filtered through a normalized least mean square filter (NLMS) to obtain a PPG signal after removing motion artifacts; wherein the input is the PPG signal, the reference signal is the motion artifact, and the output is the PPG signal after removing motion artifacts.

[0092] Step S3: extract accurate peak values ​​and valley values ​​according to the denoised PPG signal.

[0093] In PPG signals, due to the interference of multiple peaks and valleys, the detection algorithm may mistakenly identify secondary peaks as primary peaks, or misjudge small valleys as primary valleys. Traditional peak and valley detection algorithms (such as the sliding window method) have difficulty distinguishing between primary peaks and valleys and secondary peaks and valleys, especially in some pathological conditions, where the waveform may become more complex or irregular.

[0094] To address this problem, the present invention sets up a strategy of jointly detecting peak values ​​and valley values ​​using multiple detection methods. Since different peak detection algorithms have different conditions and specific methods for detecting pulse wave peak values, for the denoised PPG signal, multiple detection algorithms can take more aspects into consideration, making it easier to detect some peak values ​​that are difficult to detect with a single algorithm. In addition, in order to avoid the risk of detecting false peak values ​​and false valley values, different detection thresholds are set according to the quality of the signal, making the detected peak values ​​and valley values ​​more reasonable and accurate.

[0095] Step S4: Calculate the R value based on the denoised PPG signal and the extracted peak and valley values, and obtain the blood oxygen saturation based on the R value.

[0096] The denoised PPG signal can be divided into two wavelengths, red light (660nm) and infrared light (940nm). These two wavelengths contain information about oxygenated hemoglobin (HbO 2 ) and deoxyhemoglobin (Hb) have different absorption characteristics. In a complete PPG cycle, the red and infrared light signals are decomposed into AC and DC components. The extraction of AC and DC components is an important prerequisite for calculating the R value. These components are closely related to the hemoglobin content and type in the blood.

[0097] Therefore, since oxygenated hemoglobin and deoxygenated hemoglobin have different absorption characteristics for red light and infrared light, the absorption difference is reflected by the ratio of the AC and DC components, and the R value is calculated using their difference, and the blood oxygen saturation (SpO) is estimated based on the R value. 2 ).

[0098] Step S41: extract AC and DC components from the PPG signal.

[0099] Step S411: In a complete heartbeat cycle, a DC component is obtained by taking an average value of the signal.

[0100] Step S412: Calculate the AC component by using the peak-to-valley difference of the PPG signal.

[0101] The AC component is the fluctuating component of the signal that occurs with the heartbeat, representing the dynamic light absorption caused by PPG. With each heartbeat, blood flow increases briefly, changing the amount of light absorbed, causing the signal to form peaks and valleys within the pulse cycle. Therefore, in a complete heartbeat cycle, the AC component is obtained by calculating the peak-to-valley difference of the PPG signal.

[0102] Step S42: Calculate the R value according to the obtained AC and DC components, as shown below:

[0103]

[0104] Among them, the numerator is the ratio of the AC component to the DC component of the red light wavelength; the denominator is the ratio of the AC component to the DC component of the infrared light wavelength; the R value reflects the relative rate of change of the red light and infrared light signals.

[0105] Due to the difference in absorption of red light and infrared light by oxygenated hemoglobin and deoxygenated hemoglobin, the ratio of oxygenated and deoxygenated hemoglobin in the blood is calculated by the R value, laying the foundation for the estimation of blood oxygen saturation.

[0106] Step S43: Based on the R value, the blood oxygen saturation is calculated using an empirical formula.

[0107] Since blood oxygen saturation (SpO 2 ) has an empirical relationship with the R value, so an empirical formula (such as the Maxim Equation) is used to convert the R value into blood oxygen saturation, as shown below:

[0108] SpO 2 =A*R 2 +B*R+C;

[0109] Among them, A, B, and C are constants, which are usually obtained based on a large amount of experimental data.

[0110] Example

[0111] In this embodiment, all data are first classified according to the number of the person who collected it, and the plateau blood oxygen saturation information of each person is cleaned and sorted. According to the amount of data and the sampling frequency of PPG of 64Hz, the sliding window of PPG is set to 120 seconds, so there are 7680 data points in one window, that is, N=7680, and the window overlap range is set to 25%. Figure 2 shown.

[0112] The unprocessed PPG is contaminated by different noises, so different methods need to be used to remove different noises to obtain a clean PPG signal, such as Figure 3 As shown in the figure, since the PPG signal is between 1 and 10 hours, a Butterworth bandpass filter with a low cutoff frequency of 1 Hz and a high cutoff frequency of 10 Hz is used for filtering, and an autoregressive model is used to predict the trend of the PPG signal. The predicted trend is then subtracted from the original signal to obtain a signal after removing the baseline drift.

[0113] The PPG signal with high and low frequency noise and baseline drift removed is decomposed into different harmonics using SVMD, and the coefficient of each harmonic is calculated with the three-axis acceleration signal through the Pearson correlation coefficient calculation formula, and then judged with the coefficient threshold. The coefficient threshold set in this embodiment is 0.75. If the coefficient of the harmonic with the three-axis acceleration signal exceeds this threshold, it is saved first until all harmonics are processed. All the saved harmonics are superimposed to obtain all harmonics containing motion artifacts. Then, the PPG signal with high and low frequency noise and baseline drift removed and the artifact harmonic signal are input into the normalized adaptive filter (NLMS), such as Figure 4 As shown in the figure, the PPG signal is the input signal and the artifact harmonic is the reference signal, which can be used to obtain the final clean PPG signal.

[0114] Next, this embodiment uses a combination of multiple peak detection methods to accurately extract peak values ​​and valley values ​​for each denoised PPG signal, and sets a certain threshold value to greatly improve the accuracy of detecting peak values ​​and valley values. The multiple peak detection methods used in this embodiment include:

[0115] (1) Local maximum method: Use the find_peaks function in Python to extract all local maximum points from the signal. Note that among the extracted points, remove the points whose values ​​are less than the average amplitude of the entire signal.

[0116] (2) Mean generation block of signal threshold method: The mean of the entire signal amplitude is generated into blocks, and the point containing the maximum value in each block is called the peak point.

[0117] (3) The first-order derivative method of the adaptive threshold method: First, the signal is divided into blocks every 5 seconds. Then, all signals are differentiated and points with a differential value of zero are extracted as candidate peaks. Finally, the threshold is adaptively set according to the amplitude average of the image block, and the candidate peak with an amplitude greater than the threshold becomes the peak point.

[0118] (4) Slope and function method of adaptive threshold method: First, only rising points are retained through the differentiation process, and the rest of the points are set to zero; then, all local maximum points are extracted from the improved signal; finally, the threshold is continuously updated to 70% of the median of the last five peaks, and only points with amplitudes greater than the threshold are extracted, which are relocated to the original signal to represent the peak points in the original signal.

[0119] (5) Bandpass filtering method: First, the signal is bandpass filtered and the filtered signal is truncated and squared; then, two sliding average filters are used to determine the region of interest and to find the heartbeat peaks in these regions; finally, valid peaks are screened out based on certain threshold conditions.

[0120] (6) Adjacent data point comparison method: First, the rising slope is found by comparing adjacent data points. When the number of consecutive rising points exceeds a certain threshold, it is considered that a possible peak position has been found. Then, based on certain conditions, it is confirmed whether these possible peaks are true heartbeat peaks and their positions are recorded.

[0121] The above six methods are used for peak detection, and the threshold is set to 4. Only when 4 of the methods are determined to be peaks, will it be considered a true peak, otherwise it is considered a false peak. When calculating the valley value, directly multiply the entire PPG signal by -1, then use the above method to detect the peak, and set the threshold to 3, and finally you can get the accurate peak and valley values, such as Figure 5 shown.

[0122] Finally, the R value is calculated based on the denoised PPG signal and the extracted peak and valley values, and the blood oxygen saturation is calculated based on the R value. The specific process is as follows:

[0123] 1. Calculate the DC component: First, select the sampling window. Select a window that contains multiple pulse cycles (usually at least 5-10 seconds of data to ensure data stability); then, average all data points in the window to obtain the DC component, as shown below:

[0124]

[0125] Where n is the number of data points in the window and i is the value of the PPG signal at the i-th sampling point.

[0126] 2. Calculate the AC component: First, find the maximum (peak) and minimum (valley) values ​​of the PPG signal within a cycle; then, determine the peak-to-valley difference, i.e., the AC component of each cycle, through the maximum and minimum values ​​within the window, as shown below:

[0127] AC = Peak-Valley;

[0128] Among them, Peak is the peak value of the signal within a heartbeat cycle; Valley is the valley value of the signal within a heartbeat cycle.

[0129] In order to improve the stability and accuracy of the signal, the AC and DC components can be calculated separately in multiple cycles, and then the AC and DC components of multiple cycles can be averaged to eliminate accidental errors.

[0130] 3. Calculate the R value and blood oxygen saturation: First, calculate the R value based on the AC and DC components; then, substitute the calculated R into the Maxim equation to obtain the estimated blood oxygen saturation.

[0131] Therefore, the present invention adopts the above-mentioned blood oxygen estimation method based on the combination of successive variational mode decomposition and adaptive filtering, establishes an accurate blood oxygen saturation model for the photoelectric volumetric pulse wave signal, and proposes a joint denoising model based on successive variational mode decomposition and adaptive filtering to remove motion artifacts; in order to solve the problem of inaccurate detection of pulse wave peak and valley values, a method of combining multiple detection methods is proposed, and then a more stable DC and AC components are obtained through multi-cycle signals, thereby determining the R value required for calculating the blood oxygen saturation, and directly bringing the R value into the Maxim equation to obtain the blood oxygen saturation, which greatly improves the accuracy of estimating the blood oxygen saturation.

[0132] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A blood oxygen estimation method based on successive variational mode decomposition and adaptive filtering, characterized in that: The following steps are involved: Step S1, removing high-frequency noise and baseline drift from the collected photoplethysmography signal PPG; Step S2, using successive variational modal decomposition and adaptive filtering to remove motion artifacts from the PPG signal; Step S3, extracting accurate peak values ​​and valley values ​​according to the denoised PPG signal; Step S4, calculating the R value based on the denoised PPG signal and the extracted peak and valley values, and obtaining the blood oxygen saturation based on the R value, wherein the R value refers to the absorbance ratio of red light and infrared light in the blood.

2. The blood oxygen estimation method based on successive variational mode decomposition and adaptive filtering according to claim 1, characterized in that: In step S1, the collected PPG signal is preprocessed to remove high and low frequency noise and baseline drift. The specific process is as follows: Step S11, setting a sliding window and window overlap, processing the collected PPG signal and acceleration signal in blocks, and analyzing the time variation characteristics of the signal; Step S12: For the segmented PPG data, use a Butterworth bandpass filter to filter to remove the noise of the high and low frequency environments, and use an autoregressive model to fit the trend of the signal, and then subtract the fitted trend from the signal to obtain a PPG signal after removing the baseline drift.

3. The blood oxygen estimation method based on successive variational mode decomposition and adaptive filtering according to claim 1, characterized in that: In step S2, successive variational mode decomposition is an improvement based on variational mode decomposition, wherein the objective function of variational mode decomposition is as follows: Where k is the number of modes to be decomposed, u k (t) is the kth mode, w k is the center frequency of the kth mode; f(t) is the original signal, and j represents the imaginary unit; The objective function of variational mode decomposition is expressed as an optimization problem with constraints and solved. The reconstruction condition is added, that is, the sum of all modes is equal to the original signal f(t), as shown below: The VMD optimization problem is then expressed in Lagrangian form with constraints as follows: Among them, α is a weight factor used to balance the reconstruction error and bandwidth constraint, and γ is a Lagrange multiplier used to introduce the reconstruction constraint.

4. The blood oxygen estimation method based on successive variational mode decomposition and adaptive filtering according to claim 3, characterized in that: The PPG signal is subjected to a method combining successive variational mode decomposition and adaptive filtering to remove motion artifacts. The specific process is as follows: First, based on variational mode decomposition, successive variational mode decomposition uses a step-by-step decomposition strategy to process signals. The basic idea is to gradually separate the modal components of different frequency bands from complex signals by successive decomposition until the decomposition of the entire signal is completed. Then, the Pearson correlation coefficient between the separated harmonic signal and the triaxial acceleration is calculated as follows: Among them, pearson represents the Pearson correlation coefficient, X i Indicates harmonic data, Y i represents acceleration data, represents the mean of the harmonic data, represents the mean value of acceleration data; Finally, the original signal and the motion artifact signal are passed through a normalized minimum mean square filter to obtain the PPG signal after removing the motion artifact; wherein the input is the PPG signal, the reference signal is the motion artifact, and the output is the PPG signal after removing the motion artifact.

5. The blood oxygen estimation method based on successive variational mode decomposition and adaptive filtering according to claim 4, characterized in that: Each decomposed harmonic will have three Pearson correlation coefficients for the three-axis acceleration. The harmonics with Pearson correlation coefficients greater than the threshold are selected for reconstruction, that is, the signals are directly superimposed and the parts of the harmonics that are more correlated with the three-axis acceleration are retained, thereby obtaining most of the motion artifact signals in the original signal.

6. The blood oxygen estimation method based on successive variational mode decomposition and adaptive filtering according to claim 1, characterized in that: In step S3, based on the denoised PPG signal, a multiple detection method combined strategy is used to perform peak and valley detection; Among them, the multiple peak detection method includes but is not limited to: local maximum method, first-order derivative method of adaptive threshold method, slope and function method of adaptive threshold method, and bandpass filtering method.

7. The blood oxygen estimation method based on successive variational mode decomposition and adaptive filtering according to claim 1, characterized in that: In step S4, the R value is calculated based on the denoised PPG signal and the extracted peak and valley values, and the blood oxygen saturation is obtained based on the R value, including the following steps: Step S41, extracting AC and DC components from the PPG signal; Step S42, calculating the R value according to the obtained AC and DC components; Step S43: Based on the R value, the blood oxygen saturation is calculated using an empirical formula.

8. The blood oxygen estimation method based on successive variational mode decomposition and adaptive filtering according to claim 7, characterized in that: In step S41, AC and DC components are extracted. The specific process is as follows: Step S411: In a complete heartbeat cycle, the DC component is obtained by taking the average value of the signal, as shown below: Where n is the number of data points in the window, and i is the value of the PPG signal at the i-th sampling point; Step S412: Calculate the AC component by the peak-to-valley difference of the PPG signal, as shown below: AC = Peak-Valley; Among them, Peak is the peak value of the signal within a heartbeat cycle; Valley is the valley value of the signal within a heartbeat cycle.

9. The blood oxygen estimation method based on successive variational mode decomposition and adaptive filtering according to claim 7, characterized in that: In step S42, the R value is calculated based on the obtained AC and DC components as follows: Among them, the numerator is the ratio of the AC component to the DC component of the red light wavelength; the denominator is the ratio of the AC component to the DC component of the infrared light wavelength; the R value reflects the relative rate of change of the red light and infrared light signals.

10. The blood oxygen estimation method based on successive variational mode decomposition and adaptive filtering according to claim 7, characterized in that: In step S43, based on the empirical relationship between blood oxygen saturation and R value, an empirical formula is used to convert the R value into blood oxygen saturation, as shown below: SpO2=A*R 2 +B*R+C; Among them, A, B, and C are constants, which are usually obtained based on a large amount of experimental data.

Images