Heart rate identification and abnormal heart rate detection method based on PPG and BP signals and application

Abstract

The invention provides a heart rate recognition and abnormal heart rate detection method based on PPG and BP signals. The heart rate recognition and abnormal heart rate detection method comprises the steps that 1, a blood pressure waveform, SV, an electrocardiogram and a photoplethysmography signal are obtained; 2, preprocessing the data acquired in the step 1; step 3, performing feature extraction on each preprocessed signal; and 4, performing comprehensive analysis based on the extracted features, and identifying whether the heart rate of the patient is normal or not. According to the method, the robustness and accuracy of heart rate recognition are enhanced through the multi-signal fusion technology, meanwhile, the reliable monitoring capacity of the heart rate is improved through machine learning and a signal processing algorithm, and compared with independent electrocardiogram detection, the method can obviously improve the patient monitoring accuracy and reduce the false alarm rate.

Description

Technical Field

[0001] This invention relates to the field of electrocardiography technology, and in particular to a method for heart rate recognition and abnormal heart rate detection based on PPG and BP signals. Background Technology

[0002] On an electrocardiogram (ECG) monitor, HR stands for heart rate, representing the number of times the heart beats per minute, and is one of the fundamental indicators of cardiac function. In hospitals, a patient's heart rate (HR) is a vital sign, obtainable through different types of signals, such as an electrocardiogram (ECG) or blood pressure waveform (BP). Currently, heart rate is typically acquired from a single signal (e.g., from an ECG lead). Artifacts such as patient movement or sensor disconnection can lead to false alarms due to inaccurate heart rate measurements. Furthermore, abnormal heart rate is a crucial indicator of cardiovascular disease, and accurate identification is essential for clinical diagnosis. Existing algorithms often rely on linear feature extraction, which may not adequately capture the complexity of heart rate changes. Therefore, how to accurately measure heart rate, avoid false alarms, and detect abnormal heart rates is a pressing problem that needs to be solved. Summary of the Invention

[0003] In view of the shortcomings of the prior art, the present invention proposes a heart rate recognition and abnormal heart rate detection method based on PPG and BP signals. By combining physiological information from multiple sensors, it can reduce the impact of single sensor failure and detect abnormal heart rates simultaneously.

[0004] The technical solution adopted in this invention includes the following steps:

[0005] Step 1: Obtain physiological signals from the sample patients;

[0006] Step 2: Preprocess the physiological signals acquired in Step 1 to obtain standardized synchronization signals for each physiological signal; among which, the standardized synchronization signals include blood pressure waveform standardized signal, stroke volume standardized signal, electrocardiogram standardized signal, and photoplethysmography standardized signal.

[0007] Step 3: Construct a heartbeat recognition and anomaly detection model based on the standardized synchronization signals of various physiological signals;

[0008] Step 4: Input the physiological signals of the target patient into the heartbeat recognition and abnormality detection model to identify the heartbeat and obtain the heartbeat abnormality classification.

[0009] Therefore, the present invention employs the above-mentioned method for heart rate recognition and abnormal heart rate detection based on PPG and BP signals, which has the following beneficial effects:

[0010] First, this invention achieves complementary redundancy by integrating four types of physiological signals: PPG, BP, ECG, and SV. When a single signal is distorted due to motion artifacts, poor sensor contact, or environmental noise, the remaining signals can provide cross-validation, significantly reducing the probability of false alarms. For example, PPG signals are easily affected by peripheral blood flow fluctuations, while BP signals are sensitive to changes in central arterial pressure. Joint analysis of the temporal characteristics of both can effectively distinguish between true heart rate fluctuations and noise interference, reducing the overall false alarm rate of the system by 35% compared to single ECG detection (based on clinical validation data).

[0011] Second, this invention employs nonlinear feature extraction methods such as wavelet transform, FFT frequency domain analysis, and Lyapunov exponent to comprehensively capture the time-domain, frequency-domain, and chaotic characteristics of signals. By combining the random forest algorithm with ensemble learning of multi-source features, it solves the problem of insufficient representation of complex heart rate variability patterns by traditional linear models. Experiments show that this method achieves a sensitivity of 92.3% and a specificity of 89.7% for identifying abnormal cardiac events such as atrial fibrillation and sinus tachycardia (compared to 76.5% and 82.1% for the control group using a single ECG feature, respectively).

[0012] Third, this invention effectively suppresses baseline drift and transient noise through a time signal correction module (such as ECG-PPG-based delay compensation) and adaptive threshold processing technology. In simulated motion interference tests, the system maintains a heart rate detection error of ≤2 BPM even in a 5-10 Hz high-frequency noise environment, which is superior to the 5-8 BPM error of traditional filtering schemes.

[0013] Fourth, this invention supports multi-parameter joint analysis (such as the correlation between pulse pressure and PPG waveform width), which can simultaneously assess cardiovascular status and provide multi-dimensional indicators for early warning. Furthermore, the feature fusion framework is compatible with new sensor data access and can be extended to multimodal monitoring scenarios such as blood oxygen saturation (SpO2) or impedance cardiography (ICG) in the future.

[0014] Fifth, through wavelet coefficient reconstruction and parallel computing architecture design, the system processing latency is controlled within 50ms, meeting the real-time monitoring needs of intensive care, while the hardware resource utilization rate is reduced by 40% compared with the traditional multi-channel independent analysis scheme.

[0015] In summary, this invention achieves a highly robust and accurate heart rate monitoring system through multi-source signal fusion, nonlinear feature modeling, and machine learning optimization. It demonstrates significant performance advantages in complex clinical scenarios and provides reliable technical support for the accurate diagnosis of cardiovascular diseases.

[0016] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0017] Figure 1 This is a flowchart of the method proposed in this invention. Detailed Implementation

[0018] In the description of this invention, it should also be noted that, unless otherwise expressly specified and limited, these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various alterations or modifications to the invention, and these equivalent forms also fall within the scope defined by the appended claims.

[0019] This invention proposes a heart rate recognition and abnormal heart rate detection method based on PPG and BP signals. Its core idea is to independently detect heartbeats in different signals, correct the time delay between signals, and combine the detection of heartbeats in different signals through a voting mechanism.

[0020] like Figure 1 As shown, the method proposed in this invention includes the following steps:

[0021] Step 1: Acquire physiological signals from the sample patients; these physiological signals include blood pressure waveform, stroke volume, electrocardiogram, and photoplethysmography (PPG) signals.

[0022] Step 2: Preprocess the physiological signals collected in Step 1;

[0023] Step 3: Based on the preprocessed physiological signals, construct a heartbeat abnormality prediction model;

[0024] Step 4: Input the physiological signals of the target patient into the heart rhythm abnormality prediction model to obtain the heart rhythm abnormality classification category output by the heart rhythm abnormality prediction model.

[0025] The following sections will describe each of the above steps in detail:

[0026] 1. Signal Acquisition

[0027] First, various physiological signals of the patient are collected using a vital signs monitor and a three-lead sensor, including blood pressure waveform (BP), stroke volume (SV), electrocardiogram (ECG), and photoplethysmography (PPG) signals.

[0028] 2. Signal preprocessing

[0029] Step 2.1: Denoise the acquired physiological signals by wavelet transform to obtain the denoised reconstructed signals;

[0030] Wavelet transform is performed on each acquired signal to obtain wavelet coefficients w. j,k :

[0031]

[0032] Where x(t) represents the acquired raw physiological signals (ECG, PPG, BP, SV signals); ψ j,k (t) represents the mother wavelet basis function; j represents the scaling parameter, used to control the frequency resolution of the wavelet. When j increases, the wavelet is compressed, allowing analysis of high-frequency components (noise, motion artifacts); when j decreases, the wavelet is stretched, allowing analysis of low-frequency components (P-waves and T-waves in ECG); k represents the translation parameter, i.e., the position of the wavelet on the time axis, used for local time analysis; 2 j / 2 This represents the normalization factor, ensuring consistent wavelet energy across different scales; w j,k These are wavelet coefficients used to represent the energy distribution at scale j and location k.

[0033] The wavelet coefficients are thresholded using a soft thresholding method according to the following formula:

[0034]

[0035] in, These are the wavelet coefficients after soft thresholding, where λ is the threshold parameter, typically... σ is the noise standard deviation, and N is the signal length; wavelet coefficients can be smoothed and denoised by using soft thresholding.

[0036] Wavelet coefficients after soft thresholding The physiological signals (PPG, BP, SV, ECG) were reconstructed separately to obtain the denoised reconstructed signals:

[0037]

[0038] in, The denoised reconstructed signal (including ECG, PPG, BP, and SV reconstructed signals) is used for subsequent feature extraction.

[0039] Step 2.2: Perform signal standardization and delay correction on the denoised reconstructed signal to obtain a standardized synchronization signal;

[0040] Optionally, step 2.2 includes:

[0041] Step 2.2.1: Normalize the denoised reconstructed signal;

[0042] Since the units of various physiological signals are different, such as ECG signal in mV, BP signal in mmHg, and PPG in light intensity change, it is necessary to unify the dimensions of different signals (normalize to the 0-1 range).

[0043] Step 2.2.2: Perform delay correction on the normalized ECG reconstructed signal and PPG reconstructed signal;

[0044] To ensure that features of different signals (such as the pulse amplitude of PPG and the systolic blood pressure of BP) have equal weight in the machine learning model, thereby enhancing the comparability of features, delay correction is applied to ECG and PPG signals.

[0045] τ delay =argτmax(corr(ECG(t),PPG(t+τ)))

[0046] Wherein, ECG(t) is the amplitude of the electrocardiogram signal at time t, used to record cardiac electrical activity, with the R peak corresponding to ventricular depolarization (heartbeat trigger point); PPG(t) is the amplitude of the photoplethysmogram signal at time t, used to reflect peripheral blood flow pulsation, with the peak corresponding to arterial vasodilation (time delay of heartbeat conduction to the periphery); τ is the time offset, representing the delay of the PPG signal relative to the ECG signal, with a value range of τ∈[-T,T], where T represents the maximum possible signal delay; corr(ECG(t),PPG(t+τ)) is the cross-correlation function value of the ECG and PPG signals at time offset τ, used to quantify the similarity between the two signals;

[0047] Calculation formula:

[0048] corr(ECG(t),PPG(t+τ))=∑tECG(t)·PPG(t+τ)corr(ECG,PPG)(τ)=t∑ECG(t)·PPG(t+τ).

[0049] 3. Feature Extraction

[0050] The following sections describe the extraction and optimization of PPG features, blood pressure features, SV features, and ECG signals.

[0051] Step 3.1: Extract features from the preprocessed PPG signals to obtain the PPG signal set features;

[0052] Specifically, when extracting features from PPG signals, the Peak detection algorithm can be used to extract the peak position and pulse interval T of the PPG signal, where the pulse interval T is the time difference between adjacent PPG peaks, i.e. the interval between two heartbeats.

[0053] Based on the pulse interval T, the heart rate HR obtained from the PPG signal is calculated using the following formula. PPG :

[0054]

[0055] Where T represents the pulse interval;

[0056] The waveform features of the preprocessed PPG signal are extracted, and the waveform width PW is obtained by the following formula:

[0057] PW=T d -T a

[0058] Among them, T d T represents the end time point of the PPG pulse waveform. a Indicates the start time point of the PPG pulse waveform;

[0059] The pulse amplitude PA is obtained using the following formula:

[0060] PA = max(PPG) - min(PPG)

[0061] Where max(PPG) represents the maximum value of the PPG signal within one pulse cycle (corresponding to the peak value of arterial vasodilation), and min(PPG) represents the minimum value of the PPG signal within the same pulse cycle (corresponding to the baseline value of vasoconstriction).

[0062] Based on the obtained heart rate HR PPG Waveform width (PW) and pulse amplitude (PA) construct the characteristics of the PPG signal set:

[0063] F PPG = (PA, HR, PW)

[0064] Among them, F PPG This is the feature set of the PPG signal.

[0065] Step 3.2: Extract features from the preprocessed BP signal to obtain the BP signal set features;

[0066] Step 3.2.1: Calculate systolic and diastolic blood pressure based on the preprocessed BP signal:

[0067] BP systolic =max(BP signal)

[0068] BP diastolic =min(BP signal)

[0069] Among them, BP systolic BP represents systolic blood pressure. diastolic Indicates diastolic blood pressure;

[0070] Step 3.2.2: Calculate the pulse pressure difference based on systolic and diastolic blood pressure:

[0071] BP stroke =BPsystolic -BP diastolic

[0072] Among them, BP stroke Indicates pulse pressure difference;

[0073] Step 3.2.3: Calculate the RMS root mean square of the BP signal using the following formula:

[0074]

[0075] Where N represents the total number of BP signal samples collected, and BP[i] represents the i-th sample;

[0076] Step 3.2.4: Based on the obtained pulse pressure difference BP stroke systolic blood pressure (BP) systolic diastolic blood pressure (BP) diastolic and the root mean square RMS of the BP signal BP The characteristics that constitute the BP signal set are:

[0077] F BP =(RMS, BP) stroke BP diastolic BP systolic )

[0078] Among them, F BP The feature set of the BP signal;

[0079] Step 3.3: Use the value of each stroke output as a feature of the SV signal set;

[0080] The stroke volume (SV) value F is read directly from the three-lead sensor. SV and F SV As a characteristic of the SV signal set.

[0081] Step 3.4: Extract features from the preprocessed ECG signal to obtain the ECG signal set features;

[0082] Step 3.4.1: Denoise the preprocessed ECG signal using the following formula to obtain the denoised EGG signal:

[0083] x clean (t)=x(t)-x noise (t)

[0084] Where, x clean x(t) is the denoised ECG signal, i.e., the denoised EGG signal, and x(t) is the preprocessed ECG signal. noise (t) represents the noise signal, which is 0.03mV;

[0085] Step 3.4.2: Perform baseline correction on the EGG denoised signal using the following formula to obtain the EGG corrected signal:

[0086] x corrected (t)=x clean (t)-b(t)

[0087] Where, x corrected b(t) is the baseline-corrected signal, i.e., the EGG-corrected signal, and b(t) is the baseline drift, which is 0.032mV.

[0088] Step 3.4.3: Calculate the mean μ of the EGG correction signal using the following formula:

[0089]

[0090] Where N represents the total number of sample points of the acquired ECG signal;

[0091] Step 3.4.4: Calculate the RMS value of the EGG signal based on the mean μ and the EGG correction signal:

[0092]

[0093] Step 3.4.5: Calculate the skewness of the ECG signal based on the mean μ and the EGG correction signal.

[0094]

[0095] Where σ represents the standard deviation of the corrected ECG signal, i is an index variable representing the i-th data point, and i = 1, 2, ..., N, where N is the total number of sample points of the corrected ECG signal.

[0096] Step 3.4.6: Calculate the kurtosis of the ECG signal based on the mean μ and the EGG correction signal.

[0097]

[0098] Step 3.4.7: Based on the denoised EGG signal, obtain the denoised ECG time-domain signal through FFT (Fast Fourier Transform):

[0099]

[0100] Where x(f) is the frequency domain representation of the signal, x clean，n Let f be the value of the nth signal sample after denoising, N be the total number of signal sample points, f be the frequency component, and j be the imaginary unit. The complex exponential term is used to convert a time-domain signal to the frequency domain, representing the amplitude and phase of the signal at different frequency components.

[0101] Step 3.4.8: Process the denoised ECG time-domain signal using CWT (Continuous Wavelet Transform) to obtain the wavelet transform coefficients:

[0102]

[0103] Among them, w a，b x represents the wavelet transform coefficients. clean (t) is the denoised ECG time-domain signal, ψ is the mother wavelet function, a is the wavelet scale and a = 2, b is the shift and b = 0.025s; This is a normalization factor used to ensure consistent wavelet energy across different scales;

[0104] Step 3.4.9: Calculate the Lyapunov exponential feature based on the denoised ECG time-domain signal:

[0105]

[0106] Where, x clean (t) represents the amplitude of the denoised ECG time-domain signal at time t, x clean (0) represents the amplitude of the denoised ECG time-domain signal at the initial moment;

[0107] Step 3.4.10: Obtain the maximum heart rate (HRmax) and minimum heart rate (HRmin) during the monitoring period, and construct the ECG signal set features based on the features calculated in steps 3.4.3-3.4.9:

[0108] F ECG =(μ, RMS, Skewness, Kurtosis, FFT, Lyapunov, CWT, HRmax, HRmin)

[0109] Here, FFT refers to the frequency domain representation of the ECG signal, CWT refers to the obtained wavelet transform coefficients, and Lyapunov refers to the Lyapunov exponential feature.

[0110] 4. Heart rate recognition and abnormality detection

[0111] Step 4.1: Based on set features F SV F PPG F BP The overall set characteristics of the j-th patient are obtained:

[0112] F j =[F SV F PPG F BP ]

[0113] F′j =F j A

[0114] Among them, F j F′ is a total set of features formed by splicing together features from multiple sources of physiological signals, used to reflect the patient's original physiological state; j The optimized feature representation is ultimately used to improve the accuracy and robustness of heart rate abnormality detection; A represents the patient's activity state, which is a mathematical tool for feature engineering used to extract more effective classification information.

[0115]

[0116] Step 4.2: Based on the total set features of all patients and the heart rate status label sequence of all patients, make predictions to obtain the final prediction results.

[0117] Specifically, firstly, the total set of features of all patients is combined into a transform feature set X. initial If F′ n Given that X contains m features and there are n patients in total, then X initial It can be represented as an n×m matrix; secondly, obtain the heart rate status label sequence Y for each patient within the current monitoring period. initial ;

[0118] X initial = [F′1, F′2, ..., F′] n ]

[0119] Y initial = [0, 1, ..., 1]

[0120] Where 0 represents a normal heart rate and 1 represents an abnormal heart rate;

[0121] Finally, first X initial Y initial The data is input into a random forest model to obtain preliminary prediction results y. pred，initial Then y pred，initial and ECG signal set characteristics F ECG The intermediate feature matrix X, formed by concatenating (including mean, RMS, FFT, CWT, Lyapunov exponent, etc.), final Then X final Y initial The data is input into a random forest model to obtain the final identified heart rate and output the classification category. The classification category indicates whether the heartbeat is normal or abnormal, and can be represented by 0 and 1, such as 1 for abnormal and 0 for normal.

[0122] y pred，initial =RF[X initial Y initial ]

[0123] X final =[y pred，initial F ECG ]

[0124] y pred，final =RF[X final Y initial ]

[0125] Where RF represents the Random Forest model, y pred，initial Y represents the initial prediction result. pred，final F represents the final prediction result. ECG This indicates the characteristics of the ECG signal set.

[0126] According to the final prediction result y pred，final Determine if the patient's heart rate is abnormal;

[0127]

[0128] Based on the predicted results of the sample patients and the original label Y initial Adjust the parameters of the random forest model until the preset number of model iterations or the accuracy of the classification category reaches the accuracy threshold. Use the random forest model after the last adjustment of the model parameters as the heartbeat recognition and anomaly detection model.

[0129] Clinical Cases

[0130] To verify the performance of the method proposed in this invention, heart rate recognition and abnormality detection based on multi-physiological signal fusion were performed on 53 patients, and the following results were obtained. These patients were all 53 consecutive post-cardiac surgery patients enrolled in the hospital's cardiac surgery department, excluding those with severe heart failure (EF < 30%) and those with incomplete data.

[0131] First, the baseline information (gender, age, underlying diseases, monitoring scenario) of each patient is shown in Table 1.

[0132] Table 1: Patient Demographics and Clinical Characteristics (N=53)

[0133]

[0134] Secondly, the results of the heart rate recognition and abnormality detection method proposed in this invention for the above 53 patients, and the overall performance comparison with the results obtained by traditional ECG monitoring, are shown in Table 2. As can be seen from Table 2, the sensitivity and specificity of this invention are higher than those of traditional ECG monitoring, and the false alarm rate is significantly reduced. Therefore, the method proposed in this invention can achieve highly robust and accurate heart rate monitoring.

[0135] Table 2: Overall Performance Comparison (N=53)

[0136]

[0137]

[0138] The results of the classification and comparison according to the type of heart arrhythmia are shown in Table 3. It can be seen that the specificity of the present invention has been improved to 91.7%.

[0139] Table 3: Subgroup Analysis (by Arrhythmia Type)

[0140]

[0141] Finally, this invention also analyzes a case of atrial fibrillation false positive. The patient was a 71-year-old woman with hypertension and diabetes. ECG, PPG, BP, and SV signals were acquired from this woman in a resting state (A=1.0). Using the method proposed in this invention, the false positive of atrial fibrillation caused by baseline drift in ECG was corrected through synchronous analysis of PPG-BP signals. The results are shown in Table 4. It can be seen that this invention demonstrates significant performance advantages in complex clinical scenarios, providing reliable technical support for the accurate diagnosis of cardiovascular diseases.

[0142] Table 4: Comparison of Signal Features

[0143]

[0144] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for heart rate recognition and abnormal heart rate detection based on PPG and BP signals, characterized in that, Includes the following steps: Step 1: Obtain multiple physiological signals from the sample patients; Step 2: Preprocess the physiological signals of each sample to obtain the standardized synchronization signals of the physiological signals of each sample; Step 3: Establish a heartbeat recognition and anomaly detection model based on the standardized synchronization signal of the sample's physiological signals; Step 4: Input the target patient's physiological signals into the heartbeat recognition and abnormality detection model to identify the target patient's heartbeat and obtain the heartbeat abnormality classification category.

2. The method for heart rate recognition and abnormal heart rate detection based on PPG and BP signals as described in claim 1, characterized in that, Step 2 includes the following specific steps: Step 21: Denoise and reconstruct the physiological signals of each sample obtained by wavelet transform to obtain the denoised ECG reconstructed signal, PPG reconstructed signal, BP reconstructed signal and SV reconstructed signal; Step 22: Normalize the ECG reconstructed signal, PPG reconstructed signal, BP reconstructed signal, and SV reconstructed signal; Step 23: Perform delay correction on the normalized ECG reconstructed signal and PPG reconstructed signal, and output the BP normalized signal, SV normalized signal, ECG normalized signal and PPG normalized signal.

3. The heart rate recognition and abnormal heart rate detection method based on PPG and BP signals as described in claim 1, characterized in that, Step 3 includes the following specific steps: Step 31: Extract features from the PPG normalized signal to obtain the PPG signal set features; Step 32: Extract features from the BP normalized signal to obtain the features of the BP signal set; Step 33: Use the value of the stroke volume in the SV normalized signal as the feature of the SV signal set; Step 34: Extract features from the ECG normalized signal to obtain the ECG signal set features; Step 35: Based on the PPG signal set features, BP signal set features, and SV signal set features, determine the total set features of the j-th patient; Step 36: Input the total set features of all sample patients, the set features of ECG signals, and the heart rate status label sequences of all patients into the random forest model for model training, and use the trained random forest model as the heartbeat recognition and abnormality detection model.

4. The heart rate recognition and abnormal heart rate detection method based on PPG and BP signals as described in claim 3, characterized in that, Step 31 specifically includes: Step 311: Determine the peak position and pulse interval of the PPG normalized signal; Step 312: Determine heart rate based on pulse interval; Step 313: Extract the waveform features of the PPG normalized signal to obtain the waveform width; Step 314: Determine the pulse amplitude based on the difference between the maximum and minimum peak values ​​of the PPG normalized signal within the same pulse interval period; Step 315: The heart rate, waveform width, and pulse amplitude are spliced ​​together to obtain the PPG signal set characteristics.

5. The method for heart rate recognition and abnormal heart rate detection based on PPG and BP signals as described in claim 3, characterized in that, Step 32 specifically includes: Step 321: Determine systolic and diastolic blood pressure based on the BP normalized signal; Step 322: Determine the pulse pressure based on the difference between systolic and diastolic blood pressure; Step 323: Calculate the root mean square RMS of the BP normalized signal; Step 324: Concatenate the RMS root mean square, pulse pressure, diastolic pressure, and systolic pressure to obtain the BP signal set characteristics.

6. The method for heart rate recognition and abnormal heart rate detection based on PPG and BP signals as described in claim 3, characterized in that, Step 34 specifically includes: Step 341: Calculate the mean, root mean square RMS, skewness, and kurtosis of the EGG normalized signal, and obtain the ECG time domain signal; Step 342: Process the denoised ECG time-domain signal using continuous wavelet transform to obtain new wavelet transform coefficients; Step 343: Determine the Lyapunov exponential characteristics of the denoised ECG time-domain signal; Step 344: Determine the maximum and minimum heart rates during the monitoring period; Step 345: Concatenate the mean, RMS root mean square of the EGG signal, skewness, kurtosis, denoised ECG time domain signal, Lyapunov exponential features, new wavelet transform coefficients, maximum heart rate and minimum heart rate to obtain the ECG signal set features.

7. The heart rate recognition and abnormal heart rate detection method based on PPG and BP signals as described in claim 6, characterized in that, Step 341 includes: Step 3411: Denoise the EGG normalized signal to obtain the EGG denoised signal; Step 3412: Perform baseline correction on the EGG denoised signal to obtain the EGG corrected signal; Step 3413: Calculate the average value of all EGG correction signals to obtain the mean value of the EGG correction signals; Step 3414: Based on the mean and the EGG correction signal, obtain the RMS root mean square, skewness, and kurtosis of the EGG signal; Step 3415: Based on the denoised EGG signal, obtain the denoised ECG time domain signal through FFT.

8. The method for heart rate recognition and abnormal heart rate detection based on PPG and BP signals as described in claim 6, characterized in that, Step 36 includes the following specific steps: Step 361: Optimize the overall set features of each sample patient based on their activity status to obtain the optimized overall set features: Step 362: Based on the optimized aggregate features of all sample patients, obtain the transformed feature set X. initial And obtain the heart rate status label sequence Y for each sample patient in the current monitoring period. initial ; Step 363: Transform the feature set X initial and Y initial The data is input into a random forest model to obtain preliminary prediction results y. pred，initial ; Step 364: The preliminary prediction result Y pred，initial The features are concatenated with the ECG signal set features to obtain a new feature matrix. The new feature matrix is ​​then re-inputted into the random forest model to output the final prediction result and classification category. Step 365: Based on the final prediction results, classification categories, and heart rate state label sequences, adjust the model parameters to obtain the trained random forest model.

9. A heart rate recognition and abnormal heart rate detection device based on PPG and BP signals, characterized in that, The device includes: The signal acquisition module is used to acquire multiple physiological signals from the sample patient. The preprocessing module is used to preprocess the physiological signals of each sample to obtain standardized synchronization signals of the physiological signals of each sample. The model building module is used to build a heartbeat recognition and anomaly detection model based on standardized synchronization signals of sample physiological signals. The detection module is used to input the physiological signals of the target patient into the heartbeat recognition and abnormality detection model, identify the heartbeat of the target patient, and obtain the heartbeat abnormality classification category.

10. A readable storage medium storing a program or instructions that, when executed by a processor, implement a heart rate recognition and abnormal heart rate detection method based on PPG and BP signals as described in any one of claims 1-8.

Images