A method and system for processing exercise electrocardiograms

Abstract

The application discloses a motion electrocardio processing method and system, and relates to the technical field of electrocardio signal processing. The motion electrocardio processing method provided by the application comprises the following steps: acquiring electrocardio signals under different motion states; performing empirical mode decomposition on the motion electrocardio to decompose the electrocardio signals into a plurality of intrinsic mode functions, and performing high-pass filtering operation on the low-frequency intrinsic mode functions to suppress low-frequency noise in the motion electrocardio; then, performing variational mode decomposition on the remaining components to more accurately extract variational mode functions of different frequencies, and performing center frequency judgment on each variational mode function; for the mode functions smaller than the center frequency, non-local mean noise reduction is adopted; for the mode functions greater than or equal to the center frequency, singular value decomposition is adopted to reduce noise and realize baseline interference suppression; finally, the mode functions after noise reduction are reconstructed, the wavelet threshold method is adopted to reduce noise for the electromyographic interference noise in the motion electrocardio, so that a clean motion electrocardio signal waveform after noise interference suppression is obtained, and subsequent further analysis is facilitated.

Description

Technical Field

[0001] This invention relates to the field of exercise electrocardiogram (ECG) processing technology, and in particular to an exercise ECG processing method and system. Background Technology

[0002] Cardiovascular disease is a major global concern due to its high incidence, disability rate, and mortality rate. Currently, cardiovascular disease accounts for 27% of deaths worldwide, gradually becoming a leading cause of death and seriously threatening sustainable development. Electrocardiography (ECG) is a commonly used method for diagnosing cardiovascular disease. Compared to static in-hospital testing, during exercise, due to increased cardiac load, symptoms such as coronary heart disease and myocardial ischemia are more easily revealed in the ECG signal. Therefore, ECG signals during exercise can more accurately reflect the health status of the heart.

[0003] However, patient activity and complex acquisition environments generate strong interference noise. This noise, coupled into the electrocardiogram (ECG) signal, obscures the detailed features of the ECG, increases the difficulty of signal extraction, reduces signal quality, and may even affect the doctor's diagnosis, making it difficult to detect potential diseases during exercise. Therefore, researching and developing an effective exercise ECG processing method to suppress noise interference is particularly important, as this will help improve signal quality and reduce the risk of misdiagnosis. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for processing electrocardiograms during exercise, which can suppress noise interference and obtain accurate and stable electrocardiogram signal waveforms.

[0005] To achieve the above objectives, the present invention provides the following solution:

[0006] A method for processing exercise electrocardiograms, comprising:

[0007] Acquire electrocardiogram (ECG) signals during exercise;

[0008] Empirical mode decomposition was performed on the exercise electrocardiogram to obtain several intrinsic mode function components;

[0009] The frequency magnitudes of several intrinsic mode function components are determined, and a high-pass filtering operation is performed based on the low-frequency components of the intrinsic mode functions to suppress low-frequency noise in exercise electrocardiograms;

[0010] After determining the remaining components obtained after suppressing low-frequency noise, variational mode decomposition is performed on the remaining components;

[0011] Extract variational mode functions at different frequencies to determine the magnitude of the center frequency;

[0012] Determine the relationship between the center frequency of each variational mode function and the preset center frequency;

[0013] For mode functions with frequencies less than the center frequency, nonlocal mean denoising is used; for mode functions with frequencies greater than or equal to the center frequency, singular value decomposition denoising is used to achieve baseline interference suppression.

[0014] The denoised mode function is reconstructed, and wavelet thresholding is used to suppress electromyographic interference noise in the exercise ECG. Finally, a clean exercise ECG signal waveform with noise interference suppressed is obtained.

[0015] Preferably, the step of determining the frequency magnitudes of several intrinsic mode function components and performing a high-pass filtering operation based on the low-frequency components of the intrinsic mode functions to suppress low-frequency noise in exercise electrocardiograms specifically includes:

[0016] The instantaneous frequency and envelope of each intrinsic mode function component are determined, and the intrinsic mode functions are classified to determine their frequency range and energy distribution;

[0017] Obtain the number of intrinsic mode functions and the length of the electrocardiogram signal;

[0018] Adaptive high-pass filtering is performed by dynamically adjusting the filter parameters based on the frequency characteristics of the low-frequency components of the intrinsic mode function.

[0019] The intrinsic mode functions are reconstructed by weighted summation of all filtered intrinsic mode functions.

[0020] Preferably, determining the remaining components obtained after suppressing low-frequency noise and performing variational mode decomposition on the remaining components specifically includes:

[0021] Perform detrending preprocessing on the remaining components;

[0022] Initial parameters are set for the variational mode decomposition, including the number of modes (K) and the penalty parameter (α);

[0023] The number of modes and the value of the penalty parameter are updated iteratively using the Lagrange multiplier method and the augmented Lagrange function;

[0024] In each iteration, the bandwidth of each modal function is minimized by adjusting the spectrum of the modal components to make their bandwidth as narrow as possible;

[0025] When the change in modal components is less than a preset threshold or the preset maximum number of iterations is reached, the convergence condition is met, and appropriate values ​​for the number of modes and the penalty parameter are obtained, thus achieving optimal variational mode decomposition.

[0026] Preferably, extracting the variational mode functions at different frequencies and determining the magnitude of the center frequency specifically includes:

[0027] Perform a discrete Fourier transform on each extracted variational mode function to convert the mode function in the time domain to the frequency domain;

[0028] The amplitude squared operation is performed on the modal function in the frequency domain to obtain the energy distribution of the signal at each frequency;

[0029] Normalized power spectral density, by finding the maximum peak value in the power spectral density curve, preliminarily determines the dominant frequency component of the mode function;

[0030] The frequency bandwidth range is determined based on the initially determined peak frequency.

[0031] Further calculation of the spectral centroid is performed, and the power spectral density-weighted frequency average is used as an accurate estimate of the center frequency.

[0032] Verify the rationality of the center frequency and check whether the center frequency is close to the previously determined peak value of the main frequency. If the center frequency deviates significantly from the main frequency, the spectrum bandwidth needs to be reassessed to ensure the accuracy of the spectrum analysis.

[0033] Preferably, determining the magnitude of the center frequency of each variational mode function relative to a preset center frequency specifically includes:

[0034] Determine the center frequency of each variational mode function. If the center frequency is less than the preset center frequency, it is classified as a low-frequency mode. If the center frequency is greater than or equal to the preset center frequency, it is classified as a high-frequency mode.

[0035] Preferably, for mode functions with frequencies less than a preset center frequency, nonlocal mean denoising is used; for mode functions with frequencies greater than or equal to the preset center frequency, singular value decomposition denoising is used to achieve baseline interference suppression. Specifically, this includes:

[0036] For low-frequency mode functions, a nonlocal mean denoising algorithm is used. First, for each low-frequency mode function, all local blocks in the signal are traversed, and blocks similar to the current block are searched.

[0037] The weights are calculated based on the similarity between the searched similar blocks and the current block, and the weight coefficients are represented by a Gaussian function.

[0038] The value of the current block is updated by using a weighted average of similar blocks based on the aforementioned weights.

[0039] By combining the blocks that have undergone nonlocal mean processing, the noise-reduced low-frequency mode function can be reconstructed.

[0040] For high-frequency mode functions, singular value decomposition is used for noise reduction, transforming each high-frequency mode function into a Hankel matrix;

[0041] Perform singular value decomposition on the constructed matrix to decompose the matrix into singular value matrices and corresponding left and right singular vector matrices;

[0042] Sort the decomposed singular values ​​and then calculate the difference sequence between adjacent singular values;

[0043] The cutoff point is determined based on the difference sequence, and the singular value part of the corresponding electrocardiogram signal is retained;

[0044] By using the retained main singular values ​​and their corresponding singular vector matrices, the signal is reconstructed, thus obtaining the denoised high-frequency mode function.

[0045] Preferably, the reconstructed and denoised mode function is used to suppress electromyographic interference noise in the exercise ECG using a wavelet thresholding method, finally obtaining a clean exercise ECG signal waveform after noise interference suppression, specifically including:

[0046] The low-frequency and high-frequency mode functions after nonlocal mean and singular value decomposition noise reduction are superimposed to reconstruct the preliminarily denoised ECG signal.

[0047] Wavelet transform is performed on the reconstructed electrocardiogram signal to decompose the signal into wavelet coefficients of multiple sub-bands with different frequencies;

[0048] An adaptive threshold calculation is applied to the wavelet coefficients obtained from the decomposition.

[0049] Based on the noise estimation results of wavelet coefficients at each scale, the thresholds at different scales are adaptively calculated.

[0050] The wavelet threshold function is applied to process the wavelet coefficients at each scale to obtain the motion ECG signal waveform after noise interference suppression.

[0051] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0052] The exercise ECG processing method provided by this invention first acquires ECG signals under different exercise states. Then, empirical mode decomposition (EMD) is performed on the exercise ECG signals, decomposing them into several intrinsic mode functions (IMFs). High-pass filtering is applied to the low-frequency IMFs to suppress low-frequency noise in the exercise ECG. Next, variational mode decomposition (VMD) is performed on the remaining components to more accurately extract variational mode functions of different frequencies. The center frequency of each variational mode function is determined. Furthermore, nonlocal mean denoising is used for mode functions with frequencies lower than the center frequency, while singular value decomposition (SVD) is used for denoising to suppress baseline interference for mode functions with frequencies greater than or equal to the center frequency. Finally, the denoised mode functions are reconstructed. Wavelet thresholding is used to reduce electromyographic interference noise in the exercise ECG, resulting in a clean exercise ECG waveform with noise suppression, facilitating further analysis.

[0053] The present invention also provides a sports electrocardiogram processing system, the system comprising:

[0054] The exercise ECG acquisition module is used to acquire ECG signals under different exercise conditions;

[0055] Storage module, used to store computer control programs;

[0056] The processing module is connected to the exercise ECG acquisition module and the storage module respectively, and is used to retrieve and execute the computer control program based on the exercise ECG signal to implement the exercise ECG processing method provided above, so as to obtain a clean exercise ECG signal waveform after noise suppression.

[0057] Preferably, the processing device includes:

[0058] The exercise ECG acquisition module is used to acquire ECG signal waveform data under different exercise states;

[0059] The Empirical Mode Decomposition (EMD) module is used to obtain intrinsic mode function components and perform high-pass filtering to suppress low-frequency noise in exercise ECG.

[0060] The variational mode decomposition module is used to perform variational mode decomposition on the remaining components obtained after suppressing low-frequency noise, and to set an appropriate number of modes and penalty parameters;

[0061] The center frequency determination module is used to determine the magnitude of the center frequency of variational mode functions at different frequencies and to verify the degree of deviation between the center frequency and the dominant frequency.

[0062] The nonlocal mean denoising module is used to suppress noise in low-frequency mode functions while preserving the main characteristics of the signal;

[0063] The singular value decomposition module is used to suppress noise in high-frequency mode functions. It uses the difference decomposition method to retain the main singular value components.

[0064] The wavelet threshold denoising module is used to suppress electromyographic interference noise in exercise electrocardiograms. It adaptively calculates thresholds at different scales and performs threshold denoising processing on different frequency subbands.

[0065] Preferably, the storage device is a computer-readable storage medium.

[0066] Since the technical effects achieved by the exercise electrocardiogram processing device provided by the present invention are the same as those achieved by the exercise electrocardiogram processing method provided by the present invention, they will not be described again here. Attached Figure Description

[0067] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0068] Figure 1 A flowchart of the exercise electrocardiogram processing method provided by the present invention;

[0069] Figure 2 A flowchart illustrating the exercise electrocardiogram processing method provided in an embodiment of the present invention;

[0070] Figure 3 This is an original exercise electrocardiogram waveform in an embodiment of the present invention;

[0071] Figure 4 This is a noise-reduced waveform obtained by an exercise electrocardiogram processing method in an embodiment of the present invention;

[0072] Figure 5 This is a schematic diagram of the exercise electrocardiogram processing system provided by the present invention;

[0073] Figure 6 This is a schematic diagram of the internal structure of the processing module in the exercise electrocardiogram processing system provided by the present invention. Detailed Implementation

[0074] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0075] The purpose of this invention is to provide a method and system for processing electrocardiograms during exercise, which can suppress noise interference and obtain accurate and stable electrocardiogram signal waveforms.

[0076] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0077] like Figure 1 As shown, the exercise electrocardiogram processing method of the present invention includes:

[0078] Step 100: Acquire motion ECG signal. Before acquiring the motion ECG signal, this invention needs to determine the object to be sampled so that the motion ECG signal can be obtained by collecting the ECG signal of the object under motion. The specific implementation steps are as follows:

[0079] To determine the experimental subjects to be collected, since testing a set of experimental data takes about 20 minutes, and considering that the participants need to have sufficient physical strength to withstand the exercise stress and maintain that physical strength for more than 20 minutes, we selected healthy undergraduate and graduate students aged 18 to 24 years old with no history of mental illness or psychological disorders as the test subjects for this experiment.

[0080] To determine the types of movement of the experimental subjects to be collected, 10 different movement states were first set, mainly including lying flat, standing upright, squatting, raising hands, jumping in place, deep breathing, walking slowly at a speed of 3km / h, walking briskly at a speed of 5km / h, jogging slowly at a speed of 6km / h, and jogging briskly at a speed of 8km / h.

[0081] Determine the collection environment, ensuring that the experimental site has good ventilation and appropriate temperature conditions, and also requires an indoor environment with a treadmill.

[0082] Ensure consistent data acquisition duration for each action. Under each exercise state, the duration of ECG signal acquisition should remain consistent. It is recommended to continuously acquire data for 1 minute under each state to ensure data sufficiency and stability.

[0083] Step 101: Perform empirical mode decomposition on the exercise ECG to obtain several intrinsic mode function components. For example, the implementation process of this step is as follows:

[0084] Step 1011: Traverse the ECG signal data, identify and mark all local maxima and local minima in the signal;

[0085] Step 1012: Construct upper and lower envelopes. For all local extreme points, construct envelopes using spline interpolation. The envelopes tightly wrap around all data points of the signal and are smooth with no intersections.

[0086] Step 1013: Calculate the mean curves of the upper and lower envelopes, subtract the mean curves from the original signal to obtain the preliminary mode function, iterate the preliminary mode function to obtain the first intrinsic mode function (IMF).

[0087] Step 1014: Process the remaining signal. Remove the first IMF from the original signal to obtain the remaining signal r1(t). Use r1(t) as the new input and repeat the above steps.

[0088] For example, in this step, r1(t) is used as the new input signal. The above steps are repeated to extract the second IMF and further subtract it to obtain a new residual signal. This process continues iterating until the residual signal becomes a monotonic function or contains a few extreme points, making further decomposition impossible. Through the above repeated decomposition, the exercise ECG signal is finally decomposed into several intrinsic mode function components and a residual component r. n (t), these components represent different frequency components in the original signal, and are used in subsequent electrocardiogram signal processing steps.

[0089] Step 102: Determine the frequency magnitudes of several intrinsic mode function components, and perform a high-pass filtering operation based on the low-frequency components of the intrinsic mode functions. For example, such as... Figure 2 As shown, the implementation process of this step can be as follows:

[0090] Step 1021: Analyze each Intrinsic Mode Function (IMF) component, calculate its instantaneous frequency and envelope, and then determine its frequency range and energy distribution.

[0091] Step 1022: Obtain and count the number of all IMFs and the total length of the ECG signal, so as to enable dynamic adjustment and adaptive processing of subsequent filter parameters.

[0092] Step 1023: Combining the instantaneous frequency and envelope information from the IMF, identify the frequency range of the low-frequency components and dynamically adjust the cutoff frequency f of the filter. c .

[0093] Step 1024: The filter order n is adaptively adjusted according to the number of IMFs and the signal length.

[0094] Step 1025: Process the low-frequency IMF components based on the designed adaptive high-pass filter.

[0095] Step 1026: All filtered IMFs are weighted, accumulated, and reconstructed. The weighting coefficients are adaptively calculated based on the energy proportion and frequency information of the IMFs.

[0096] Step 1027: Weight all IMF components and reconstruct the ECG signal to ensure that the effective signal components are preserved to the greatest extent while suppressing baseline drift noise in exercise ECG.

[0097] During this step, the weighting coefficient w for each IMF is determined based on its energy distribution and frequency characteristics during the reconstruction process. i This coefficient can be adaptively calculated based on the IMF's energy share and frequency information:

[0098]

[0099] Where E i f(w) represents the energy of the IMF. i ) is a frequency-dependent weighting function.

[0100] Step 103: Determine the remaining components obtained after suppressing low-frequency noise, and perform variational mode decomposition on the remaining components. Specifically, the implementation process of this step can be as follows:

[0101] Step 1031: Use a polynomial fitting method to de-trend the remaining components.

[0102] Step 1032: Set the number of modes (K) according to the frequency range to be decomposed from the ECG signal, and determine the value of the penalty parameter (α) through cross-validation.

[0103] Step 1033: Set the initial conditions for the Lagrange multipliers and the augmented Lagrange function, including the estimated values ​​of the initial mode functions and the initial values ​​of the Lagrange multipliers, and iteratively update the number of modes and the values ​​of the penalty parameters.

[0104] Step 1034: In each iteration, optimize the frequency and bandwidth of the spectrum of each modal component and update the modal function to minimize its bandwidth.

[0105] Step 1035: If the change in modal components between two consecutive iterations is less than a preset threshold or reaches the preset maximum number of iterations, the convergence condition is met, the optimal number of modes and penalty parameters are obtained, the corresponding modal components are output, and the optimal result of variational mode decomposition is achieved.

[0106] In this step, the initial estimates of the mode functions and the initial values ​​of the Lagrange multipliers are set to 0, allowing the algorithm to gradually adjust the multiplier values ​​from an equilibrium state and converge to the optimal solution. If the change in the mode components between two consecutive iterations is less than a preset threshold, which is set within a very small range of 10... -4 Up to 10 -6 This ensures the algorithm stops iterating when it is sufficiently close to the optimal solution. The preset maximum number of iterations is set between 100 and 500.

[0107] Step 104: Extract variational mode functions at different frequencies and determine the magnitude of the center frequency. The implementation process for this step is as follows:

[0108] Step 1041: Discretize each variational mode function into a sequence of sampling points and perform a fast Fourier transform to obtain a frequency domain signal, representing the amplitude and phase of the variational mode function at different frequencies.

[0109] Step 1042: Calculate the square of the amplitude of the frequency domain signal and observe the energy distribution of the variational mode function at different frequencies.

[0110] Step 1043: Normalize the calculated amplitude square, i.e., the power spectral density. In the normalized power spectral density curve, find the maximum peak point to preliminarily determine the dominant frequency component of the mode function, and then analyze the peak value around the dominant frequency w. p Find the frequency point w where the energy on both sides decays to a certain set threshold (e.g., 3dB). l and w h Determine the frequency bandwidth Δw, and adjust the bandwidth range according to the energy concentration of the mode function to ensure that the energy proportion within the frequency band reaches the predetermined standard (such as 95% of the energy is concentrated within the frequency band).

[0111] Step 1044: Calculate the centroid of the spectrum, i.e., the weighted average of the frequencies, as the center frequency w. c Accurate estimation of the centroid frequency w c With peak clock speed w p The comparisons are made, and the deviations are calculated. If the deviations are large, the spectral bandwidth range needs to be reassessed and the spectral centroid recalculated to ensure the accuracy of the spectral analysis.

[0112] As can be seen in step 104, the frequency bandwidth Δw is calculated using the following formula:

[0113] Δw=w h -w l

[0114] Among them, w l and w h These represent the lower and upper limits of the frequency bandwidth, respectively, with Δw reflecting the width of the mode function's bandwidth. The above process accurately determines the center frequency of the variational mode function through Fourier transform and spectral analysis, ensuring the accuracy and reasonableness of the final result.

[0115] Step 105: Determine the relationship between the center frequency of each variational mode function and the preset center frequency. If the center frequency is less than the preset center frequency, it is classified as a low-frequency mode. If the center frequency is greater than or equal to the preset center frequency, it is classified as a high-frequency mode.

[0116] Step 106: For low-frequency mode functions with a center frequency less than the preset value, nonlocal mean denoising is used; for high-frequency mode functions with a center frequency greater than or equal to the preset value, singular value decomposition is used to suppress baseline interference. For example, ... Figure 2 As shown, the implementation process for this step can be as follows:

[0117] Step 1061: Nonlocal means denoising algorithm, firstly divide the signal into a series of local blocks of fixed size, for each local block B i Search for a local block B that is similar to it across the entire range of the signal. j .

[0118] Step 1062: Use the Gaussian function to represent the searched similar block B. j With current block B i The similarity between blocks is calculated, and weight coefficients are assigned to blocks that are closer in distance and have higher similarity.

[0119] Step 1063: Calculate a weighted average of all similar blocks using the weights mentioned above, and update the current block B. i The value of this weighted average operation can suppress noise while preserving the main characteristics of the signal.

[0120] Step 1064: Singular Value Decomposition Algorithm. First, each high-frequency mode function is transformed into a Hankel matrix to characterize the temporal features of the signal.

[0121] Step 1065: Perform singular value decomposition on the constructed Hankel matrix, decomposing the matrix into singular value matrices and corresponding left and right singular vector matrices. Sort the decomposed singular values ​​in descending order, and then calculate the difference sequence between adjacent singular values. The difference sequence is used to identify the boundary between the main components and noise components in the signal.

[0122] Step 1066: Determine the cutoff point r based on the difference sequence, retain the first r singular values, and reconstruct the denoised high-frequency mode function using the retained main singular values ​​and their corresponding singular vector matrices.

[0123] Step 106 demonstrates the application of nonlocal mean denoising and singular value decomposition in processing low-frequency and high-frequency mode functions, ensuring that baseline interference noise is effectively suppressed after the entire mode function is smoothed.

[0124] Step 107: Reconstruct the denoised mode function, use wavelet thresholding to suppress electromyographic interference noise in the exercise ECG, and finally obtain the clean exercise ECG signal waveform after noise interference suppression. The implementation process of this step can be as follows:

[0125] Step 1071: Weight all low-frequency mode functions and high-frequency mode functions to form a preliminary noise-reduced ECG signal, and perform wavelet transform on the reconstructed ECG signal.

[0126] Step 1072: Based on the characteristics of the exercise ECG signal, select "db4" as the wavelet basis function to obtain the wavelet coefficients of different frequency subbands, and perform adaptive threshold calculation.

[0127] Step 1073: Use the median absolute deviation of wavelet coefficients to estimate the noise level at each scale, and adaptively calculate the threshold at different scales. For high-frequency subbands containing more noise components, use a higher threshold for stronger noise reduction, and for low-frequency subbands containing the main signal components, use a lower threshold to preserve the signal details.

[0128] Step 1074: Perform inverse wavelet transform on the wavelet coefficients after thresholding, and recombine the wavelet coefficients at each scale to recover the time domain signal.

[0129] The exercise ECG signal processed in step 107 has a higher signal-to-noise ratio. For example... Figure 3 The image shows the original waveform of an electrocardiogram (ECG) signal acquired during a slow jog at 6 km / h. As can be seen, the ECG signal contains significant noise interference, obscuring its characteristic waveform. After noise reduction using the exercise ECG processing method described in this paper, the resulting waveform is as follows. Figure 4 As shown, the noise in the ECG signal is greatly suppressed after algorithm processing, while the key features of the ECG signal are preserved, making it suitable for further medical diagnosis and analysis.

[0130] Based on the above description, the working principle of the exercise electrocardiogram processing method provided by the present invention is as follows:

[0131] The target object was identified, and ECG signals were collected under different motion states to obtain raw ECG waveforms, which were then plotted. Next, empirical mode decomposition (EMD) was performed on the motion ECG, decomposing the ECG signal into several intrinsic mode functions (IMFs). High-pass filtering was applied to the low-frequency IMFs to suppress low-frequency noise in the motion ECG. Then, variational mode decomposition (VMD) was performed on the remaining components to more accurately extract variational mode functions of different frequencies. The center frequency of each variational mode function was determined. For mode functions with frequencies lower than the center frequency, nonlocal mean denoising was used; for mode functions with frequencies greater than or equal to the center frequency, singular value decomposition (SVD) was used to suppress baseline interference. Finally, the denoised mode functions were reconstructed. Wavelet thresholding was used to reduce EMG interference noise in the motion ECG. The clean motion ECG waveform after noise suppression was plotted for further analysis.

[0132] The present invention also provides a sports electrocardiogram processing system, such as Figure 5 As shown, the system includes:

[0133] The 500 exercise ECG acquisition module is used to collect ECG signals from test subjects under different exercise conditions.

[0134] Storage module 501 is used to store computer control programs.

[0135] The processing module 502 is connected to the exercise ECG acquisition module 500 and the storage module 501 respectively. It is used to retrieve and execute the computer control program based on the exercise ECG signal to implement the exercise ECG processing method provided above and obtain a clean exercise ECG signal waveform after noise suppression.

[0136] The processing device 502 includes:

[0137] The exercise ECG acquisition module 5021 is used to acquire ECG signal waveform data under different exercise states;

[0138] The empirical mode decomposition module 5022 is used to obtain the intrinsic mode function components and perform high-pass filtering to suppress baseline drift interference in exercise electrocardiogram.

[0139] The variational mode decomposition module 5023 is used to perform variational mode decomposition on the remaining components obtained after suppressing baseline drift noise interference, and to set an appropriate number of modes and penalty parameters;

[0140] The center frequency determination module 5024 is used to determine the magnitude of the center frequency of variational mode functions at different frequencies and to verify the degree of deviation between the center frequency and the dominant frequency.

[0141] The nonlocal mean denoising module 5025 is used to suppress noise in low-frequency mode functions while preserving the main characteristics of the signal.

[0142] The singular value decomposition module 5026 is used to suppress noise in the high-frequency mode function. It uses the difference decomposition method to retain the main singular value part and obtain the noise-reduced high-frequency mode function.

[0143] The wavelet threshold denoising module 5027 is used to suppress residual noise in exercise electrocardiograms, adaptively calculate thresholds at different scales, and perform threshold denoising processing on different frequency subbands.

[0144] Furthermore, storage device 501 may be a computer-readable storage medium.

[0145] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0146] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for processing exercise electrocardiograms, characterized in that, include: Acquire electrocardiogram (ECG) signals during exercise; Empirical mode decomposition is performed on the electrocardiogram signal under the aforementioned motion state to obtain several intrinsic mode function components; The method involves determining the frequency magnitudes of several Intrinsic Mode Function (IMF) components and performing high-pass filtering on the low-frequency components of the IMFs to suppress low-frequency noise in exercise electrocardiograms (ECGs). This includes: analyzing each IMF component, calculating its instantaneous frequency and envelope to determine its frequency range and energy distribution; acquiring and counting the number of all IMFs and the total length of the ECG signal; identifying the frequency range of low-frequency components by combining the instantaneous frequency and envelope information of the IMFs, and dynamically adjusting the cutoff frequency of the filter; adaptively adjusting the filter order based on the number of IMFs and the signal length; processing the low-frequency IMF components based on the designed adaptive high-pass filter; weighting and reconstructing all filtered IMFs, with the weighting coefficients adaptively calculated based on the energy proportion and frequency information of the IMFs; and reconstructing the ECG signal by weighting and accumulating all IMF components to obtain the remaining components after suppressing low-frequency noise. Variational mode decomposition is performed on the remaining components, including: using a polynomial fitting method to de-trend the remaining components; setting the number of modes according to the frequency range to be decomposed from the ECG signal, and determining the value of the penalty parameter through cross-validation; setting the initial conditions for the Lagrange multipliers and augmented Lagrange functions, including the estimated value of the initial mode function and the initial value of the Lagrange multipliers, and iteratively updating the number of modes and the value of the penalty parameter; in each iteration, optimizing the frequency and bandwidth of the spectrum of each mode component, and updating the mode function to minimize its bandwidth; if the change in the mode component between two consecutive iterations is less than a preset threshold or reaches the preset maximum number of iterations, the convergence condition is met, the optimal number of modes and penalty parameter are obtained, the corresponding mode components are output, and the optimal result of variational mode decomposition is achieved. Extract variational mode functions at different frequencies to determine the magnitude of the center frequency; Determine the relationship between the center frequency of each variational mode function and the preset center frequency; For mode functions with frequencies lower than the center frequency, nonlocal mean denoising is used; for mode functions with frequencies greater than or equal to the center frequency, singular value decomposition denoising is used to achieve baseline interference suppression. Specifically, this includes: For low-frequency mode functions, a nonlocal mean denoising algorithm is used. First, for each low-frequency mode function, all local blocks in the signal are traversed, and blocks similar to the current block are searched. The weights are calculated based on the similarity between the searched similar blocks and the current block, and the weight coefficients are represented by a Gaussian function. The value of the current block is updated by using a weighted average of similar blocks based on the aforementioned weights. By combining the blocks that have undergone nonlocal mean processing, the noise-reduced low-frequency mode function can be reconstructed. For high-frequency mode functions, singular value decomposition is used for noise reduction, transforming each high-frequency mode function into a Hankel matrix; Perform singular value decomposition on the constructed matrix to decompose the matrix into singular value matrices and corresponding left and right singular vector matrices; Sort the decomposed singular values ​​and then calculate the difference sequence between adjacent singular values; The cutoff point is determined based on the difference sequence, and the singular value part of the corresponding electrocardiogram signal is retained; By using the retained main singular values ​​and their corresponding singular vector matrices, the signal is reconstructed, thereby obtaining the denoised high-frequency mode function; The denoised mode function is reconstructed, and wavelet thresholding is used to suppress electromyographic interference noise in the exercise ECG. Finally, a clean exercise ECG signal waveform with noise interference suppressed is obtained.

2. The method for processing exercise electrocardiograms according to claim 1, characterized in that, Extracting the variational mode functions at different frequencies and determining the magnitude of the center frequency specifically includes: Perform a discrete Fourier transform on each extracted variational mode function to convert the mode function in the time domain to the frequency domain; The amplitude squared operation is performed on the modal function in the frequency domain to obtain the energy distribution of the signal at each frequency; Normalized power spectral density, by finding the maximum peak value in the power spectral density curve, preliminarily determines the dominant frequency component of the mode function; The frequency bandwidth range is determined based on the initially determined peak frequency. Further calculation of the spectral centroid is performed, and the power spectral density-weighted frequency average is used as an accurate estimate of the center frequency. Verify the rationality of the center frequency and check whether the center frequency is close to the previously determined peak value of the main frequency. If the center frequency deviates significantly from the main frequency, the spectrum bandwidth needs to be reassessed to ensure the accuracy of the spectrum analysis.

3. The method for processing exercise electrocardiograms according to claim 1, characterized in that, The determination of the magnitude of the center frequency of each variational mode function relative to the preset center frequency specifically includes: Determine the center frequency of each variational mode function. If the center frequency is less than the preset center frequency, it is classified as a low-frequency mode. If the center frequency is greater than or equal to the preset center frequency, it is classified as a high-frequency mode.

4. The method for processing exercise electrocardiograms according to claim 1, characterized in that, The denoised mode function is reconstructed, and wavelet thresholding is used to suppress electromyographic interference noise in the exercise ECG. Finally, a clean exercise ECG signal waveform with noise suppression is obtained, specifically including: The low-frequency and high-frequency mode functions after nonlocal mean and singular value decomposition noise reduction are superimposed to reconstruct the preliminarily denoised ECG signal. Wavelet transform is performed on the reconstructed electrocardiogram signal to decompose the signal into wavelet coefficients of multiple sub-bands with different frequencies; An adaptive threshold calculation is applied to the wavelet coefficients obtained from the decomposition. Based on the noise estimation results of wavelet coefficients at each scale, the thresholds at different scales are adaptively calculated. The wavelet threshold function is applied to process the wavelet coefficients at each scale to obtain the motion ECG signal waveform after noise interference suppression.

5. A sports electrocardiogram processing system, characterized in that, include: The exercise ECG acquisition module is used to acquire ECG signals during exercise. Storage module, used to store computer control programs; The processing module is connected to the exercise ECG acquisition module and the storage module respectively, and is used to retrieve and execute the computer control program based on the ECG signal under the exercise state, so as to implement the exercise ECG processing method as described in any one of claims 1-4, and obtain a clean exercise ECG signal waveform after noise suppression.

6. The exercise electrocardiogram processing system according to claim 5, characterized in that, The processing module includes: The exercise ECG acquisition module is used to acquire ECG signal waveform data under different exercise states; The Empirical Mode Decomposition (EMD) module is used to obtain intrinsic mode function components and perform high-pass filtering to suppress low-frequency noise in exercise ECG. The variational mode decomposition module is used to perform variational mode decomposition on the remaining components obtained after suppressing low-frequency noise, and to set an appropriate number of modes and penalty parameters; The center frequency determination module is used to determine the magnitude of the center frequency of variational mode functions at different frequencies and to verify the degree of deviation between the center frequency and the dominant frequency. The nonlocal mean denoising module is used to suppress noise in low-frequency mode functions while preserving the main characteristics of the signal; The singular value decomposition module is used to suppress noise in high-frequency mode functions. It uses the difference decomposition method to retain the main singular value components. The wavelet threshold denoising module is used to suppress electromyographic interference noise in exercise electrocardiograms. It adaptively calculates thresholds at different scales and performs threshold denoising processing on different frequency subbands.

7. The exercise electrocardiogram processing system according to claim 5, characterized in that, The storage module is a computer-readable storage medium.

Images