ECG multi-source interference classification identification and hierarchical combined suppression method

Abstract

This application discloses a method for classifying, identifying, and jointly suppressing multi-source interference in ECG. The method includes: preprocessing ECG and IMU signals to obtain a signal set; extracting features from the signal set and identifying interference types; obtaining interference intensity levels based on the interference types and classifying motion intensity levels based on the interference intensity levels; eliminating linear mutual interference components and compensating for nonlinear mutual interference in the signal set through decoupling operations; performing multi-level hierarchical suppression processing on the decoupled ECG signal to obtain purified ECG signal data; suppressing motion artifacts and labeling the IMU signal to obtain interference distribution data and adjust parameter configurations; and obtaining filter parameter data for quality assessment. Implementing this embodiment classifies multi-source interference in ECG, eliminates coupling mutual interference between ECG and IMU signals, and balances interference suppression effectiveness with ECG feature fidelity, making it suitable for real-time monitoring in wearable devices.

Description

Technical Field

[0001] This application relates to the field of electromyography (EMG) signal technology, specifically to a method, system, electronic device, and computer-readable storage medium for ECG multi-source interference classification, identification, and hierarchical joint suppression. Background Technology

[0002] In the field of wearable ECG monitoring, ECG signals (electrocardiogram, a waveform signal of cardiac electrical activity recorded by electrodes on the body surface, whose core features include QRS wave, ST segment, and T wave, and is the core basis for cardiovascular health assessment) are susceptible to multi-source interference, such as power frequency interference, electromyographic interference, motion artifacts, baseline drift, and coupling interference from IMU signals (Inertial Measurement Unit, which in this solution is a 6-axis inertial sensor, integrating a three-axis accelerometer and a three-axis gyroscope, used to collect inertial data of human chest movement state and posture changes). This leads to distortion of key ECG features of QRS wave, ST segment, and T wave, seriously reducing the accuracy of arrhythmia analysis and cardiovascular health assessment. Existing ECG signal anti-interference technologies generally suffer from the following defects: interference is not classified, and the use of "one-size-fits-all" filtering cannot distinguish between different types of interference such as power frequency, electromyography, motion artifacts, and mutual interference. The uniform use of fixed filtering can easily lead to the loss of useful signals or incomplete interference suppression. The coupling interference between IMU signal and ECG signal was not considered: In miniaturized integrated devices, the two signals interfere with each other due to power coupling and electromagnetic crosstalk. Traditional single-channel filtering cannot eliminate this type of coupling noise, which becomes the main cause of signal distortion under strong interference. Existing technologies lack a graded suppression mechanism and do not dynamically adjust the filtering strategy according to the intensity of interference and the state of motion. They are prone to over-filtering in the resting state and insufficient suppression in the motion state, and cannot balance the anti-interference effect with the fidelity of ECG characteristics. Therefore, the industry urgently needs an anti-interference method that can automatically classify and identify multi-source interference and perform hierarchical joint suppression according to the type and intensity of interference, in order to solve the problems of coarse filtering, inability to eliminate mutual interference, poor adaptability to scenarios, and difficulty in real-time implementation of existing technologies. Summary of the Invention

[0003] To address the problems existing in the prior art, this application solves the technical problems of coexistence of multiple sources of interference and easy distortion of features in wearable ECG signals by combining feature recognition, coupling and mutual interference decoupling, hierarchical suppression and closed-loop calibration, and is used to output high-fidelity IMU signal data and ECG signal data. The purpose of this application is to provide a method, system, electronic device and computer-readable storage medium for ECG multi-source interference classification, recognition and hierarchical joint suppression.

[0004] In a first aspect, the ECG multi-source interference classification, identification, and hierarchical joint suppression method described in this application includes the following steps: The synchronously acquired ECG and IMU signals are preprocessed to obtain a preprocessed signal set; Feature set extraction and interference type identification are performed on the signal set; The interference intensity level is obtained based on the interference type, and the motion intensity level is classified based on the interference intensity level. The linear mutual interference components of the signal set are eliminated and the nonlinear mutual interference is compensated by decoupling operation to obtain the decoupled ECG signal and IMU signal. The decoupled ECG signal is subjected to multi-level hierarchical suppression processing to obtain purified ECG signal data; Motion artifact suppression and annotation are performed on the IMU signals in the signal set to obtain interference distribution data; The purified ECG signal data and the interference distribution data are adjusted by parameter configuration to obtain filter parameter data; The quality of the filtered parameter data is evaluated to obtain high-fidelity ECG signal data and IMU signal data.

[0005] Preferably, the preprocessing of the synchronously acquired ECG and IMU signals to obtain a preprocessed signal set includes: The ECG signal and the IMU signal are read to obtain the original signal dataset; The original signal dataset is filtered to obtain a first signal set; The sampling rate of the first signal set is unified to obtain the second signal set; The third signal set is obtained by detecting and eliminating signals from the second signal set. The third signal set is timestamped to obtain the fourth signal set; The fourth signal set is smoothed to obtain a preprocessed signal set.

[0006] Preferably, the step of extracting features from the signal set and identifying the interference type, obtaining the interference intensity level, and classifying the motion intensity level according to the interference intensity level includes: The interference type is determined from the correlation values ​​in the signal set, and an interference classification label is obtained; The interference classification labels are weighted and quantitative indicators are calculated to obtain the interference intensity level; The frequency distribution of the IMU signal in the interference intensity level is detected to obtain the motion intensity level.

[0007] Preferably, the step of eliminating linear mutual interference components and compensating for nonlinear mutual interference in the signal set through decoupling operations to obtain the decoupled ECG signal and IMU signal includes: Adaptive calculations are performed on the signal set to obtain the distribution parameters of linear interference; The distributed parameters are separated to obtain the intermediate signal after removing linear interference. The intermediate signal is combined with a second-order Volterra series model to analyze the nonlinear interference and calculate the compensation value of the nonlinear interference to obtain the compensated signal data. The compensated signal data is then detected to obtain the decoupled ECG signal and IMU signal.

[0008] Preferably, the step of performing multi-level suppression processing on the decoupled ECG signal to obtain purified ECG signal data includes: The decoupled ECG signal is subjected to power frequency interference suppression through an adaptive notch filter to obtain intermediate signal data; The intermediate signal data is processed using the EEMD algorithm and an adaptive soft threshold to suppress electromyographic interference and obtain a transition signal. The transition signal is processed using the VMD algorithm and IMU motion reference to suppress motion artifacts and baseline drift, resulting in purified ECG signal data.

[0009] Preferably, the step of performing motion artifact suppression and annotation on the IMU signals in the signal set to obtain interference distribution data, and adjusting the parameters of the purified ECG signal data and the interference distribution data to obtain filter parameter data includes: The IMU signals in the signal set are labeled to obtain labeled interference distribution data; The labeled interference distribution data is analyzed to determine the specific time period and amplitude range of the interference, and the interference feature data after positioning is obtained. Simultaneously, the ECG signal and the IMU signal in the signal set are compared synchronously to obtain the signal matching result; The signal matching results are processed in layers to obtain filter parameter data.

[0010] Preferably, the step of performing quality assessment on the filtered parameter data to obtain high-fidelity ECG signal data and IMU signal data includes: The ECG signal data and the IMU signal data in the filter parameter data are extracted to obtain signal component data through signal decomposition technology; Calculate the signal-to-noise ratio (SNR) value for the signal component data to obtain the noise percentage data; The noise proportion data is labeled to obtain labeled component state data. The integrity of feature points is determined from the marked component state data to obtain feature fidelity evaluation result data. Based on the evaluation results of the feature fidelity, the integrity of the feature points is determined, the state data of the comprehensive evaluation is obtained, and then the signal-to-noise ratio and the feature fidelity are determined to obtain high-fidelity ECG signal data and IMU signal data.

[0011] Secondly, the ECG multi-source interference classification, identification, and hierarchical joint suppression system described in this application includes: The data acquisition module preprocesses the synchronously acquired ECG and IMU signals to obtain a preprocessed signal set. The feature extraction module extracts features from the signal set and identifies the type of interference. The interference intensity level is obtained based on the interference type, and the motion intensity level is classified based on the interference intensity level. The decoupling operation module eliminates linear mutual interference components and compensates for nonlinear mutual interference in the signal set through decoupling operations, thereby obtaining the decoupled ECG signal and IMU signal. The suppression processing module performs multi-level hierarchical suppression processing on the decoupled ECG signal to obtain purified ECG signal data. The purification module performs motion artifact suppression and annotation on the IMU signals in the signal set to obtain interference distribution data. It then adjusts the parameters of the purified ECG signal data and the interference distribution data to obtain filter parameter data. The quality assessment module performs a quality assessment on the filtered parameter data to obtain high-fidelity ECG signal data and IMU signal data.

[0012] Thirdly, this application also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the ECG multi-source interference classification, identification and hierarchical joint suppression method as described in the first aspect.

[0013] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the ECG multi-source interference classification, identification, and hierarchical joint suppression method as described in the first aspect.

[0014] Implementing this embodiment enables automatic classification and identification of multi-source interference, accurately locating the source of interference. By jointly judging the time domain, frequency domain, and cross-correlation features, it can automatically distinguish the types of interference such as mutual interference, power frequency interference, electromyography, motion artifacts, and baseline drift, and quantify their intensity, providing a precise basis for graded suppression and avoiding blind filtering. The hierarchical joint suppression provides more accurate anti-interference and higher fidelity. It adopts a three-level hierarchical filtering architecture, which uses corresponding algorithms to suppress different interferences step by step. In strong motion scenarios, it can significantly improve the ECG signal-to-noise ratio, while fully preserving key features such as QRS wave and ST segment, with high feature fidelity. Eliminate the coupling and mutual interference between IMU signal and ECG signal at the source. By constructing a mutual interference model and adaptive decoupling, the electromagnetic crosstalk and power supply coupling noise caused by integrated hardware are effectively eliminated, solving the core pain point that traditional methods cannot handle mutual interference. By combining IMU signals to achieve bidirectional closed-loop optimization, the suppression effect is stronger. The IMU signal is fed back to the artifact suppression stage of the ECG signal in a closed loop, which significantly improves the accuracy of artifact removal under motion conditions. This joint optimization is more stable and reliable than the filtering effect of a single ECG signal. It is highly wearable and adaptable to different environments. It can adaptively adjust filtering parameters according to the intensity of movement and interference, and supports hierarchical power consumption management. The algorithm has been optimized for lightweight operation and can run in real time on a low-power MCU, meeting the engineering requirements of wearable devices for long battery life, low latency, and dynamic monitoring. Attached Figure Description

[0015] Figure 1 This is a flowchart of a method for classifying, identifying, and hierarchically suppressing ECG multi-source interference as described in this application; Figure 2 This is a schematic diagram of the structural composition of the ECG multi-source interference classification, identification, and hierarchical joint suppression system described in this application; Figure 3 This is a framework diagram of the terminal device provided in the embodiments of this application. Detailed Implementation

[0016] Example 1 like Figure 1 As shown, the ECG multi-source interference classification, identification, and hierarchical joint suppression method described in this application includes: S101. Preprocess the synchronously acquired ECG and IMU signals to obtain the preprocessed signal set; S102. Extract feature sets from the signal set and identify interference types; The interference intensity level is obtained based on the type of interference, and the motion intensity level is classified according to the interference intensity level. S103. The linear mutual interference components of the signal set are eliminated and the nonlinear mutual interference is compensated through decoupling operation to obtain the decoupled ECG signal and IMU signal. S104. Perform multi-level hierarchical suppression processing on the decoupled ECG signal to obtain purified ECG signal data. S105. Perform motion artifact suppression and labeling on the IMU signals in the signal set to obtain interference distribution data. Adjust the parameters of the purified ECG signal data and the interference distribution data to obtain filter parameter data. S106. Evaluate the quality of the filter parameter data and obtain high-fidelity ECG signal data and IMU signal data.

[0017] Furthermore, S101 includes: The ECG signal and the IMU signal are read to obtain the raw signal dataset; The original signal dataset is filtered to obtain the first signal set; The sampling rate of the first signal set is unified to obtain the second signal set; The third signal set is obtained by detecting and eliminating signals from the second signal set. The third signal set is timestamped to obtain the fourth signal set; The fourth signal set is smoothed to obtain the preprocessed signal set.

[0018] In a specific embodiment, the raw signal dataset is obtained by initially reading the synchronously acquired data of ECG and IMU signals; The DC component in the original signal dataset is processed using a high-pass filter to obtain the first signal set after removing the DC component. Based on the sampling rate difference of the first signal set, the sampling rate of the ECG signal and the IMU signal is unified by interpolation to obtain the second signal set after sampling rate alignment; For abnormal data in the second signal set, if the detected signal value exceeds the preset threshold range, it is marked as invalid data and removed to obtain the third signal set after removing abnormal data. By comparing the timestamps of the third signal set and adjusting the time axes of the ECG and IMU signals, the fourth signal set after time synchronization is determined. Based on the characteristics of the fourth signal set, a sliding window method is used to smooth the signal and obtain the preprocessed signal set.

[0019] Specifically, in S101, when preprocessing the synchronously acquired ECG and IMU signals, the DC removal process is achieved by calculating the mean of the signal and subtracting the mean from the ECG and IMU signals. For example, for the ECG signal, assuming the acquired raw data is a sequence of 1000 points, its mean is calculated to be 0.5mV. Subtracting 0.5mV from each data point can eliminate the DC component and obtain a signal with zero mean. This process can be automatically completed by the program through a simple subtraction operation, with low algorithm complexity and strong real-time performance. Resampling alignment is performed because the ECG signal and IMU signal may have different sampling rates, such as 500Hz for the ECG signal and 100Hz for the IMU signal. Linear interpolation is used to upsample the IMU signal to 500Hz. Specifically, the interpolation point is calculated based on the time axis. For example, if the inertial measurement unit data at time points 0.01 seconds and 0.02 seconds are 1.2 and 1.8 respectively, then the data at the interpolation point at 0.015 seconds is 1.5. Time alignment of the two signals is achieved in this way. The program can automatically calculate the interpolation based on the timestamp to ensure synchronization. Outlier removal is performed using a statistical method based on three times the standard deviation. Assuming the standard deviation of the ECG signal after DC removal is 0.3mV, data exceeding the mean ±0.9mV are considered outliers and replaced with the average of neighboring points. For example, if a value is 1.2mV, exceeding the threshold, it is replaced with the average of the two points before and after it, which is 0.4mV. This process is automatically completed by the algorithm traversing the data. Combined with the physiological characteristics of the ECG signal, it ensures that outliers are removed while signal characteristics are preserved. From DC offset removal to time resampling and outlier removal to improve data quality, each step can be automated through programming to obtain time-aligned and quality-optimized signals, providing a data foundation for analyzing the correlation between heart rate variability and exercise status.

[0020] Specifically, in S101, the calculation for removing the DC component of the signal by mean subtraction is as follows:

[0021] Where x is the original ECG signal, in mV; The original ECG signal mean value, in mV; The signal is the ECG signal after DC removal, in mV; N is the total number of signal sampling points, dimensionless, a positive integer. In the calculation of aligning the sampling rates of ECG and IMU signals through linear interpolation:

[0022] Where t is the target time for interpolation, in seconds; t i ,ti+1 The units are adjacent sampling times of the original signal, in seconds; y i ,y i+1 For the original signal at t i ,t i+1 The value at the location is in mV; y is the interpolated signal value. In identifying and replacing signal outliers using the 3σ criterion:

[0023]

[0024] Where x is the signal value at the current sampling point, in mV; σ is the signal mean, in mV; σ is the signal standard deviation, in mV; x -1 x +1 The signal values ​​of the adjacent sampling points before and after the current point, in mV; The signal value after replacing the outlier is in mV. In the process of smoothing and denoising signals by using a sliding window averaging method:

[0025] Where L is the length of the smoothing window, a dimensionless, positive odd number; x k The values ​​of each sampling point within the window are in mV; x s The signal value is after smoothing, in mV.

[0026] Furthermore, S102 includes: The interference type is determined from the correlation values ​​in the signal set, and the interference classification label is obtained; The interference classification labels are weighted to calculate quantitative indicators and obtain the interference intensity level; Frequency distribution detection is performed on the IMU signal in the interference intensity level to obtain the motion intensity level.

[0027] In a specific embodiment, the ECG signal and IMU signal in the signal set are decomposed using time-domain analysis methods to obtain the basic characteristics of the signal in the time dimension and obtain a time-domain feature set. Based on the time-domain feature set, the ECG signal and IMU signal are transformed using frequency-domain analysis methods to extract the signal distribution information in the frequency dimension and determine the frequency-domain feature set. For the time-domain feature set and the frequency-domain feature set, the cross-correlation analysis method is used to calculate the correlation between the ECG signal and the IMU signal, obtain the cross-feature between the two, and obtain the cross-correlation feature set; Based on the cross-correlation feature set, if the correlation value between the ECG signal and the IMU signal exceeds the preset threshold range, it is determined to be a specific type of interference, and an interference classification label is obtained. The interference classification labels are weighted by the key values ​​in the cross-correlation feature set to calculate the quantitative index of interference intensity and determine the interference intensity level. The IMU signal in the interference intensity level is detected. If the frequency distribution of the frequency domain feature set in the IMU signal is concentrated in the preset high frequency range, it is classified as a higher motion intensity, and the motion intensity level is obtained. The interference intensity level and motion intensity level are mapped in a unified manner through data integration methods, and a comprehensive classification result is output.

[0028] Specifically, in S102, features can be extracted and classified and graded by an automated algorithm when processing the preprocessed ECG and IMU signals. For time-domain feature extraction, the peak interval of the ECG signal is automatically calculated. For example, in a signal with a length of 2000 points, by detecting the position of the R wave peak, assuming that the intervals between adjacent peaks are 0.8 seconds, 0.82 seconds and 0.79 seconds respectively, the average interval is calculated to be 0.803 seconds as a heart rate-related feature. The algorithm uses the sliding window method to traverse the data to ensure that the changes in each cycle are captured. In frequency domain feature extraction, the signal is converted into a spectrum using Fast Fourier Transform (FFT). For example, after transforming the ECG signal, the dominant frequency component is found to be concentrated around 1.2Hz, indicating the heart rate-dominant frequency. Simultaneously, the power spectral density in the 0.5 to 5Hz frequency band is extracted, and the total power is calculated to be 2.3mV. 2 , used to evaluate signal energy distribution; For cross-correlation feature extraction, the correlation coefficient between ECG signal and IMU signal is calculated. For example, by using the normalized cross-correlation algorithm, the maximum correlation coefficient is 0.75, indicating that there is a strong correlation between the two within a specific time window with a time delay of 0.02 seconds, which is used to identify potential sources of interference. Interference type identification is achieved through frequency domain abnormal peak detection, such as detecting an abnormal power of 1.8mV at 8Hz in the spectrum of an ECG signal. 2 Based on database comparison, it was determined to be power interference. The interference intensity was quantified by the abnormal power ratio, which was calculated to be 15%. The program automatically output the classification result as "power interference, medium intensity". Motion intensity levels are determined based on IMU signals, and the root mean square (RMS) value of the acceleration signal is extracted. For example, the RMS value within a 10-second window is calculated to be 2.5 m / s². 2The algorithm automatically classifies exercise into "medium intensity" by using preset thresholds (low intensity: x < 1.5, medium intensity: 1.5 ≤ x < 3, high intensity: x ≥ 3) and combines ECG signal characteristics to analyze the impact of exercise on heart rate. All steps are completed automatically by the algorithm.

[0029] Specifically, in S102, the time interval between adjacent R waves is calculated as follows:

[0030] Where, Δ nR f is the number of sampling points between adjacent R waves, dimensionless; s R is the signal sampling rate, in Hz; R is the heartbeat interval, in seconds. In calculating the power spectral density of an ECG signal using FFT:

[0031] Where x is the input ECG signal, in mV; X is the complex FFT result of the ECG signal, in mV; N is the number of signal sampling points, dimensionless; and P is the power spectral density, in mV². In calculating the normalized cross-correlation coefficient between the ECG signal and the IMU signal:

[0032] Where x is the ECG signal, unit: mV; y is the IMU signal (acceleration / angular velocity); Cov(x,y) is the covariance between the ECG signal and the IMU signal; σ x σ is the standard deviation of the ECG signal. y ρ is the standard deviation of the IMU signal; ρ is the normalized cross-correlation coefficient, dimensionless, with a range of [-1, 1]. In measuring the power frequency interference intensity by the proportion of power frequency power:

[0033] Among them, P 50 This refers to the signal power in the 50Hz power frequency band; P all The total power of the ECG signal across the entire frequency band; I represents the power frequency interference intensity, dimensionless and percentage-based. In classifying exercise intensity levels using triaxial acceleration RMS values:

[0034] Among them, a x ,a y ,a z For IMU triaxial acceleration; W is the number of sampling points in the RMS calculation window, dimensionless, a positive integer; a rms This is the root mean square value of acceleration (motion intensity). Grading threshold: Low intensity: a rms <1.5m / s 2 Medium strength: 1.5 m / s 2 ≤a rms <3m / s 2 High strength: a rms ≥3m / s 2 ; In one embodiment, the exercise intensity is further divided into four levels: resting (X<0.2g), low intensity (0.2g≤x<0.6g), medium intensity (0.6g≤x<1.2g), and high intensity (X≥1.2g).

[0035] Furthermore, S103 includes: Adaptive calculations are performed on the signal set to obtain the distribution parameters of linear interference; The distributed parameters are separated to obtain the intermediate signal after removing linear interference; The intermediate signal is combined with a second-order Wolter series model to analyze the nonlinear interference and calculate the compensation value of the nonlinear interference to obtain the compensated signal data. The compensated signal data is then detected to obtain the decoupled ECG signal and IMU signal.

[0036] In a specific embodiment, the ECG and IMU signals in the signal set are filtered using a preprocessing method to remove high-frequency noise and baseline drift, thereby obtaining pre-cleaned signal data. Based on the preliminary cleaned signal data, a mutual interference model of coupled interference is constructed. The coupling coefficients of ECG signals and IMU signals in the signal set are adaptively calculated using a recursive least squares algorithm to determine the distribution parameters of linear interference. To address the distributed parameters of linear interference, a decoupling method is employed to separate the ECG signal and the IMU signal, thereby obtaining the intermediate signal after removing the linear interference. Based on the intermediate signal, the nonlinear interference is modeled and analyzed using the second-order Volterra series model, and the compensation value of the nonlinear interference is calculated to obtain the compensated signal data. If residual interference components are detected to exceed the preset threshold range for the compensated signal data, secondary processing is performed by iteratively adjusting the compensation parameters to obtain the decoupled ECG signal and IMU signal. The waveform features of the decoupled ECG and IMU signals are extracted using time-domain analysis methods to obtain signal changes at key time points and determine whether the signal quality meets the conditions for subsequent analysis.

[0037] Specifically, in S103, during the decoupling process of the coupling and interference model of the ECG signal and IMU signal in the signal set, the ECG signal and IMU signal are acquired through the signal acquisition device. Assuming that the sampling frequency of the ECG signal is 250Hz, the sampling frequency of the IMU signal is 100Hz, and the acquisition time is 10 seconds, 2500 ECG signal data points and 1000 IMU signal data points are obtained. For unified processing, the IMU signal is upsampled to 250Hz through linear interpolation to obtain 2500 data points. A linear mutual interference model is constructed. Assuming that the ECG signal is linearly interfered with by the IMU signal, a linear coupling interference model between the ECG signal and the IMU signal is established, where a is the linear coupling coefficient. The coupling coefficient 'a' is estimated using the RLS adaptive algorithm, and 'a' is calculated. est The estimated value, for example, at the 1000th sampling point, a est The convergence to 0.47 and the reduction of the sum of squared errors to 0.002 indicate that the estimation accuracy is high. Using the estimated a est A linear decoupling operation is performed to obtain a preliminarily decoupled ECG signal. At this point, the linear interference component is effectively weakened, assuming that the signal-to-noise ratio is increased from 15dB to 20dB. To further address nonlinear interference, a second-order Volterra series model is introduced, where b1 and b2 are estimated using the least squares method. Assuming that b1 = 0.03 and b2 = 0.001 are obtained, the nonlinear residual is reduced to 0.0005 after compensation, indicating that the nonlinear interference is significantly suppressed. By combining linear decoupling and nonlinear compensation, the decoupled ECG and IMU signals were obtained. The peak waveform error was reduced from the original 0.1mV to 0.02mV, and the heart rate calculation error was reduced from 5bpm to 1bpm, verifying the reliability of the decoupling effect.

[0038] Specifically, in S103, a linear coupling interference model between the ECG signal and the IMU signal is established:

[0039] Where, x obs For observing ECG signals with coupling interference; x true y represents the interference-free real ECG signal; a is the linear coupling coefficient between the ECG signal and the IMU signal, which is dimensionless; y is the IMU signal. In adaptive estimation of linear coupling coefficients using the RLS algorithm:

[0040] Where e is the coupling coefficient estimation error; x obsFor observing ECG signals with coupling interference; a est The linear coupling coefficient is estimated and is dimensionless; y is the IMU signal; K is the RLS algorithm gain coefficient. In eliminating linear coupling interference between ECG and IMU signals:

[0041] Where, x lin This is to eliminate the ECG signal after linear coupling; the meanings of the remaining characters are the same as above. In compensating for nonlinear coupling disturbances using a second-order Volterra model:

[0042] Where b1 is the first-order nonlinear coupling coefficient, dimensionless; b2 is the second-order nonlinear coupling coefficient, dimensionless; y is the IMU signal, unit: m / s²; x non For nonlinear coupling interference components; x dec To eliminate the decoupled ECG signal after linear + nonlinear coupling.

[0043] Furthermore, S104 includes: The decoupled ECG signal is subjected to power frequency interference suppression through an adaptive notch filter to obtain intermediate signal data. The intermediate signal data is processed using the EEMD algorithm and an adaptive soft threshold to suppress electromyographic interference and obtain the transition signal. The transition signal is processed using the VMD algorithm and IMU motion reference to suppress motion artifacts and baseline drift, resulting in purified ECG signal data.

[0044] In a specific embodiment, various types of interference in the ECG signal are initially identified, and the specific distribution range of power frequency interference, electromyographic interference, and motion artifacts is determined by comparing them with a pre-established interference type library, thus obtaining a classified list of interferences. Based on the classified interference list and combined with motion intensity data, a graded processing strategy is implemented for ECG signals. An adaptive notch filter is used to suppress power frequency interference in a targeted manner, and intermediate signal data after removing power frequency interference is obtained. For the intermediate signal data after removing power frequency interference, the Empirical Mode Decomposition (EMD) algorithm is used to separate the electromyographic interference. The separated interference components are filtered out by an adaptive soft threshold to obtain the transition signal with the electromyographic interference removed. Based on the transition signal after removing electromyographic interference, and combined with the IMU signal as a reference, the motion artifacts are decomposed by the variational mode decomposition algorithm to obtain the separated motion-related components. For the separated motion-related components, combined with the distribution characteristics of baseline drift, motion artifacts and baseline drift are jointly suppressed by signal reconstruction methods to obtain a preliminarily purified ECG signal; Based on the initially purified ECG signal, if residual interference components are detected to exceed the preset threshold range, the residual interference is filtered out a second time through signal smoothing processing to determine the purified ECG signal data. For the purified ECG signal data, key points of the signal waveform are marked using time-domain feature extraction methods to obtain the time nodes of waveform changes and determine whether the signal meets the conditions for subsequent processing.

[0045] Specifically, in S104, a three-level suppression is performed on the decoupled ECG signal. The first level is to suppress power frequency interference through an adaptive notch filter. The specific method is to use a notch filter with a center frequency of 50Hz and a bandwidth of 1Hz, using an IIR structure. The filter parameters are adaptively adjusted to match the actual interference frequency of the input signal. The analysis process is to perform a spectrum analysis on the input ECG signal and find a significant peak at 50Hz. After confirming the power frequency interference, the filter adjusts the Q value to 30 in real time to ensure that the suppression depth reaches more than 40dB, and the 50Hz component of the output signal is significantly weakened. To address electromyographic interference, a second classification method was adopted, and the EEMD algorithm and adaptive soft thresholding method were optimized to decompose the ECG signal into multiple IMF components. The number of decomposition layers was set to 8, and the noise standard deviation was 0.2. High-frequency IMFs (IMF1 and IMF2) were identified as the main components of electromyographic interference through spectrum analysis. Soft thresholding was applied to these IMFs, with the threshold dynamically adjusted to 1.5 times the signal standard deviation. Interfering IMFs were removed during signal reconstruction, reducing electromyographic noise energy by approximately 60%. To address motion artifacts and baseline drift, a third-level approach was adopted. The VMD algorithm was combined with IMU motion reference, with the number of decomposed modes set to 5 and the penalty factor α set to 2000. The peak motion acceleration (2.5 m / s²) was obtained from IMU data, and its correlation with ECG baseline drift was found to be 0.8. Low-frequency modes were extracted as drift components and removed. Simultaneously, motion artifact-related modes were smoothed with a window length of 50 ms to obtain purified ECG signal data, resulting in an improvement in signal-to-noise ratio of approximately 15 dB. By forming a logical chain through spectrum analysis, parameter optimization, and data fusion, the signal is gradually purified, and the output of each stage can be used as input for subsequent heart rate variability analysis, thus possessing business relevance.

[0046] Specifically, in S104, the adaptive notch filter suppresses 50Hz power frequency interference:

[0047] wherein, H(z) is the transfer function of the notch filter; z 1 represents a delay of 1 sampling period, which is dimensionless; z 2 represents a delay of 2 sampling periods, which is dimensionless; r is the radius of the filter pole, dimensionless, and 0<r<1 (the typical value is 0.99); ω0 is the angular frequency corresponding to 50Hz power frequency; f s is the sampling rate of ECG signals; In the suppression of electromyographic interference by using optimized EEMD combined with adaptive soft thresholding:

[0048] wherein, x is the input ECG signal; IMF i is the i-th intrinsic mode function; r i is the decomposition residual component; is the IMF component after soft threshold processing; sgn() is the sign function, which is dimensionless; T is the soft threshold; λ is the threshold adjustment coefficient; σ i is the standard deviation of the i-th IMF component; In the suppression of motion artifacts and baseline drift through VMD combined with IMU reference:

[0049] wherein, x is the input ECG signal; u k is the k-th modal component after VMD decomposition; u1 is the low-frequency baseline drift modal component; x clean is the purified ECG signal after baseline drift removal.

[0050] Further, S105 comprises: annotating the IMU signals in the signal set to obtain annotated interference distribution data; analyzing the annotated interference distribution data, determining the specific time period and amplitude range of the interference, and acquiring positioned interference feature data; simultaneously performing synchronous comparison on the ECG signals and the IMU signals in the signal set to acquire a signal matching result; performing layered processing on the signal matching result to acquire filtering parameter data.

[0051] In a specific embodiment, real-time motion data is acquired through IMU signals, and motion artifacts are preliminarily annotated in combination with an ECG signal processing flow, so as to obtain annotated interference distribution data; according to the annotated interference distribution data, the precise positioning of motion artifacts is analyzed by a signal comparison method, the specific time period and amplitude range of the interference are determined, and positioned interference feature data are acquired; For the interference feature data after localization, the IMU signal and ECG signal are synchronously compared through a two-way calibration mechanism to obtain the calibrated signal matching result; Based on the calibrated signal matching results, the filter parameter configuration is dynamically adjusted, and the motion intensity and interference level are processed in layers to determine the adjusted filter parameter data. By adjusting the filtering parameters, motion artifacts in the ECG signal are filtered out, and the filtered intermediate signal data is obtained. If interference exceeding the preset threshold is still detected in the intermediate signal data, the residual interference is processed again using a signal smoothing method to obtain the processed ECG signal data.

[0052] Specifically, in S105, when processing the IMU signal, it is fed back to the ECG signal processing link to optimize the motion artifact recognition and suppression effect. The specific implementation method during the bidirectional closed-loop calibration process is as follows: Triaxial acceleration data was collected using an IMU sensor with a sampling frequency of 100Hz. The data was smoothed using a Kalman filter, and motion characteristic parameters were extracted, such as an acceleration variance of 1.2 m / s². Combined with the time-domain fluctuation analysis of the ECG signal, the correlation coefficient between the two was calculated to be 0.75. After confirming the influence of motion artifacts, IMU data is used as a reference input to the ECG processing algorithm. An adaptive filtering structure is adopted, and the filter bandwidth is dynamically adjusted to the range of 0.5Hz to 2Hz. Motion-related abnormal fluctuations in the ECG signal are calibrated in real time, forming a closed-loop feedback mechanism to ensure the accuracy of signal processing. The parameters of each filtering module are adaptively adjusted based on the motion intensity level and interference intensity, and hierarchical power consumption management is performed. The specific method is as follows: Motion intensity was analyzed using IMU data and categorized into three levels: low, medium, and high, corresponding to average acceleration values ​​of 0.5 m / s², 1.5 m / s², and 3 m / s², respectively. Simultaneously, interference intensity was assessed on the ECG signal. When the noise variance was calculated to be 0.3, the computational complexity of the filtering module was automatically adjusted. For example, under low-intensity motion, the sampling rate was reduced to 50 Hz, the filter order was reduced to 2nd order, and power consumption was reduced by approximately 30%. Under high-intensity motion, the sampling rate is increased to 200Hz, the filter order is increased to 4th order to ensure processing accuracy, and the power consumption allocation is increased to 1.5 times the base value. The analysis process forms a dynamic balancing strategy by monitoring the system resource utilization and signal quality indicators in real time. Logically, it is related to the subsequent signal feature extraction business to ensure the stable operation of the system in different scenarios.

[0053] Specifically, in S105, the correlation between ECG and acceleration signals is calculated to locate motion artifacts:

[0054] Where x is the ECG signal; a is the IMU acceleration signal; Cov(x,a) is the covariance between the ECG signal and the acceleration signal; σ x σ is the standard deviation of the ECG signal. a ρ is the standard deviation of the acceleration signal. xa is the correlation coefficient between the ECG signal and motion artifacts, dimensionless, range [-1,1]; In adaptively adjusting the filter bandwidth based on motion intensity:

[0055] Where B is the real-time bandwidth of the filter; B0 is the base bandwidth of the filter; H is the bandwidth adjustment coefficient; a rms This is the root mean square value of acceleration.

[0056] Furthermore, S106 includes: Signal component data are extracted from the ECG and IMU signal data in the filter parameter data using signal decomposition techniques. Calculate the signal-to-noise ratio (SNR) of the signal component data to obtain the noise percentage data; The noise proportion data is labeled to obtain the labeled component state data. The integrity of feature points is determined by analyzing the labeled component state data to obtain the feature fidelity evaluation result data. Based on the feature fidelity evaluation results, the integrity of feature points is determined, and comprehensive evaluation status data is obtained. Then, the signal-to-noise ratio and feature fidelity are determined to obtain high-fidelity ECG signal data and IMU signal data.

[0057] In a specific embodiment, the processed ECG signal data and IMU signal data in the filter parameter data are obtained, and the main components are extracted by signal decomposition technology to obtain the decomposed signal component data. For the decomposed signal component data, frequency domain analysis is used to calculate the signal-to-noise ratio and determine the noise ratio in the signal. Based on the noise ratio data in the signal, compare it with the preset signal-to-noise ratio threshold. If the noise ratio data exceeds the threshold range, it is marked as an unqualified signal component, and the marked component status data is obtained. For the labeled component state data, the evaluation parameters of feature fidelity are obtained, and the integrity of feature points is determined by time-domain waveform comparison method to obtain the evaluation result data of feature fidelity. Based on the feature fidelity evaluation results, the data is compared with the preset fidelity threshold. If the feature point integrity meets the threshold requirements, it is marked as a qualified signal component, and the comprehensive evaluation status data is obtained. By comprehensively evaluating the state data, it is determined whether the signal-to-noise ratio and feature fidelity both meet the output conditions. If both are met, the signal component data are integrated to obtain high-fidelity ECG signal data and IMU signal data. Based on high-fidelity ECG and IMU signal data, various quality detection indicators are recorded and stored in the signal archive database to obtain archived signal record data.

[0058] Specifically, in S106, the process of evaluating the quality of the processed ECG and IMU signal data and outputting high-fidelity ECG and IMU signal data is implemented as follows: The ECG and IMU signal data were analyzed using signal processing algorithms to extract the signal-to-noise ratio (SNR). Assuming the calculated SNR value is 25.5dB, the frequency domain analysis method was used to decompose the signal using Fast Fourier Transform (FFT). The energy proportion of the main frequency component in the range of 0.8Hz to 3.5Hz was analyzed, and the proportion was found to be 82.3%, which was used as a preliminary evaluation index of feature fidelity. The calculated SNR value is compared with the preset threshold of 20dB, and the feature fidelity ratio is compared with the preset threshold of 80%. If both conditions are met, proceed to the next step. If the requirements are not met, the signal is further denoised using a wavelet transform algorithm. The Daubechies wavelet basis is selected, the decomposition level is set to 5 levels, and soft thresholding is applied to the high-frequency noise coefficient with a threshold of 0.15. The SNR value and feature fidelity ratio are recalculated until the threshold requirements are met. If the signal quality assessment passes, the system automatically marks the processed ECG and IMU signals as high-fidelity signals and outputs them to the subsequent heart rate variability analysis module to analyze the standard deviation of the heart rate interval time series. The assumed value is 45.2ms, which serves as a logical association with cardiovascular health assessment services to ensure the reliability of the signal in subsequent applications.

[0059] Specifically, in S106, the signal-to-noise ratio of the purified ECG signal is calculated to evaluate the noise reduction effect:

[0060] Where SNR is the signal-to-noise ratio of the ECG signal; P sig Effective ECG signal power; P noi denoted as residual noise power; lg is the logarithm to base 10, dimensionless. In calculating the energy proportion of characteristic frequency bands in electrocardiograms and evaluating waveform fidelity:

[0061] Where F represents the electrocardiogram characteristic fidelity, which is dimensionless; P 0.8-3.5Hz Power in the 0.8-3.5Hz ECG characteristic frequency band; P total The total power of the ECG signal across the entire frequency band; pass / fail criteria: SNR≥20dB, F≥80%.

[0062] In one embodiment, the implementation has been ported and tested across all scenarios on the STM32L431 low-power MCU, and the core performance indicators are as follows: After mutual interference suppression, the cross-correlation coefficient between the ECG signal and the IMU signal decreased by more than 95%; ECG signal signal-to-noise ratio improvement ≥25dB, up to 40dB; fidelity of key features such as QRS wave and ST segment ≥98%; The measurement error of the IMU signal is reduced by ≥80%, and the drift during long-term acquisition is ≤0.5° / h; The algorithm has a single-frame computation time of ≤10ms and an end-to-end latency of ≤40ms, which fully meets the real-time requirements. Full-scenario test dataset: Complete waveform data of raw and processed signals have been collected under different sports scenarios such as resting, walking, jogging, running, and cycling, including test cases with different interference intensities.

[0063] Example 2 like Figure 2 As shown, an ECG multi-source interference classification, identification, and hierarchical joint suppression system includes: Data acquisition module 1 preprocesses the synchronously acquired ECG and IMU signals to obtain a preprocessed signal set; Feature extraction module 2 extracts features from the signal set and identifies the type of interference; The interference intensity level is obtained based on the type of interference, and the motion intensity level is classified according to the interference intensity level. Decoupling operation module 3 eliminates linear mutual interference components of the signal set and compensates for nonlinear mutual interference through decoupling operation, to obtain the decoupled ECG signal and IMU signal; Suppression processing module 4 performs multi-level hierarchical suppression processing on the decoupled ECG signal to obtain purified ECG signal data; The purification module 5 performs motion artifact suppression and annotation on the IMU signals in the signal set to obtain interference distribution data. It then adjusts the parameters of the purified ECG signal data and the interference distribution data to obtain filter parameter data. Quality assessment module 6 performs quality assessment on the filter parameter data to obtain high-fidelity ECG signal data and IMU signal data.

[0064] Example 3 like Figure 3As shown, this embodiment provides a device 30, which includes a processor 300 and a memory 301; Memory 301 is used to store program code 302 and transfer program code 302 to the processor; The processor 300 is used to execute the steps in the above-mentioned ECG multi-source interference classification, identification and hierarchical joint suppression method according to the instructions in the program code 302.

[0065] For example, computer program 302 may be divided into one or more modules / units; one or more modules / units are stored in memory 301 and executed by processor 300 to complete this application; one or more modules / units may be a series of computer program instruction segments capable of performing specific functions, which are used to describe the execution process of computer program 302 in terminal device 30.

[0066] Terminal device 30 may be a computing device such as a desktop computer, laptop, handheld computer, or cloud server; terminal device may include, but is not limited to, processor 300 and memory 301.

[0067] Those skilled in the art will understand that Figure 3 This is merely an example of terminal device 30 and does not constitute a limitation on terminal device 30. It may include more or fewer components than shown, or combine certain components, or different components. For example, terminal device may also include input / output devices, network access devices, buses, etc.

[0068] The processor 300 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0069] The memory 301 can be an internal storage unit of the terminal device 30, such as a hard disk or memory of the terminal device 30; the memory 301 can also be an external storage device of the terminal device 30, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. equipped on the terminal device 30.

[0070] Furthermore, the memory 301 may include both internal storage units and external storage devices of the terminal device 30; the memory 301 is used to store computer programs and other programs and data required by the terminal device, and the memory 301 may also be used to temporarily store data that has been output or will be output.

[0071] Example 4 This embodiment provides a computer-readable storage medium storing a computer program; when the computer program is executed by a processor, it implements all the steps of the ECG multi-source interference classification, identification and hierarchical joint suppression method as described in Embodiment 1. Computer-readable storage media include, but are not limited to: ROM, RAM, disk, optical disc, USB flash drive, solid-state drive, cloud storage media; the program can be run by being loaded into any electronic device.

[0072] The above description is merely a preferred embodiment of one or more embodiments of this specification and is not intended to limit the scope of one or more embodiments of this specification. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments of this specification should be included within the protection scope of one or more embodiments of this specification.

Claims

1. A method for joint classification, identification, and hierarchical suppression of ECG multi-source interference, characterized in that, include: The synchronously acquired ECG and IMU signals are preprocessed to obtain a preprocessed signal set; Feature set extraction and interference type identification are performed on the signal set; The interference intensity level is obtained based on the interference type, and the motion intensity level is classified based on the interference intensity level. The linear mutual interference components of the signal set are eliminated and the nonlinear mutual interference is compensated by decoupling operation to obtain the decoupled ECG signal and IMU signal. The decoupled ECG signal is subjected to multi-level hierarchical suppression processing to obtain purified ECG signal data; Motion artifact suppression and annotation are performed on the IMU signals in the signal set to obtain interference distribution data; The purified ECG signal data and the interference distribution data are adjusted by parameter configuration to obtain filter parameter data; The quality of the filtered parameter data is evaluated to obtain high-fidelity ECG signal data and IMU signal data.

2. The ECG multi-source interference classification, identification, and hierarchical joint suppression method according to claim 1, characterized in that, The preprocessing of the synchronously acquired ECG and IMU signals to obtain a preprocessed signal set includes: The ECG signal and the IMU signal are read to obtain the original signal dataset; The original signal dataset is filtered to obtain a first signal set; The sampling rate of the first signal set is unified to obtain the second signal set; The third signal set is obtained by detecting and eliminating signals from the second signal set. The third signal set is timestamped to obtain the fourth signal set; The fourth signal set is smoothed to obtain a preprocessed signal set.

3. The ECG multi-source interference classification, identification, and hierarchical joint suppression method according to claim 1, characterized in that, The step of extracting features from the signal set, identifying the interference type, obtaining the interference intensity level, and classifying the motion intensity level according to the interference intensity level includes: The interference type is determined from the correlation values ​​in the signal set, and an interference classification label is obtained; The interference classification labels are weighted and quantitative indicators are calculated to obtain the interference intensity level; The frequency distribution of the IMU signal in the interference intensity level is detected to obtain the motion intensity level.

4. The ECG multi-source interference classification, identification, and hierarchical joint suppression method according to claim 1, characterized in that, The process of eliminating linear mutual interference components and compensating for nonlinear mutual interference in the signal set through decoupling operations to obtain decoupled ECG and IMU signals includes: Adaptive calculations are performed on the signal set to obtain the distribution parameters of linear interference; The distributed parameters are separated to obtain the intermediate signal after removing linear interference. The intermediate signal is combined with a second-order Volterra series model to analyze the nonlinear interference and calculate the compensation value of the nonlinear interference to obtain the compensated signal data. The compensated signal data is then detected to obtain the decoupled ECG signal and IMU signal.

5. The ECG multi-source interference classification, identification, and hierarchical joint suppression method according to claim 1, characterized in that, The step of performing multi-level suppression processing on the decoupled ECG signal to obtain purified ECG signal data includes: The decoupled ECG signal is subjected to power frequency interference suppression through an adaptive notch filter to obtain intermediate signal data; The intermediate signal data is processed using the EEMD algorithm and an adaptive soft threshold to suppress electromyographic interference and obtain a transition signal. The transition signal is processed using the VMD algorithm and IMU motion reference to suppress motion artifacts and baseline drift, resulting in purified ECG signal data.

6. The ECG multi-source interference classification, identification, and hierarchical joint suppression method according to claim 1, characterized in that, The process involves suppressing motion artifacts and labeling the IMU signals in the signal set to obtain interference distribution data. Then, the purified ECG signal data and the interference distribution data are compared and their parameters are adjusted to obtain filter parameter data, including: The IMU signals in the signal set are labeled to obtain labeled interference distribution data; The labeled interference distribution data is analyzed to determine the specific time period and amplitude range of the interference, and the interference feature data after positioning is obtained. Simultaneously, the ECG signal and the IMU signal in the signal set are compared synchronously to obtain the signal matching result; The signal matching results are processed in layers to obtain filter parameter data.

7. The ECG multi-source interference classification, identification, and hierarchical joint suppression method according to claim 1, characterized in that, The process of evaluating the quality of the filtered parameter data to obtain high-fidelity ECG signal data and IMU signal data includes: The ECG signal data and the IMU signal data in the filter parameter data are extracted to obtain signal component data through signal decomposition technology; Calculate the signal-to-noise ratio (SNR) value for the signal component data to obtain the noise percentage data; The noise proportion data is labeled to obtain labeled component state data. The integrity of feature points is determined from the marked component state data to obtain feature fidelity evaluation result data. Based on the evaluation results of the feature fidelity, the integrity of the feature points is determined, the state data of the comprehensive evaluation is obtained, and then the signal-to-noise ratio and the feature fidelity are determined to obtain high-fidelity ECG signal data and IMU signal data.

8. A system for classifying, identifying, and hierarchically suppressing ECG multi-source interference, characterized in that, include: The data acquisition module preprocesses the synchronously acquired ECG and IMU signals to obtain a preprocessed signal set. The feature extraction module extracts features from the signal set and identifies the type of interference. The interference intensity level is obtained based on the interference type, and the motion intensity level is classified based on the interference intensity level. The decoupling operation module eliminates linear mutual interference components and compensates for nonlinear mutual interference in the signal set through decoupling operations, thereby obtaining the decoupled ECG signal and IMU signal. The suppression processing module performs multi-level hierarchical suppression processing on the decoupled ECG signal to obtain purified ECG signal data. The purification module performs motion artifact suppression and annotation on the IMU signals in the signal set to obtain interference distribution data. It then adjusts the parameters of the purified ECG signal data and the interference distribution data to obtain filter parameter data. The quality assessment module performs a quality assessment on the filtered parameter data to obtain high-fidelity ECG signal data and IMU signal data.

9. An electronic device, characterized in that, The system includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor causes the processor to perform the steps of the ECG multi-source interference classification, identification and hierarchical joint suppression method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the ECG multi-source interference classification, identification and hierarchical joint suppression method as described in any one of claims 1 to 7.

Images