CN115944303B - Method, system and storage medium for online compression of electrocardiogram pulse signal - Google Patents

Abstract

The invention discloses an on-line electrocardio pulse signal compression method, which comprises the following steps: the method comprises the steps of collecting electrocardiosignals and pulse signals, performing analog-to-digital conversion and real-time filtering processing on the signals, firstly improving the existing adaptive filter, and adding a high-frequency noise estimation part; then, combining the data updating process of the cache region of the microprocessor system with the self-adaptive filtering principle, and calculating in a sliding window iteration mode to realize real-time filtering of the pulse signals of the first points; key points of the electrocardio pulse signals are extracted, compressed, encoded and transmitted on line; and after the upper computer decodes the received data, recovering the electrocardio pulse signals by adopting a cubic spline interpolation method, and performing performance evaluation by utilizing the minimum mean square error and the compression ratio. The compression ratio is effectively improved, and the compressed data can be applied to heart rate and pulse rate extraction of the electrocardio pulse signals. The intelligent wearable device with high real-time requirements can be used for intelligent wearable devices such as a bracelet and a watch.

Description

Method, system and storage medium for online compression of electrocardiogram pulse signal

Technical Field

The present invention relates to the field of signal processing and compression, and in particular to an online compression method, system and storage medium for an electrocardiogram pulse signal.

Background technique

The ECG pulse signal contains rich physiological and pathological information of the human body, which is used to monitor clinical physiological parameters such as blood oxygen saturation, heart rate, heart rate variability, and blood pressure. At present, some wearable devices, such as smart watches and bracelets, extract pulse rhythm from pulse signals for the assessment of human health in daily life. Long-term data collection will generate massive amounts of data. For wearable devices with limited storage capacity and processing speed, a large amount of data brings a heavy burden to the system's transmission, storage, and processing. Therefore, at present, considering the limited storage space of wearable devices, CPU processing speed and data transmission volume, battery capacity and other factors, wearable devices can only calculate and transmit pulse rate or blood oxygen saturation, but not store the original ECG pulse signal for a long time. However, there is still a lot of information about human health in the original ECG pulse signal. Therefore, it is very important to cache and transmit the original ECG pulse signal.

At present, there are three types of biomedical data compression technologies: direct compression, transformation-based compression, and parameter extraction-based compression. Direct compression takes a long time to compress data and has a low compression ratio; transformation-based compression often requires complex calculations, and the data compressed by direct compression and transformation-based compression cannot be processed directly and must be processed after decompression and recovery; parameter extraction-based compression technology is easily affected by noise, and the choice of parameters has a great influence on the compression results.

Summary of the invention

The embodiments of the present application provide an online compression method, system and storage medium for ECG pulse signals, thereby solving the technical problem in the prior art that it is difficult to achieve both a high compression ratio and accurate compression results for ECG pulse signal data of wearable devices. This ensures that after the ECG pulse signal is processed with a high compression ratio, its compression result is less affected by the compression and can still be used accurately.

An online compression method for an electrocardiogram pulse signal comprises the following steps:

S01: Collecting ECG signals and pulse signals, performing analog-to-digital conversion on the signals, and obtaining ECG pulse signals on the human body surface;

S02: Design an adaptive filter, improve the existing adaptive filter, and add a high-frequency noise estimation part;

S03: Online adaptive filtering, combining the data update process in the microprocessor system cache with the adaptive filtering principle, and using a sliding window iteration method to perform calculations to achieve real-time filtering of the ECG pulse signal;

S04: Compression and encoding of ECG pulse signal, the extreme points and inflection points in the ECG pulse signal are collectively referred to as key points, the key points of the ECG pulse signal are extracted online to achieve compression of the ECG pulse signal; the compression process of the ECG pulse signal is as follows: determine the extreme value according to the definition of the extreme point, calculate the straight line between two adjacent extreme points by Lagrange's middle value theorem, subtract the ECG pulse signal from the straight line to obtain the difference curve, extract the extreme points of the difference curve, and achieve compression of the ECG pulse signal; encode the compressed ECG pulse signal, the ECG pulse signal encoding process is as follows: encode the position and amplitude corresponding to the key parameters of the ECG pulse signal according to the transmission rules of the microprocessor to ensure that the encoded data length is an integer multiple of 1 byte. If the length of a data point is not an integer multiple of 1 byte, fill the high bit of the data with 0 until the condition is met. For a microprocessor with a voltage resolution of 10 bits, the position and amplitude corresponding to each key parameter after encoding occupy 3 bytes, and the encoded data is sent in the order of position first and amplitude later;

S05: decoding and recovery, decoding the compressed ECG pulse signal, the ECG pulse signal decoding process is as follows: extracting and dividing the received data according to the position of the packet header and the packet tail, each 3 bytes of data corresponds to a key parameter, wherein the first byte corresponds to the position of the key parameter, the second byte corresponds to the high 8 bits of the key parameter amplitude, and the third byte corresponds to the low 8 bits of the key parameter amplitude; recovering the amplitude of the key parameter according to the high 8 bits and the low 8 bits of information to achieve decoding of the ECG pulse signal; recovering the ECG pulse signal: using a cubic spline interpolation method to interpolate the decoded ECG pulse signal key parameters to achieve recovery of the ECG pulse signal;

S06: Evaluate the performance of the algorithm and apply the algorithm. Use root mean square error and compression ratio to evaluate the performance of the algorithm. At the same time, based on the extracted key points, calculate the pulse main wave peak and the pulse wave main wave interval, and calculate the heart rate.

Preferably, high-frequency noise estimation is performed on the signal in step S02, including: based on the characteristic that mean filtering can effectively filter out high-frequency noise in the signal, mean filtering is combined with a microprocessor data update process; when the buffer length is M, the maximum and minimum values of the current buffer are removed, and the mean of the remaining M-2 data is calculated to complete the high-frequency noise estimation in the signal.

Preferably, step S03 is online adaptive filtering, which combines the data update process of the microprocessor system cache with the principle of adaptive filtering, and uses a sliding window iteration method to perform calculations to achieve real-time filtering of the ECG pulse signal, specifically:

Based on sliding window iteration, the real-time filtering of the data is converted into filtering of the entire current buffer area; when a new sampling point slides into the data buffer area, the data in the buffer area is updated, and the data processing of the new sampling point is completed based on the original M-1 sampling points.

Preferably, the online adaptive filtering specifically includes an initialization process and an iteration process:

Initialization process: When the current data sampling point is less than or equal to the length of the data buffer (0≤n≤M-1), each latest sampling point is placed in the nth byte of the buffer; then, the sampling values in the data buffer are arranged in ascending order, and the mean is calculated to estimate the high-frequency noise of the data; the high-frequency noise estimation result is recorded as [0,…,x ₀ ,…,x _n ]; the output after filtering out the high-frequency noise is used as the reference input x _n of the adaptive filter, and the data is further subjected to online adaptive filtering operation, and the result is recorded as [0,…,y ₀ ,…,y _n ]; then, the filter coefficients are updated until the sampling point n=M-1, completing the initialization process;

Iterative process: When the current data sampling point is greater than the length of the data buffer (n>M), the data in the buffer are shifted left by one bit, and each latest sampling point is placed in the Mth byte of the buffer; then, the sampling values in the data buffer are arranged in ascending order, and the mean is calculated to estimate the high-frequency noise of the data; the high-frequency noise estimation result is recorded as [ _xn (n-M+1),…, _xn (M),…, _xn (n)]; the output after filtering out the high-frequency noise is used as the reference input _xn of the adaptive filter, and the data is further subjected to online adaptive filtering operation, and the result is recorded as [ _yn (n-M+1),…, _yn (M),…, _yn (n)]; then, the filter coefficients are updated until the sampling is completed, and the iterative process is completed.

Preferably, the compression of the ECG pulse signal in step S04 is achieved by extracting key points in the signal:

The key points in the signal are composed of extreme points and inflection points. According to the definition of extreme points, at least three points are used to judge the extreme values. The current data buffer is recorded as buffer. If buffer(M-2) is greater than buffer(M-1) and buffer(M-3), then buffer(M-2) is the maximum value. If buffer(M-2) is less than buffer(M-1) and buffer(M-3), then buffer(M-2) is the minimum value. Two arrays E and L are set to store the amplitude and position corresponding to the extreme points respectively. Buffer(j) is a known extreme point. , the amplitude and position of buffer(j) are stored in E(3) and L(3) respectively; in the process of continuous data acquisition and processing, if buffer(M-2) is not an extreme point, the amplitude E(3) of the previous extreme point buffer(j) remains unchanged, and the position L(3) decreases by 1; if buffer(M-2) is an extreme point, the amplitude E(3) of the previous extreme point buffer(j) is shifted left by three bits and stored in E(0), and the position L(3) is shifted left by three bits and stored in L(0), and the amplitude and position of the current extreme point buffer(M-2) are stored in E(3) and L(3);

The process of extracting the inflection point is as follows: after the extreme point is determined, the straight line between two adjacent extreme points is calculated using the Lagrange middle value theorem, the ECG pulse signal is subtracted from the straight line to obtain a difference curve, and the extreme point of the interpolation curve is calculated, which is the inflection point of the ECG pulse signal. Its corresponding parameters are stored in E(1), E(2) and L(1), L(2).

Preferably, the microprocessor transmits and sends data in integer multiples of 1 byte. During the encoding process, it is necessary to ensure that the encoded data is an integer multiple of 1 byte. If the length of a data is not an integer multiple of 1 byte, the high bit is padded with zeros until the condition is met; after the ECG pulse signal is filtered in the microprocessor, the data type becomes a floating point type, and the data type is forced to be converted into an integer type, and key parameters are further extracted from the data; for a microprocessor with an input voltage resolution of 10 bits, the position and amplitude of each key parameter after encoding occupy 3 bytes, among which the first byte corresponds to the position of the key parameter, and the last two bytes correspond to the amplitude of the key parameter.

Preferably, the microprocessor sends and receives the encoded key parameters in the order of position first and amplitude later, divides the received data according to the position of the packet header and the packet tail, and every 3 bytes correspond to the characteristic parameters of a key point; sets array D and array P to store the position and amplitude of the key point after decoding, wherein the position of the key parameter corresponding to the first byte is stored in D(M-1), the second byte corresponds to the high 8 bits of the key parameter amplitude, and the third byte corresponds to the low 8 bits of the key parameter amplitude; the amplitude of the key parameter is restored according to the high 8 bits and the low 8 bits of information, and the restored amplitude of the key parameter is stored in P(M-1).

An online compression system for electrocardiograph pulse signals comprises: a data acquisition module for collecting electrocardiograph signals and pulse signals; a filtering processing module for adaptively filtering signals; a compression module for compressing, encoding and sending signals; and a decompression module for receiving, storing and decoding signals.

Preferably, the filtering processing module includes: a high-frequency noise estimation module, an adaptive filtering algorithm module, and a filter module.

A computer storage medium stores a computer program, which realizes the above-mentioned online compression method of electrocardiogram pulse signal when executed.

One or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages:

1. It does not rely on any complex mathematical transformation and has a fast algorithm. The proposed algorithm is verified using ECG pulse signals and time is used as a measure of speed. The proposed method is expected to be used in smart wearable devices with high real-time requirements such as bracelets and watches.

2. According to the waveform characteristics of the ECG pulse signal, a compression method based on time domain characteristics is proposed; the compressed signal retains the key features (extreme values and inflection points) of the original signal, and more characteristic information can be derived from these features, such as pulse wave propagation time, pulse wave propagation speed, etc. These characteristic information includes key signal parameters such as heart rate, pulse rate, wave height, and area.

3. The data update process of the microprocessor system cache is combined with the principle of adaptive filtering, and the sliding window iteration method is used for calculation to realize real-time filtering of the ECG pulse signal. The characteristic parameters of the ECG pulse signal are extracted online, which improves the compression ratio, and the compressed data has little impact, within the acceptable error range between the restored data and the original data.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a flow chart of an online compression method for an electrocardiogram pulse signal of the present application;

FIG2 is a flowchart of the adaptive filtering of the present application;

FIG3 is a microprocessor data cache update process of the present application;

Figure 4 is a schematic diagram of the key points in the pulse signal

Fig. 5 is a schematic diagram of key point preservation;

FIG6 is a schematic diagram of an extreme value extraction process;

FIG7 is a flowchart of a key point extraction process;

FIG8 is a schematic diagram of an inflection point extraction process;

FIG9 is a schematic diagram of a key point parameter encoding process;

FIG10 is a schematic diagram of a key point parameter decoding process;

FIG11 is a schematic diagram showing the comparison of signal recovery under different interpolation methods;

FIG12 is a waveform comparison of the pulse signal and the ECG signal before compression and after reconstruction;

FIG13 is a flow chart of a dynamic threshold method for extracting peak values;

Figure 14 is a comparison of pulse rates before and after compression;

Figure 15 is the real-time pulse rate obtained in the hardware system;

Figure 16 shows the process of signal compression and decompression.

Detailed ways

The embodiments of the present application provide a method, system and storage medium for online compression of electrocardiogram pulse signals, which solve the technical problem in the prior art that it is difficult to achieve both high compression ratio and accurate compression results for electrocardiogram pulse signal data of wearable devices.

The technical solution in the embodiment of the present application is to solve the above-mentioned crosstalk problem, and the overall idea is as follows: improve the traditional adaptive filter and add a high-frequency noise estimation part; combine the data update process of the microprocessor system cache area with the adaptive filtering principle, and use a sliding window iteration method to perform calculations to achieve real-time filtering of the ECG pulse signal; extract the key points of the ECG pulse signal online to achieve compression and encoding of the ECG pulse signal; decode the key parameters and use the cubic spline interpolation method to restore the ECG pulse signal.

According to the characteristics of the adaptive filtering method, the noise in the signal is suppressed to effectively improve the compression ratio, and the compressed data can be used to extract the heart rate and pulse rate of the ECG pulse signal.

In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

As shown in Figure 16, the process of ECG pulse signal compression and recovery is divided into data sampling compression and data decompression. In the signal processing process, the first half works on the lower computer, including: signal sampling, adaptive filtering, signal compression, signal encoding and sending; the second half works on the upper computer, including: signal reception and storage, signal decompression and signal recovery.

As shown in FIG1 , an online compression method for an electrocardiogram pulse signal comprises the following steps:

S01: Collecting ECG signals and pulse signals, performing analog-to-digital conversion on the signals, and obtaining ECG pulse signals at the human body surface.

The ECG and pulse signals are captured by optical sensors, amplified and filtered in the lower computer module, and then sampled and compressed.

The pulse signal acquisition system was used to collect signals from 30 test subjects. Before the experiment, the consent of the test subjects was obtained in advance and a written consent form was signed to ensure the health of the test subjects. Under the supervision of the hospital ethics committee, the signal sampling of the 30 test subjects was carried out for 4 minutes at a sampling frequency of 1000 Hz.

Gaussian function is used to construct the pulse signal, which is synthesized by three Gaussian functions. Let a pulse wave be x(k). According to the characteristics of the pulse signal, the clean pulse wave is synthesized by three Gaussian functions, corresponding to the main wave, pre-dicrotic wave and dicrotic wave of the actual pulse wave, and its expression is:

Among them, _Ai determines the height of the Gaussian function image, _A1 = 0.8, _A2 = 0.5, _A3 = 0.4; _pi determines the image peak position, _p1 = 0.25, _p2 = 0.45, _p3 = 0.70; _bi determines the image width, _b1 = 0.012, _b2 = 0.010, _b3 = 0.030;

The electrocardiogram signal is constructed by using Gaussian function, and the electrocardiogram signal is synthesized by six Gaussian functions. Let a pulse wave be E(k) and the calculation formula is as follows:

Wherein, _Ai determines the height of the Gaussian function image, _A1 = 0.04, _A2 = -0.01, _A3 = 0.95, _A4 = -0.14, _A5 = 0.01, _A6 = 0.17; _μi is the expected value of the image, _μ1 = 68.6, _μ2 = 95.4, _μ3 = 100.5, _μ4 = 106.2, _μ5 = 126.9, _μ6 = 170.4; _σLi , _σ1i are the standard deviations of the image, _σL1 = 12, _σL2 = 0, _σL3 = 4.9, _σL4 = 7.3, _σL5 = 0, _σL6 = 0, _σ11 = 12.3, _σ12 = 337.1, _σ13 = 9.6, _σ14 = 2.2, _σ15 =42.9,σ ₁₆ =60.3;

S02: Filtering of ECG and pulse signals. First, the high-frequency noise of the signal is estimated, and then the noise in the signal is removed by an adaptive filtering algorithm and a filter.

As shown in FIG. 2 , the filtering method of the present application is composed of three parts: high-frequency noise estimation, a digital filter with adjustable coefficients, and an adaptive algorithm for correcting filter parameters.

The high-frequency noise of the actually collected noisy signal is estimated, which is expressed by formula (3):

Among them, _xn is the output after filtering out high-frequency noise, that is, the input end of the adaptive filter; buffer is the data buffer of the microprocessor and the data in the buffer are arranged in ascending order; M is the length of the data buffer.

The noise in the signal is removed by the adaptive filter, that is, the minimum value of the system output error _en is found. _En is the output error of the adaptive filter, and its expression is:

e _n = d _n - _yn (4)

Among them, d _n is the main input of the adaptive filter; y _n is the output of the adaptive filter.

The least mean square error (LMS) algorithm is an iterative method between the initial signal and the reference signal. It minimizes the mean square error through continuous iteration. The LMS algorithm updates and corrects the parameters at time n+1 through the coefficients of the filter at time n. The calculation formula is as follows:

w _n+1 = w _n +2ux _n s _n (5)

Among them, u represents the convergence factor of the adaptive filter, which controls the stability and convergence speed of the LMS algorithm; w _n is the coefficient of the filter corresponding to the sampling point, expressed as w _n =[w ₀ (n),w ₁ (n)…,w _M-1 (n)] ^T , M is the order of the filter; s _n is related to the noise in the input signal x(k).

The update of the weight vector w _n depends on the input data _sn . Therefore, if it is assumed that the successive input data are independent, that is, w _n and _sn are independent, the mathematical expectation of the weight vector of the LMS filtering algorithm can be obtained as follows:

Where _w0 is the optimal Wiener solution of the weight vector; R is the autocorrelation matrix E[x(n)x ^T (n) of the input vector; let _vn = E( _wn ) _-w0 , then equation (6) can be expressed as:

v _n+1 =(I-2uR)v _n (7)

The correlation matrix R can be decomposed into:

R＝QΛQ ^T (8)

Wherein, the matrix Q is a transformed unitary matrix, and each column vector thereof is a set of orthogonal eigenvectors related to the eigenvalues of the matrix R; the matrix Λ is a diagonal matrix, and all the eigenvalues of the matrix R are its diagonal elements, and these eigenvalues are all positive real numbers, denoted as [λ ₁ ,λ ₂ ,…,λ _n ]; multiplying equation (7) by Q ^T on the left yields:

Q ^T v _n+1 = (I-2uR)Q ^T v _n (9)

Let c _n =Q ^T v _n , the initial value of c _n is c ₀ , then equation (9) can be simplified to:

c _n =(I-2uΛ ⁿ )c ₀ (10)

Expanding equation (10) yields M independent scalar equations, namely:

In order to ensure the convergence of the coefficients in the average sense, it is necessary to satisfy The solution is that the step size factor of the LMS algorithm meets the convergence condition:

Where λ _max is the maximum eigenvalue of the correlation matrix R. When the number of iterations tends to infinity, the weight vector w _n of the filter approaches the Wiener solution w ₀ .

In the process of data acquisition, some noise is inevitably collected. The noise in the signal is mainly baseline drift, power frequency interference and electromyographic interference. By using the above method to suppress noise after signal acquisition and before compression, power frequency interference and electromyographic interference can be effectively avoided, that is, the interference of noise on extreme value information can be avoided.

S03: Using sliding window iteration to perform calculations to achieve real-time filtering of the pulse signal.

The data update process of the microprocessor system cache is combined with the principle of adaptive filtering, and the sliding window iteration method is used for calculation to achieve real-time filtering of the pulse signal. According to the relationship between the current data sampling point and the length of the data cache, the real-time filtering is divided into two steps: initialization process and iteration process.

As shown in Figure 3, assuming that the buffer can store M sampling points, each time a new sampling point is collected, data needs to be updated. First, remove the earliest cached (leftmost) sampling point data(0); then, move data(1) to data(M-1) one position to the left; finally, place the new sampling point data(M) at the rightmost end. If the data stream is fixed, the buffer is like a window that slides bit by bit in the data stream. This process is called sliding window. Each time the window slides, only the leftmost and rightmost values in the buffer change, and the other sampling points only change their positions while the values remain unchanged.

Initialization process:

When the number of sampling points is less than or equal to the length of the buffer (0≤n≤M-1), the latest sampling point is placed in the nth byte of the buffer. Then, the high-frequency noise is estimated according to formula (3), the filter coefficients are updated according to formula (5), and the filtered result is recorded as [0,…,y ₀ ,…,y _n ] until the sampling point n＝M-1, completing the initialization process.

Iteration process:

When the number of sampling points is greater than the length of the buffer (n>M-1), the iteration process begins. As in the initialization process, the data in the buffer and the intermediate variables are updated through the sliding window. After the data update is completed, the data in the data buffer is processed, and the calculation process is the same as the initialization process. The result after adaptive filtering is recorded as [y _n (n-M+1),…,y _n (M),…,y _n (n)]; until the sampling is completed, the iteration process is completed.

Based on sliding window iteration, the real-time filtering of data is converted into filtering of the entire current buffer area; when a new sampling point slides into the data buffer area, the data in the buffer area is updated, and the data processing of the new sampling point is completed on the basis of the original M-1 sampling points; according to the characteristic that mean filtering can effectively filter out high-frequency noise in the signal, mean filtering and microprocessor data update process are combined to remove the maximum and minimum values of the current buffer area, and the mean of the remaining M-2 data is calculated to complete the estimation of high-frequency noise in the signal.

S04: extract key points from the filtered signal, perform signal compression, encode and send in the microprocessing unit.

For common signals without discontinuities (such as random curves with inflection points and extreme points such as sine, cosine, triangle waves, etc.), in signal processing, the extreme values and inflection points in the signal play a very important role in further signal processing. It is very important to retain these features as much as possible when compressing the signal so that these features can be maintained after the signal is restored.

In a preferred implementation, the key points in step S04 include extreme value points.

As shown in Figure 4, the proposed compression algorithm retains the extreme value features in the signal. The extraction of extreme values is easily affected by power frequency interference and electromyographic interference. The designed adaptive filter is used to suppress the interference of these noises on the extreme value information. The extreme point is determined by the starting point A, the main wave peak B, the main wave trough C, the tidal wave peak D, the tidal wave trough E, the dicrotic wave peak and the end point A'. The amplitude information of the pulse wave in the time domain is represented by _Hi , _ti is the position information corresponding to _Hi , and PPI refers to the length of the pulse wave or the period of the pulse wave. Moreover, more characteristic information can be derived from these characteristic information, such as the main wave phase period ratio _t1 / _H1 , the propagation time of the pulse wave, the propagation speed of the pulse wave, etc. These derived characteristic information can be used to monitor blood pressure and other physiological indicators. Therefore, the extreme point is used as part of the compression result.

As shown in Figure 5, the position and amplitude information of the feature points are stored in the initialization array E and array L. From the definition of extreme points, we know that at least three points are needed to determine whether the sampling point is an extreme value. As shown in Figure 6, if y(N-2) is a minimum value, then:

y(N-2)≤y(N-1) (13)

y(N-2)≤y(N-3) (14)

If y(N-2) is a maximum value, then:

y(N-2)≥y(N-1) (15)

y(N-2)≥y(N-3) (16)

Where: N is the length of the data buffer.

As shown in Figure 7, the data is updated. y(j) is a known extreme point, and the amplitude and position of y(j) are stored in E(3) and L(3) respectively. After the latest sampling point filtering is completed, three sampling points are used to determine whether y(N-2) is an extreme point. If y(N-2) is an extreme point, E(3) and L(3) are shifted left by four bits, and the amplitude and position information of y(N-2) are stored in E(3) and L(3). In addition, the information of other feature points y(p) and y(q) in E(0) and L(3) can also be calculated. If y(N-2) is not an extreme point, array E remains unchanged and L(3) is reduced by 1. After the operation is completed, all data in the data buffer are shifted left by one bit to complete the update.

In a preferred implementation, the key point in step S04 includes an inflection point.

As shown in Figure 4, there are still some key points on the curve between adjacent extreme points that play an important role in signal recovery, such as point a and point b on curve AB, which are collectively referred to as inflection points. Therefore, inflection points are also part of the compression result.

As shown in Figure 8, a new difference curve is obtained by subtracting the straight line between points AB from curve AB, and points a and b are the peak and trough positions of the difference curve. Therefore, the peak and trough points of the difference curve are used as part of the compression result.

In a preferred implementation, the key points in step S04 are encoded and sent in the micro-processing unit.

The position and amplitude of the key parameters extracted from the signal are encoded and sent in the microprocessor unit (MCU). As we all know, MCU stores and sends data in one byte (8 bits). It should be noted that the length of the encoded data should be an integer multiple of 1 byte. If the length of the data is not an integer multiple of 1 byte, the high bit is filled with 0 until the condition is met. For example, En and Ln are the amplitude and position of the nth feature point. If the resolution of the input voltage is 10 bits and the sampling frequency of the signal is 200Hz, the information of the nth feature point needs to occupy 24 bits. Therefore, the parameters of a key point occupy 3 bytes. As shown in Figure 9, in order to avoid data loss during transmission, add a header and a tail to the encoded data, and send the encoded data to the host computer module wirelessly or wired.

S05: The host computer receives, stores, and decodes the ECG pulse signal and recovers the data sample.

Receiving, storing and decoding the compressed signal are completed in the host computer module. The microprocessor sends and receives the encoded key parameters in the order of position first and amplitude second, and divides the received data according to the position of the packet header and the packet tail. Every 3 bytes correspond to the characteristic parameters of a key point. The process is shown in Figure 10. Arrays P and D are used to store the amplitude and position of the decoded key points. For example, according to the 3-byte data received according to the voltage resolution, the first byte of data corresponds to the position of the key point, which is stored in D(M-1), and the last two bytes correspond to the high 8 bits and low 8 bits of the key parameter amplitude respectively. The second byte of data is multiplied by 256 and added to the third byte of data to obtain the amplitude of the key point parameter, which is stored in P(M-1).

In a preferred implementation, the data samples are restored in step S05 using a cubic spline interpolation method.

The cubic spline interpolation method is a method of shape-preserving piecewise cubic interpolation based on the values at the neighboring grid points at the interpolation value at the query point. As shown in Figure 11, the schematic diagram of the recovery results of curve AB using different interpolation methods based on the compression results shows that the cubic spline interpolation method produces the smallest error between the original signal and the reconstructed signal. Therefore, the cubic spline interpolation method can be used to recover the compressed signal. The compressed data is recovered by cubic spline interpolation and used for further analysis and processing.

S06: Evaluate performance using minimum mean square error and compression ratio.

In order to evaluate the performance of the algorithm, some indicators are cited, such as the minimum mean square error (MSE) and compression ratio (CR), which are calculated as follows:

Where: n is the index of the data, N is the length of the signal, R _n and S _n correspond to the original signal and the reconstructed signal respectively.

When CR is the same, the smaller the MSE, the better the performance of the method. Similarly, when MES is the same, the larger the CR, the better the performance of the method.

Two other compression methods, such as turning point compression (TP) algorithm and wavelet decomposition compression (WT) algorithm, are used to compare with the proposed algorithm to evaluate the performance of the algorithm. Among them, for TP compression, the first sampling point is retained, three points (y1, y2, y3) are selected from the first point, and the amplitude difference (y2-y1, y3-y2) of these three points is found. If (y2-y1)(y3-y2)<0. The sampling point y2 is a turning point, and y2 is retained. Otherwise, the last sampling point y3 is retained. The signal is restored by cubic spline interpolation. For WT compression, the sampled signal is separated into high-frequency and low-frequency components by the 'db4' wavelet. The low-frequency component is retained and used to restore the signal.

The proposed method and the above two methods are used to process ECG pulse signals, and the results are shown in Table 1. As can be seen from the table, the compression ratio of the proposed method is significantly better than that of the other two methods. In the pulse signal, the compression ratio of the proposed method can reach 14.571; in the ECG signal, the compression ratio of the proposed method can reach 52.786. Considering that the TP compression algorithm requires three reference points for comparison during the compression process, when the amount of data is too large, the real-time performance of the algorithm is seriously affected, and the WT compression algorithm requires complex mathematical calculations during the compression process, the amount of calculation also affects the real-time performance of the algorithm. Although the MSE of the proposed method is the largest among the three methods, the real-time performance of the proposed method is easier to achieve within the allowable error range.

Table 1 ECG pulse signal compression results

As shown in Figure 12, it is a comparison of the proposed method before and after compression in the ECG pulse signal. As shown in Figure 12 (a), the restored pulse signal is basically consistent with the original signal. Although the compression ratio of the proposed method in the ECG signal is significantly greater than that of the pulse signal, it can be observed in Figure 12 (b) that the reconstructed signal has a large deviation from the original signal at point A.

Furthermore, the proposed method was applied to the pulse signal collected by the self-developed acquisition system, and the statistical results are shown in Table 2. The statistical results of 30 sets of sampling data are expressed in the form of "mean ± standard deviation". As can be seen from the table, the compression ratios corresponding to the three methods are 2.81 ± 0.33, 10.15 ± 0.10, and 41.36 ± 7.49, respectively. The compression ratio of the proposed method is the largest, but according to the different physiological characteristics of each person, the signal processed each time is also different, which is reflected in the larger standard deviation. Relatively speaking, this method can achieve a significant result in practical applications.

Table 2 Measured pulse signal compression results

The compression method proposed in this embodiment includes the time domain feature points of the signal before compression, which can be used for time domain feature analysis of ECG pulse signals, such as heart rate and pulse rate. Heart rate is defined as the number of heart beats per minute, that is, the heart rate range of a normal adult is 60-100 beats/minute in a calm state. Some studies have shown that pulse rate can replace heart rate to assess human health. The pulse rate can be calculated by the difference between adjacent peaks, and its calculation formula is as follows:

Where: fs is the sampling frequency, PPI is the pulse wave main wave interval.

As shown in Figure 13, first, 0.3 times the sum of the maximum and average values of the read data is used as the final threshold. Then, the points in the data that are greater than the threshold are set to 1, and the points that are less than the threshold are set to 0 to obtain a new array X. Then, the original data read is multiplied by X to obtain a curve greater than the threshold. The array X is differentiated, and the point with a value of -1 is the starting point and end point of the curve greater than the threshold. The maximum between the two points is the peak point of the P wave.

Since the amplitude of the pulse signal is unstable, there will be many false detections of the peak points found in this way. Most of the falsely detected peak points are the peak points of the tidal wave, so the falsely detected peak points need to be removed. According to the properties of the pulse signal, it is found that the pulse width of each person does not change much in the same time period, and the difference between the peak point of the P wave and the peak point of the tidal wave will not exceed half of the width of a pulse wave. We use the difference between the two peak points as the screening condition, and screen out two points with a difference greater than 0.6 times the average pulse width. Compare the amplitudes of the two points, and the point with a larger amplitude is the peak point of the P wave.

The compression result already contains the extreme value information required for calculating the pulse rate, so when extracting the pulse rate from the compression result, it is to find the key points greater than the threshold and compare them for false detection. If there is a false detection peak point, the false detection point is removed according to the schematic flow chart shown in Figure 13, and the pulse rate is calculated according to formula (19). If there is no false detection point, the pulse rate is calculated directly according to the pulse rate equation. For any set of pulse signals in the experimental data, the pulse rate calculated by the proposed method is consistent with the result before compression, as shown in Figure 14. In addition, wemos D1 mini (ESP8266) is used to verify the accuracy of the proposed method based on the Arduino development environment. As shown in Figure 15, the two sides of "->" are the sampling time and pulse rate of the sampled signal. The results show that the method is feasible in practical applications. Therefore, it is expected to be used for health monitoring in wearable devices.

The embodiments of this specific implementation method are all preferred embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Therefore, all equivalent changes made based on the structure, shape, and principle of the present invention should be included in the protection scope of the present invention.

Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims (8)

1. A method for online compression of an electrocardiogram pulse signal, characterized in that it comprises the following steps: S01: Collecting ECG signals and pulse signals, performing analog-to-digital conversion on the signals, and obtaining ECG pulse signals on the human body surface; S02: Design an adaptive filter, improve the existing adaptive filter, and add a high-frequency noise estimation part; S03: Online adaptive filtering, combining the data update process in the microprocessor system cache with the adaptive filtering principle, and using a sliding window iteration method to perform calculations to achieve real-time filtering of the ECG pulse signal; The online adaptive filtering, combining the data update process in the microprocessor system cache with the adaptive filtering principle, and using a sliding window iteration method to perform calculations to achieve real-time filtering of the ECG pulse signal, is specifically: Based on sliding window iteration, the real-time filtering of data is converted into filtering of the entire current buffer area; when a new sampling point slides into the data buffer area, the data in the buffer area is updated, and the data processing of the new sampling point is completed on the basis of the original M-1 sampling points; The online adaptive filtering specifically includes an initialization process and an iteration process: Initialization process: When the current data sampling point is less than or equal to the length of the data buffer (0≤n≤M-1), each latest sampling point is placed in the nth byte of the buffer; then, the sampling values in the data buffer are arranged in ascending order, and the mean is calculated to estimate the high-frequency noise of the data; the high-frequency noise estimation result is recorded as [0,…,x0,…,xn]; the output after filtering out the high-frequency noise is used as the reference input xn of the adaptive filter, and the data is further subjected to online adaptive filtering operation, and the result is recorded as [0,…,y0,…,yn]; then, the filter coefficients are updated until the sampling point n=M-1, completing the initialization process; Iterative process: when the current data sampling point is greater than the length of the data buffer (n>M), the data in the buffer are all shifted left by one bit, and each latest sampling point is placed in the Mth byte of the buffer; then, the sampling values in the data buffer are arranged in ascending order, and the mean is calculated to estimate the high-frequency noise of the data; the high-frequency noise estimation result is recorded as [xn(n-M+1),…, xn(M),…, xn(n)]; the output after filtering out the high-frequency noise is used as the reference input xn of the adaptive filter, and the data is further subjected to online adaptive filtering operation, and the result is recorded as [yn(n-M+1),…, yn(M),…, yn(n)]; then, the coefficients of the filter are updated until the sampling is completed, and the iterative process is completed; S04: compress and encode the ECG pulse signal, refer to the extreme points and inflection points in the ECG pulse signal as key points, extract the key points of the ECG pulse signal online, and realize the compression of the ECG pulse signal; the compression process of the ECG pulse signal is as follows: determine the extreme value according to the definition of the extreme point, calculate the straight line between two adjacent extreme points by Lagrange's middle value theorem, subtract the ECG pulse signal from the straight line to obtain the difference curve, extract the extreme points of the difference curve, and realize the compression of the ECG pulse signal; encode the compressed ECG pulse signal, the ECG pulse signal encoding process is as follows: encode the position and amplitude corresponding to the key parameters of the ECG pulse signal according to the transmission rules of the microprocessor, and ensure that the encoded data length is an integer multiple of 1 byte. If the length of a data point is not an integer multiple of 1 byte, fill the high bit of the data with 0 until the condition is met; for a microprocessor with a voltage resolution of 10 bits, the position and amplitude corresponding to each key parameter after encoding occupy 3 bytes, and the encoded data is sent in the order of position first and amplitude later; S05: decoding and recovery, decoding the compressed ECG pulse signal, the ECG pulse signal decoding process is as follows: extracting and dividing the received data according to the position of the packet header and the packet tail, each 3 bytes of data corresponds to a key parameter, wherein the first byte corresponds to the position of the key parameter, the second byte corresponds to the high 8 bits of the key parameter amplitude, and the third byte corresponds to the low 8 bits of the key parameter amplitude; recovering the amplitude of the key parameter according to the high 8 bits and the low 8 bits of information to achieve decoding of the ECG pulse signal; recovering the ECG pulse signal: using a cubic spline interpolation method to interpolate the decoded ECG pulse signal key parameters to achieve recovery of the ECG pulse signal; S06: Evaluate the performance of the algorithm and apply the algorithm. Use root mean square error and compression ratio to evaluate the performance of the algorithm. At the same time, based on the extracted key points, calculate the pulse main wave peak and the pulse wave main wave interval, and calculate the heart rate. 2. The method for online compression of an electrocardiogram pulse signal according to claim 1, wherein the step S02 estimates the high-frequency noise of the signal, comprising: According to the characteristic that mean filtering can effectively filter out high-frequency noise in the signal, mean filtering is combined with the microprocessor data update process. When the buffer length is M, the maximum and minimum values of the current buffer are removed, and the mean of the remaining M-2 data is calculated to complete the estimation of the high-frequency noise in the signal. 3. The online compression method for ECG pulse signal according to claim 1, characterized in that the compression of the ECG pulse signal in step S04 is achieved by extracting key points in the signal: The key points in the signal are composed of extreme points and inflection points. According to the definition of extreme points, at least three points are used to judge the extreme values. The current data buffer is recorded as buffer. If buffer(M-2) is greater than buffer(M-1) and buffer(M-3), then buffer(M-2) is the maximum value. If buffer(M-2) is less than buffer(M-1) and buffer(M-3), then buffer(M-2) is the minimum value. Two arrays E and L are set to store the amplitude and position corresponding to the extreme points respectively. Buffer(j) is a known extreme point. , the amplitude and position of buffer(j) are stored in E(3) and L(3) respectively; in the process of continuous data acquisition and processing, if buffer(M-2) is not an extreme point, the amplitude E(3) of the previous extreme point buffer(j) remains unchanged, and the position L(3) decreases by 1; if buffer(M-2) is an extreme point, the amplitude E(3) of the previous extreme point buffer(j) is shifted left by three bits and stored in E(0), and the position L(3) is shifted left by three bits and stored in L(0), and the amplitude and position of the current extreme point buffer(M-2) are stored in E(3) and L(3); The process of extracting the inflection point is as follows: after the extreme point is determined, the straight line between two adjacent extreme points is calculated using the Lagrange middle value theorem, the ECG pulse signal is subtracted from the straight line to obtain a difference curve, and the extreme point of the interpolation curve is calculated, which is the inflection point of the ECG pulse signal. Its corresponding parameters are stored in E(1), E(2) and L(1), L(2). 4. The online compression method for ECG pulse signal as described in claim 1 is characterized in that the microprocessor transmits and sends data according to an integer multiple of 1 byte. During the encoding process, it is necessary to ensure that the encoded data is an integer multiple of 1 byte. If the length of a data is not an integer multiple of 1 byte, the high bit is padded with zero until the condition is met; after the ECG pulse signal is filtered in the microprocessor, the data type becomes a floating point type, and the data type is forcibly converted to an integer type, and key parameters are further extracted from the data; for a microprocessor with an input voltage resolution of 10 bits, the position and amplitude of each key parameter after encoding occupy 3 bytes, among which the first byte corresponds to the position of the key parameter, and the last two bytes correspond to the amplitude of the key parameter. 5. The online compression method for electrocardiogram pulse signal as described in claim 1 is characterized in that the microprocessor sends and receives the encoded key parameters in the order of position first and amplitude later, divides the received data according to the position of the packet header and the packet tail, and each 3 bytes correspond to the characteristic parameters of a key point; sets array D and array P to store the position and amplitude of the decoded key point, wherein the position of the key parameter corresponding to the first byte is stored in D(M-1), the second byte corresponds to the high 8 bits of the key parameter amplitude, and the third byte corresponds to the low 8 bits of the key parameter amplitude, and the amplitude of the key parameter is restored according to the high 8 bits and the low 8 bits of information, and the restored amplitude of the key parameter is stored in P(M-1). 6. A compression system for implementing the online compression method for electrocardiogram pulse signal according to any one of claims 1 to 5, characterized in that it comprises: Data acquisition module, which can complete the acquisition of ECG signals and pulse signals; A filtering processing module performs adaptive filtering on the signal; Compression module, which compresses, encodes and sends signals; The decompression module receives, stores and decodes the signal. 7. The compression system as described in claim 6 is characterized in that the filtering processing module includes: a high-frequency noise estimation module, an adaptive filtering algorithm module, and a filter module. 8. A computer storage medium having a computer program stored thereon, wherein when the computer program is executed, the online compression method for electrocardiogram pulse signal according to any one of claims 1 to 5 is implemented.