WO2020151169A1

WO2020151169A1 - Method for automatic removal of frictional sound interference of electronic stethoscope

Info

Publication number: WO2020151169A1
Application number: PCT/CN2019/091484
Authority: WO
Inventors: 蔡盛盛; 胡南; 徐兴国; 周宁
Original assignee: 苏州美糯爱医疗科技有限公司
Priority date: 2019-01-23
Filing date: 2019-06-17
Publication date: 2020-07-30
Also published as: CN109805954B; CN109805954A

Abstract

Disclosed in the present invention is a method for automatic removal of frictional sound interference of an electronic stethoscope, related to the technical field of stethoscope noise reduction. The present invention comprises performing empirical mode decomposition on data x_K in an interference time region number K in x, obtaining a number M_K of intrinsic mode function components: (I); performing low-pass filtering of IMF₂ and IMF₃, performing frame blocking, and calculating correlation coefficients of IMF₂ and IMF₃, determining a number P of data point regions recorded as: (II), zeroing all data outside of the data point regions; calculating (III), and sequentially assigning updated values to edge data segments of the P data point regions by means of cubic spline interpolation. The present invention quickly and accurately detects and locates one or more regions where frictional sound interference may have occurred in a stethoscope signal; realizing automatic removal of frictional sound in single-channel stethoscope data also realizing automatic removal of frictional sound of multiple frictional sound interference regions in the same segment of data.

Description

Method for automatically eliminating friction noise interference of electronic stethoscope

Technical field

The invention belongs to the technical field of stethoscope silencing processing, and particularly relates to a method for automatically eliminating friction noise interference of an electronic stethoscope.

Background technique

Auscultation allows physicians to understand the patient’s condition in a simple and easy way. Traditional auscultation technology is often restricted by factors such as the location of the visit and the level of medical skills. With the vigorous development of the Internet of Things (IoT) technology, various types of electronic stethoscopes continue to appear, enabling the real-time monitoring and automatic transcription of the patient’s heart and lung sound data , Cloud diagnosis and treatment and intelligent diagnosis become possible, and a series of problems presented in the research process of electronic stethoscope have also attracted attention. The electronic stethoscope converts the weak heart and lung sounds into electrical signals through the transducer on the auscultation head. In practical applications, such as the switching and adjustment of the auscultation position, the movement of the auscultation head may cause auscultation The friction sound between the head and clothing is recorded by the electronic stethoscope (for example, when a large external force is applied to the auscultation head unintentionally), which interferes with the normal auscultation signal: for example, the pediatric clinic is carrying out auscultation due to the low degree of cooperation of infant patients Frictional noise may occur in the media; another example is when the patient uses an electronic stethoscope and uploads the auscultation data to the cloud, improper operation may also cause frictional noise interference.

The appearance of friction noise makes the doctor’s electronic stethoscope use experience worse and interferes with their diagnosis results; on the other hand, when artificial intelligence is introduced for automatic auscultation, the signal quality degradation caused by friction noise will affect the subsequent heart sound positioning, heart and lungs. The effect of automatic sound diagnosis. For electronic stethoscopes designed to integrate functions such as high-precision real-time monitoring of the fetus, intelligent assessment of heart/lung function, automatic diagnosis of heart/lung diseases, etc., it can automatically locate and eliminate the friction noise in the auscultation signal. These artificial intelligence functions can One of the important prerequisites for realization. However, although the electronic stethoscope signal may be disturbed by fricative noise is an urgent problem for electronic stethoscopes, there are only a few publicly published documents or patents that consider this problem from the perspective of hardware design, not from the perspective of back-end signal processing. Solve the problem. In practice, the hardware design of an electronic stethoscope cannot get rid of the principle that the auscultation head converts vibration waves into electrical signals through contact with the human body. Therefore, the improvement of the hardware may weaken the interference of the friction sound to a certain extent, but it cannot completely prevent the friction sound from being caused by the stethoscope. Therefore, there is an urgent need to eliminate fricative interference from the perspective of signal processing.

At present, from the perspective of signal processing, the focus of existing patents related to electronic stethoscopes includes: signal preprocessing (including noise reduction, heart sound localization, cardiopulmonary sound separation, etc.) and signal intelligent analysis (fetal heart rate monitoring, heart sound-based heart Intelligent diagnosis of diseases, intelligent diagnosis of respiratory diseases based on lung sounds), etc., but there is no patent to propose a signal processing solution that automatically eliminates friction interference for the problem that the signal may be interfered by friction noise.

The invention patent "Contact electronic stethoscope that can avoid noise interference during auscultation" (application number CN200510063183.2) proposes a contact electronic stethoscope that is said to avoid noise interference during auscultation. The contact microphone on the auscultation head There is an elastic body on the outside. When the contact microphone in the auscultation head does not reach a certain pressure, it is kept at a certain distance from the human body, which can avoid contact with each other when the auscultation head moves on the surface of the body or clothing and receives harsh and irrelevant human information. Friction noise. The unsolved problems of the patent include: (1) When the microphone is not in contact with the human body, the received auscultation signal is relatively weak, which may affect the signal quality in some cases; (2) When the contact microphone in the auscultation head reaches a certain level It can still come into contact with the human body under pressure, and friction noise may still be generated at this time.

The utility model patent "A friction noise reduction device for electronic stethoscopes" (application number CN201721387654.X) proposes a friction noise reduction device for electronic stethoscopes. The sensor is closely attached to the inner wall of the fixed warehouse for Fix and support the sensor. The bottom surface of the fixed warehouse and the outer wall of the fixed warehouse are not in contact with any other shells or components, which minimizes the conduction path of friction noise and reduces the distance between the stethoscope shell and the skin and clothing of the doctor or patient to a certain extent The friction noise. The patent still considers the reduction of fricative interference from the perspective of hardware layout, and this method can only improve the anti-frictional interference performance of the auscultation signal to a certain extent. In some cases, the back end may still receive fricative interference signals.

The prior art only considers the problem of fricative interference elimination from the perspective of hardware design, but does not solve the problem from the perspective of back-end signal processing. Due to the limitation of the auscultation principle, the improvement brought by hardware design is very limited. It is necessary to design an automatic elimination method of fricative interference from the perspective of signal processing. The currently missing work in this area includes:

1. At present, there is no automatic identification and location method for the friction noise interference area: in fact, even if there is only this link in the electronic stethoscope, in the subsequent intelligent analysis stage, the friction noise interference area that is automatically identified and located can be shielded to avoid this problem. Analyze the error.

2. There is currently no signal processing method for automatically eliminating fricative interference: on the one hand, the lack of automatic identification and positioning methods for fricative interference regions makes this work impossible to carry out. On the other hand, the single-channel blind signal separation involved in this problem is A difficulty in signal processing.

Summary of the invention

The purpose of the present invention is to provide a method for automatically eliminating frictional noise interference of an electronic stethoscope, which can quickly and accurately detect and locate one or more possible occurrences in the auscultation signal by calculating the Mel cepstrum coefficient of the data unit and determining the interference time interval. A fricative interference area; and use the fricative interference area to detect the positioning results, and for single-channel auscultation data, it can better realize the automatic elimination of fricative interference.

In order to solve the above technical problems, the present invention is realized through the following technical solutions:

The invention is an automatic elimination method of friction noise interference of an electronic stethoscope, which includes:

Step 1: Read the auscultation signal sampling sequence x(n), n=1,2,...,N with a duration of N in the buffer, and express it as a vector form x;

Step 2: Using Mel Cepstrum Coefficients (MFCC) and Support Vector Machines (SVM), determine the K non-overlapping interference time intervals interfered by fricatives on x, in chronological order:

[n _K,begin ,n _K,end ],[n _K-1,begin ,n _K-1,end ],...,[n _1,begin ,n _1,end ];

Among them, K≥0;

Step 3: Determine the size of K: if K=0, go to step 8; if K>0, go to step 4;

Step 4: The K interference time intervals are formed into a stack in the manner of entering first and then exiting the last appearing interval;

Step 5: Perform empirical mode decomposition (EMD) on the data x _K on the Kth interference time interval in x to obtain M _K eigenmode function components:

Step 6: Perform low-pass filtering on IMF ₂ and IMF ₃ , then calculate the correlation coefficients of IMF ₂ and IMF ₃ in frames, and determine P non-overlapping data segments with correlation coefficients greater than the preset threshold Th _c ;

Record the data point intervals corresponding to each of the data segments as follows:

Set all data outside the interval corresponding to these data points on IMF ₂ and IMF ₃ to zero;

Step 7: Calculation

And sequentially assign updated values to the edge data segments of the P data point intervals through cubic spline interpolation;

Step 8: Update the data on the x-th interference time interval as:

Remove the time interval [n _K,begin ,n _K,end ] at the top of the stack, set K=K-1, and return to step three;

Step 9: Output the auscultation signal x that eliminates the interference of friction noise.

Preferably, it is determined in step 2 that the K non-overlapping interference time intervals interfered by fricative noise on x specifically include the following:

First, divide the data x with a length of 0.2s as a data unit, and the data units overlap by 0.1s, and calculate the Mel cepstrum coefficient of each data unit;

If the data unit is expressed as s and the length is M, the specific processing is as follows:

(1) Add a Hanning window h of length M to s, and do N _FFT point Fast Fourier Transform (FFT) on it to calculate its power spectrum:

(2) Use f _mel (f) = 2959×log ₁₀ (1+f/700) to convert the linear frequency 0～f _s /2 into _mel frequency, and divide Q continuous frequencies evenly in the _mel frequency domain Overlap the area of 50%, and accordingly construct a filter bank ψ _q , q=1, 2,...,Q containing Q triangular filters, and calculate Q weighted outputs:

(3) The MFCC of this data segment can be expressed as a Q×1-dimensional vector c, and the qth element is

Calculated

Then, divide the MFCC vector c calculated on each data unit by max(|c|), and substitute it into the linear support vector machine f(c)=sign(w ^T c+b), where w and b are the linear For the normal vector and intercept of the support vector machine, when f(c)>0, it is judged that the data unit has friction noise interference; when f(c)<0, it is judged that the data unit has no friction noise interference;

Finally, if the distance between two similar data units that detect fricative interference is not greater than 0.1s, they are merged into the same uninterrupted interference time interval interfered by fricatives, and finally K non-interferences interfered by fricatives on x are obtained. The overlapping interference time intervals, in chronological order, are: [n _K,begin ,n _K,end ],[n _K-1,begin ,n _K-1,end ],...,[n _{1, begin} ,n _1,end ].

Preferably, performing empirical mode decomposition (EMD) on the data x _K on the Kth interference time interval in x in step 5 specifically includes the following:

If x _K contains N _K points, set the number of screening S _k = 8, and the maximum number of eigenmode function components

Search for the maximum point of x _K , and use cubic spline interpolation to fit the upper envelope e _u ;

Search for the minimum point of x _K , and use cubic spline interpolation to fit its lower envelope e _l ;

Calculate h=x _K -(e _u +e _l )/2, and replace x _K with h to repeat the above-mentioned screening process S _k times and output the current eigenmode function component IMF=x _K -h;

Replace x _K with the residual vector r that removes the extracted IMF;

If ||r|| ₂ /||x _K || ₂ <10 ^-6 or the extracted eigenmode function components are equal to M _K,max ; then output M _K eigenmode function components:

Otherwise, repeat the IMF extraction.

Preferably, step 6 specifically includes the following:

Use 13-order Butterworth digital filter with upper cutoff frequency of 0.06π to perform low-pass filtering on IMF ₂ and IMF ₃ ;

Take 0.02s as a data segment and overlap by 0.01s to calculate the correlation coefficient between the corresponding data segments of IMF ₂ and IMF ₃ ;

Determine the data segments whose correlation coefficient is greater than the preset threshold Th _c , and merge them into the same data segment when the distance between adjacent data segments is less than 0.01 s, thereby obtaining P non-overlapping data segments, which correspond to each in order The data point intervals of are respectively denoted as:

Keep the values of P data segments on IMF ₂ and IMF ₃ , and set all data at other times to zero.

Preferably, in step 7, assigning updated values to the edge data segments of the P data point intervals through cubic spline interpolation in sequence includes the following: Use interval P

As an interpolation point, use cubic spline interpolation

The corresponding edge data segment

Fit the points on, so that the repair result is smooth;

According to the sequence of occurrence, the edge data segments of the P data point intervals are sequentially updated.

The present invention has the following beneficial effects:

1. The present invention quickly and accurately detects and locates one or more fricative interference regions that may appear in the auscultation signal by calculating the Mel cepstrum coefficient of the data unit and determining the interference time interval; The data x _K is decomposed by empirical mode to obtain the corresponding eigenmode function components; the correlation coefficients of adjacent eigenmode function components are calculated and P non-overlapping data point intervals greater than the preset threshold Th _c are determined, and the data points Assign updated values in intervals to eliminate interference sounds.

2. The present invention can realize the automatic elimination of friction noise interference from single-channel auscultation data, and at the same time realize the automatic elimination of friction noise in multiple friction noise interference regions in the same piece of data.

Of course, any product implementing the present invention does not necessarily need to achieve all the advantages described above at the same time.

Description of the drawings

In order to explain the technical solutions of the embodiments of the present invention more clearly, the following will briefly introduce the drawings used in the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained from these drawings without creative work.

Fig. 1 is a flowchart of a method for automatically eliminating friction noise interference of an electronic stethoscope according to the present invention;

2 is a waveform and time-frequency spectrum diagram of auscultation data originally interfered by fricatives in the first embodiment of the present invention;

Fig. 3 is a component diagram of each eigenmode function obtained after EMD decomposition in the first embodiment of the present invention;

FIG. 4 is a diagram of the reserved areas in the final result of determining IMF2 and IMF3 based on the correlation coefficient in the first embodiment of the present invention;

Fig. 5 is a waveform of auscultation data and its time-frequency spectrum after the interference of the frictional noise is finally eliminated in the first embodiment of the present invention.

detailed description

The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the protection scope of the present invention.

Please refer to FIG. 1, the present invention is an automatic elimination method for friction noise interference of an electronic stethoscope, including:

Step 1: Read the auscultation signal sampling sequence x(n), n=1, 2,...,N with a duration of N in the buffer, and express it as a vector form x;

Among them, K≥0;

Calculate the Mel Cepstrum Coefficient (MFCC) for the sliding window of the auscultation signal to extract the features used to detect the friction noise; use linear support vector machine (SVM) to realize the automatic detection and location of the friction noise interference area;

Step 3: Determine the size of K: if K=0, go to step 8; if K>0, go to step 4;

Use empirical mode decomposition (EMD) to initially separate friction and cardiopulmonary auscultation signals;

Using the calculated correlation coefficients EMD results IMF _{IMF. 3} and _2, and further determines IMF _{IMF. 3} and ₂ with cardiopulmonary auscultation signal tone corresponding to a region, to avoid loss of signal details automatically eliminate the interference sound auscultation after rubbing;

Step 7: Calculation

The cubic spline interpolation is used to ensure the smoothness of the auscultation signal after the friction noise is automatically eliminated;

Step 8: Update the data on the x-th interference time interval as:

Using the "first-in-last-out" feature of the stack, the fricative interference in multiple areas can be automatically eliminated in sequence;

(4) Add a Hanning window h of length M to s, and perform N _FFT point fast Fourier transform (FFT) on it to calculate its power spectrum:

(5) Use f _mel (f) = 2959×log ₁₀ (1+f/700) to convert the linear frequency 0～f _s /2 into _mel frequency, and divide Q continuous frequencies evenly in the _mel frequency domain. Overlap the area of 50%, and accordingly construct a filter bank ψ _q , q=1, 2,...,Q containing Q triangular filters, and calculate Q weighted outputs:

(6) The MFCC of the data segment can be expressed as a Q×1-dimensional vector c, and the qth element is

Calculated

Replace x _K with the residual vector r that removes the extracted IMF;

Otherwise, repeat the IMF extraction.

Preferably, step 6 specifically includes the following:

As an interpolation point, use cubic spline interpolation

The corresponding edge data segment

Fit the points on, so that the repair result is smooth;

Specific embodiment one:

Please refer to Figure 2 for reading a piece of auscultation data x with a duration of 3 seconds, sampling rate f _s = 4kHz, and interference by friction noise, and divide it by the point with the largest absolute value for normalization. Figure 2 shows the waveform and time Spectrogram.

The data x is divided as a data unit with a length of 0.2s, and the data units overlap by 0.1s, and the MFCC of each data unit is calculated. Assuming that a certain data unit is denoted as s and the length is M=0.2×4000=80, do the following operations:

Add a Hanning window h with a length of M=800 points to s, and do it

Point FFT to calculate the power spectrum:

Use f _mel (f)=2959×log ₁₀ (1+f/700) to convert the linear frequency 0～f _s /2 into _mel frequency, and divide Q=20 continuous intersections evenly in the _mel frequency domain. Overlap 50% of the area, and accordingly construct a filter bank ψ _q containing Q=20 triangular filters, q=1, 2,...,Q, and calculate Q=20 weighted outputs:

The MFCC of the data segment is represented as a Q×1-dimensional vector c, and the qth element is

Calculated.

Then, divide the MFCC vector c calculated on each data unit by max(|c|) and substitute it into the linear SVM: f(c)=sign(w ^T c+b), where w and b are the linear support The normal vector and intercept of the vector machine are trained from 560 labeled auscultation data segments with fricative interference and 850 auscultation data segments without fricative interference. When f(c)>0, it is judged that the data unit has fricatives. Interference: When f(c)<0, it is judged that the data unit has no friction noise interference.

Finally, if the distance between two similar data units that detect fricative interference is not greater than 0.1s, they are merged into the same uninterrupted time interval interfered by fricative sounds, and finally a time interval interfered by fricative sounds on x is obtained. : [N _begin ,n _end ], where n _begin = 1.4 seconds and n _end = 2.2 seconds, as shown in the dashed box in Figure 2.

Time interval that will be disturbed by fricative noise

The data on is expressed as x _K , which contains

Set the number of screening S _k = 8, the maximum number of eigenmode function components

Search for the maximum point of x _K , and use cubic spline interpolation to fit the upper envelope e _u ; search for the minimum point of x _K , and use cubic spline interpolation to fit the lower envelope e _l ; Calculate h=x _K -(e _u +e _l )/2, and replace x _K with h to repeat the above-mentioned screening process S _k times, and then output the current eigenmode function component IMF=x _k -h.

Replace x _K with the residual vector r of the extracted IMF, repeat the above IMF extraction process several times, when it reaches ||r|| ₂ /||x _K || ₂ <10 ^-6 or the extracted eigenmode function component It ends when equal to M _K,max , and finally MK=9 eigenmode function components are obtained:

See Figure 3.

Use a 13-order Butterworth digital filter with an upper cut-off frequency of 0.06π to perform low-pass filtering on IMF ₂ and IMF ₃ ; use 0.02s as a data segment, overlap by 0.01s, and calculate the corresponding data of IMF ₂ and IMF ₃ . Correlation coefficient between segments; determine the data segment whose correlation coefficient is greater than the preset threshold Th _c =0.45, and merge them into the same data segment when the distance between adjacent data segments is less than 0.02s, thereby obtaining P = 4 non-overlapping data segments Data segment, record their respective data point intervals in sequence as

The result is shown in Figure 4 (the selected interval is in the dashed box); the values on the P data segments are retained on IMF ₂ and IMF ₃ , and the data at other times are all set to zero.

Calculation

And in turn, the edge data segments of the P data point intervals are assigned updated values through cubic spline interpolation: given L=10, use interval P

As an interpolation point, use cubic spline interpolation

The corresponding edge data segment

The above points are fitted to make the repair result smooth; the edge data segments of the P data point intervals are sequentially updated in the order of occurrence.

Update the data on the fricative interference interval of the original data x as

Output the auscultation signal x that eliminates the fricative interference. The resulting waveform and the corresponding time-frequency diagram are shown in Figure 5. It can be seen from the figure that the interval signal characteristics after the fricative interference is eliminated are very similar to the interval signal characteristics without fricative interference. Expert inspection also proved the performance of this method.

It is worth noting that in the above system embodiment, the units included are only divided according to functional logic, but not limited to the above division, as long as the corresponding function can be realized; in addition, the specific name of each functional unit It is only for the convenience of distinguishing each other, and is not used to limit the protection scope of the present invention.

In addition, those of ordinary skill in the art can understand that all or part of the steps in the methods of the foregoing embodiments can be implemented by programs instructing related hardware, and the corresponding programs can be stored in a computer readable storage medium.

The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiment does not describe all the details in detail, nor does it limit the invention to only the described specific embodiments. Obviously, many modifications and changes can be made according to the content of this manual. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is only limited by the claims and their full scope and equivalents.

Claims

A method for automatically eliminating friction noise interference of an electronic stethoscope, which is characterized in that it includes:

Step 1: Read the auscultation signal sampling sequence x(n), n=1,2,...,N with a duration of N in the buffer, and express it as a vector form x;

Step 2: Using Mel cepstrum coefficients and support vector machines, determine the K non-overlapping interference time intervals interfered by fricatives on x, which are in chronological order:

[n K,begin ,n K,end ],[n K-1,begin ,n K-1,end ],...,[n 1,begin ,n 1,end ];

Among them, K≥0;

Step 3: Determine the size of K: if K=0, go to step 8; if K>0, go to step 4;

Step 4: The K interference time intervals are formed into a stack in the manner of entering first and then exiting the last appearing interval;

Step 5: Perform empirical mode decomposition on the data x K on the Kth interference time interval in x to obtain M K eigenmode function components:

Step 6: Perform low-pass filtering on IMF 2 and IMF 3 , then calculate the correlation coefficients of IMF 2 and IMF 3 in frames, and determine P non-overlapping data segments with correlation coefficients greater than the preset threshold Th c ;

Record the data point intervals corresponding to each of the data segments as follows:

Set all data outside the interval corresponding to these data points on IMF 2 and IMF 3 to zero;

Step 7: Calculation
And sequentially assign updated values to the edge data segments of the P data point intervals through cubic spline interpolation;

Step 8: Update the data on the x-th interference time interval as:
Remove the time interval [n K,begin ,n K,end ] at the top of the stack, set K=K-1, and return to step three;

Step 9: Output the auscultation signal x that eliminates the interference of friction noise.
The method for automatically eliminating the friction noise interference of an electronic stethoscope according to claim 1, wherein the K non-overlapping interference time intervals that are determined by the friction noise interference on x in step 2 specifically include the following:

First, divide the data x with a length of 0.2s as a data unit, and the data units overlap by 0.1s, and calculate the Mel cepstrum coefficient of each data unit;

Then, divide the MFCC vector c calculated on each data unit by max(|c|), and substitute it into the linear support vector machine f(c)=sign(w T c+b), where w and b are the linear For the normal vector and intercept of the support vector machine, when f(c)>0, it is judged that the data unit has friction noise interference; when f(c)<0, it is judged that the data unit has no friction noise interference;

Finally, if the distance between two similar data units that detect fricative interference is not greater than 0.1s, they are merged into the same uninterrupted interference time interval interfered by fricatives, and finally K non-interferences interfered by fricatives on x are obtained. The overlapping interference time intervals, in chronological order, are: [n K,begin ,n K,end ],[n K-1,begin ,n K-1,end ],...,[n 1, begin ,n 1,end ].
The method for automatically eliminating friction noise interference of an electronic stethoscope according to claim 1, wherein the empirical mode decomposition of data x K in the Kth interference time interval in x in step 5 specifically includes the following:

If x K contains N K points, set the number of screening S k = 8, and the maximum number of eigenmode function components

Search for the maximum point of x K , and use cubic spline interpolation to fit the upper envelope e u ;

Search for the minimum point of x K , and use cubic spline interpolation to fit its lower envelope e l ;

Calculate h=x K -(e u +e l )/2, and replace x K with h to repeat the above-mentioned screening process S k times and output the current eigenmode function component IMF=x K -h;

Replace x K with the residual vector r that removes the extracted IMF;

If || r || 2 / eigenmode function component || x K || 2 <10- 6 equal to or extracted M K, max; M K eigenmodes function component output:
Otherwise, repeat the IMF extraction.
The method for automatically eliminating frictional noise interference of an electronic stethoscope according to claim 1, wherein step 6 specifically includes the following:

Use 13-order Butterworth digital filter with upper cutoff frequency of 0.06π to perform low-pass filtering on IMF 2 and IMF 3 ;

Take 0.02s as a data segment and overlap by 0.01s to calculate the correlation coefficient between the corresponding data segments of IMF 2 and IMF 3 ;

Determine the data segments whose correlation coefficient is greater than the preset threshold Th c , and merge them into the same data segment when the distance between adjacent data segments is less than 0.01 s, thereby obtaining P non-overlapping data segments, which correspond to each in order The data point intervals of are respectively denoted as:

Keep the values of P data segments on IMF 2 and IMF 3 , and set all data at other times to zero.
The method for automatically eliminating fricative noise interference of an electronic stethoscope according to claim 1, wherein in step 7, assigning updated values to edge data segments of P data point intervals sequentially through cubic spline interpolation includes the following: P use

As an interpolation point, use cubic spline interpolation
The corresponding edge data segment
Fit the points on, so that the repair result is smooth;

According to the sequence of occurrence, the edge data segments of the P data point intervals are sequentially updated.