KR101400316B1 - Apparatus and method for eliminating noise of biomedical signal - Google Patents

Abstract

A biological signal noise canceling apparatus and a method thereof are disclosed. At this time, the bio-signal noise elimination apparatus includes a wavelet learning unit for receiving a noise-free biosignal and generating a probability table indicating a probability of selecting a base wavelet function through wavelet training, a bio- And a noise canceller for removing the noise from the bio-signal including the noise through wavelet transformation and wavelet reduction using the optimal wavelet function selected based on the probability table.
As described above, according to the present invention, it is possible to optimally reduce noise that may occur in various experimental environments such as EEG (Electroencephalography), EMG (electromyography), ECG (Electrocardiogram) and FNIRS (FUNCTIONAL NEAR INFRARED SPECTROSCOPY). In addition, since noise cancellation is an essential element in BCI and bio-signal-based diagnostic assistant software technology, it is highly likely to be commercialized in the future.

Description

[0001] APPARATUS AND METHOD FOR ELIMINATING NOISE OF BIOMEDICAL SIGNAL [0002]

The present invention relates to an apparatus for removing a biological signal noise and a method thereof.

BCI (Brain - Computer Interface) Researches and bio - signals are used to classify the problems of machine control and organ function by using human bio - signals in hospitals and clinic centers.

Noise-free signal measurement is essential for this bio-signal measurement. However, a living body signal in which noises are mixed due to various interference by a power source, a muscle, a measuring electrode, and a subject is measured.

Conventionally, wavelet filtering and adaptive wavelet noise cancellation techniques have been developed to remove noise from a biological signal.

Conventional studies have chosen optimization techniques for specific wavelet functions rather than using the newly developed wavelet functions. In the conventional wavelet based approach, noise removal of electrocardiogram signal was performed by applying only one base wavelet such as Daubechies to remove the noise of the electrocardiogram signal of the entire T time length.

However, optimal wavelet functions may be different according to a specific time interval in a bio-signal to be analyzed, and a method for solving the problem is needed.

That is, in the case of the bio-signal to be analyzed, a pattern in which the generation pattern of the bio-signal changes dynamically during the entire time T can be shown according to the measurement environment of the tester, the measurement state, This means that the optimal base wavelet may be different when a wavelet transform based processing method is used in a specific time interval.

Therefore, the noise reduction technique of the optimal optimal wavelet-based ECG signal can not provide optimal noise cancellation result.

SUMMARY OF THE INVENTION It is an object of the present invention to provide a biological signal noise cancellation apparatus and method for enhancing noise removal efficiency by selecting an optimal base wavelet function according to a specific time interval in an entire bio-signal for efficient noise cancellation.

According to an embodiment of the present invention, a biological signal noise elimination apparatus includes a wavelet learning unit that receives a noiseless biological signal and generates a probability table indicating probability of selecting a base wavelet function through wavelet training, And a noise canceller for removing the noise from the bio-signal including the noise through wavelet transformation and wavelet reduction using an optimal base wavelet function selected based on the probability table.

Wherein the wavelet learning unit comprises:

The method of claim 1, further comprising: dividing a total time interval of the noise-free biomedical signal into a dynamic fragment period, applying a base wavelet function to each dynamic fragment period to output a compensated signal through wavelet transform and inverse transform, It is possible to generate the probability table in which the mean square error between bio-signals is expressed by the frequency at which the base wavelet function is selected, and then the frequency is converted into a probability.

Wherein the wavelet learning unit comprises:

A matrix composed of the mean squared errors is generated, and as the value of the matrix is closer to 0, the frequency value to be selected as the optimal basis wavelet can be increased.

The noise-

The entire time interval for the noise-containing biomedical signal is subdivided into a dynamic fragment period, and then wavelet transform and reduction and wavelet inverse transformation are performed by applying optimal base wavelet functions selected using the probability table for each dynamic fragment period Thereby outputting a biomedical signal from which noise has been removed.

The noise-

The wavelet reduction can be performed by applying a thresholding technique.

The probability table includes:

And may include an n-gram table.

According to another aspect of the present invention, there is provided a method for removing a bio-signal noise in a bio-signal-noise canceling apparatus, the method comprising: receiving a no-noise bio-signal and displaying a probability of selecting a base wavelet function through wavelet training; Generating a probability table; and removing the noise from a bio-signal including noise through wavelet transformation and wavelet reduction using an optimal basis wavelet function selected based on the probability table.

Wherein the step of generating the probability table comprises:

Performing a wavelet transform and an inverse transform by applying base wavelet functions to each of the dynamic segment periods, performing a wavelet transform and an inverse transform on the dynamic time domain, Generating a mean square error matrix including values indicating a degree of reconstruction between the compensated biosignal and the noiseless biomedical signal; converting the mean square error matrix into a table of frequency units in which the base wavelet function is selected; And generating the probability table by converting a frequency included in the table of the frequency unit into a probability value.

Wherein the step of converting into the table of frequency units comprises:

Increasing the frequency 1 when the value indicating the degree of restoration is smaller than the value obtained by adding the standard deviation to the minimum value; and increasing the frequency when the value indicating the degree of restoration is greater than the added value can do.

The step of removing the noise includes:

Receiving a bio-signal including a noise, subdividing the entire time interval of the bio-signal including the noise into a dynamic segment period, applying optimal base wavelet functions selected through the probability table for each dynamic segment period, Performing wavelet transformation and wavelet reduction, and performing inverse transform processing on the biometric signal on which wavelet transformation and wavelet reduction have been performed to remove noise.

Wherein performing the wavelet reduction comprises:

The wavelet reduction can be performed by applying a thresholding technique.

According to the embodiment of the present invention, it is possible to optimally reduce noise that may occur in various experimental environments such as EEG (Electroencephalography), EMG (electromyography), ECG (Electrocardiogram), FNIRS (FUNCTIONAL NEAR INFRARED SPECTROSCOPY)

In addition, since noise cancellation is an essential element in BCI and bio-signal-based diagnostic assistant software technology, it is highly likely to be commercialized in the future.

1 is a block diagram of an apparatus for removing a biological signal noise according to an embodiment of the present invention.
2 is a flowchart illustrating a learning method for eliminating a bio-signal noise according to an embodiment of the present invention.
3 is a flowchart illustrating a method of removing a bio-signal noise according to an embodiment of the present invention.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those skilled in the art can easily carry out the present invention. The present invention may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. In order to clearly illustrate the present invention, parts not related to the description are omitted, and similar parts are denoted by like reference characters throughout the specification.

Now, a biological signal noise canceling apparatus and a method thereof will be described with reference to the drawings.

FIG. 1 is a configuration diagram of an apparatus for removing a biological signal noise according to an embodiment of the present invention, FIG. 2 is a flowchart illustrating a learning method for eliminating a biological signal noise according to an embodiment of the present invention, FIG. FIG. 4 is a flowchart illustrating a method of removing a biological signal noise according to an example.

Referring first to FIG. 1, a biological signal noise removing apparatus includes a wavelet learning unit 100 and a noise removing unit 200.

The wavelet learning unit 100 and the noise eliminator 200 apply an optimal basis wavelet selection method based on a Discrete Wavelet Transform (DWT) to remove noise of a biological signal.

The electrocardiogram signal including the noise continuously collected for the entire time T can be expressed by Equation (1).

Here, X (T) is an electrocardiogram signal including noises, S (T) is an electrocardiogram signal including no noise, and E (T) is a noise signal.

The wavelet learning unit 100 and the noise removing unit 200 divide the entire electrocardiogram signal T into N pieces of dynamic pieces and search for an optimal basis wavelet for each divided interval.

At this time, the electrocardiogram signal separated for each section can be expressed by Equation (2).

Here, k and l are indices indicating a specific time interval when they are divided into N units. And t is the time interval for subdivision when divided into N pieces.

In addition, when the electrocardiogram signal measured during the entire time T period is divided into N, it can be expressed as Equation (3).

에서 최적의 기저 웨이블릿 함수를 찾는다.At this time, the wavelet learning unit 100 and the noise removing unit 200 calculate

We find an optimal base wavelet function.

The wavelet learning unit 100 performs wavelet learning for selecting an optimal basis wavelet function for a specific interval for a biomedical signal to remove noise. The noise eliminator 200 performs noise elimination on the bio-signal based on the learning information generated by the wavelet learning unit 100.

Here, the operation of the wavelet learning unit 100 will be described with reference to FIG.

Referring to FIG. 2, the wavelet learning unit 100 receives a noise-free biomedical signal including no noise (S101).

The wavelet learning unit (100) limits a biological signal at the time of wavelet learning to a biological signal not including a noise.

In step S103, the wavelet learning unit 100 segments the noiseless bio-signal input in step S101 into dynamic sub-units, i.e., N sub-sub-units, with respect to the entire electrocardiogram signal T period.

In step S103, the wavelet learning unit 100 performs wavelet transform by applying base wavelet functions to each segmented dynamic segment unit in step S103.

At this time, the wavelet learning unit 100 performs wavelet transform on the basis wavelet functions capable of performing discrete wavelet transform (DWT) analysis.

Here, there are about 50 or more M wavelet functions that can be analyzed in MATLAB, such as Haar, Daubechies, Symets, Coiflets, Biorthogonal, Reverse biorthogonal, and Discrete approximation of Meyer.

The wavelet learning unit 100 performs a wavelet inverse wavelet transform (S107) on the wavelet coefficients filtered by the base wavelet functions, and outputs a reconstructed biomedical signal (S109).

The wavelet learning unit 100 generates a mean square error (MSE) matrix as shown in Table 1 as M × N mathematical equations (4) (S111).

Here, M is the number of the base wavelets that can perform DWT analysis, and N is the number obtained by dividing the T section into N pieces.

This average square error is a value indicating the degree of reconstruction after the original signal and wavelet transform, and the value closest to 0 is the optimal base wavelet function.

는 무잡음(noise-free) 서브 심전도 신호 구간이다.

에 특정 노이즈가 포함된 신호,즉

에 화이트 가우시안 노이즈(White Gaussian Noise) 등 여러 잡음이 섞인 신호에 대해서

는 M개 중 i번째 기저웨이블릿(mother wavelet)를 적용하여 쓰레스홀딩(thresholding) 기법을 이용하여 잡음 제거 후 IDWT(Inverse DWT)를 수행하여 재복원된 서브 심전도 신호 구간이다. In step S111, an MxN matrix is generated, where i is an index for M. That is, 1 <= i <= M, j is an index for N. That is, 1 <= j <= J. i, j are natural numbers. And

Is a noise-free sub-ECG signal section.

A signal including a specific noise, that is,

For white noise signals such as white Gaussian Noise,

Is a reconstructed sub-ECG signal interval by performing an IDWT (Inverse DWT) after noise removal using a thresholding technique using an i-th base wavelet of M pixels.

1 to 360 360 ~ 720 ... 21240-21600 db2 0.001951 0.001758 0.001825 0.002229 db2 0.002042 0.001890 0.001810 0.001830 db4 0.001889 0.001860 0.001692 0.001632 ... 0.001951 0.001758 0.001825 0.002229

The wavelet learning unit 100 generates an n-gram table by converting it into a frequency table using the MSE matrix shown in Table 1 (S113).

Here, a threshold condition such as Equation (5) is applied as a criterion for converting the MSE matrix into frequency units.

에 있어서 DWT 변환을 통하여 복원된 신호가 MxN개의 행으로 나타낸다. 그리고 특정 구간

에 적합한 기저 웨이블릿으로 선택하기 위해 M개의 기저 웨이블릿으로 적용된 MSE 값들을 이용하여 최소 값에 표준편차 값을 더한 값보다 작은 경우 빈도 1을 증가시킨다. 그렇지 않으면 특정 구간에 해당되는 기저 웨이블릿함수는 적합하지 않다는 표시로 빈도를 증가시키지 않는다. 이와 같은 방식을 MSE 행렬 요소들 모두에 적용하면 표 2와 같은 값을 구할 수 있다.The wavelet learning unit 100 uses the MSE matrix as a condition for transforming the frequency table,

The signal restored by the DWT conversion is represented by MxN rows. Then,

The MSE values applied to the M base wavelets are used to select a base wavelet suitable for the base wavelet, and the frequency 1 is increased when the minimum value is smaller than the sum of the standard deviation values. Otherwise, the base wavelet function corresponding to a certain interval does not increase frequency with an indication that it is not suitable. Applying this method to all of the MSE matrix elements yields the values shown in Table 2.

1 to 360 360 ~ 720 ... 21240-21600 ... One One One One db2 One One One One db3 One One One 0 Db4 One 0 One 0

The wavelet learning unit 100 transforms the probability values of the n-gram units as shown in Table 3 based on the frequency of Table 2 for optimal basis wavelet selection. Table 3 shows an example of bi-gram conversion.

... dmey 0.0206041179 haar - db1 0.0206041179 haar - db2 0.0198587930 ...

As described above, the wavelet learning unit 100 generates a matrix as shown in Table 1 using Equation (4). Then, based on these matrices, the frequency matrix is transformed into a frequency matrix as shown in Table 2 using Equation (5). Then, an n-gram table in which the frequency is converted into an n-gram probability as shown in Table 3 is generated and output.

Here, an n-gram table is used as an example of the probability-converted table, but the present invention is not limited thereto. The wavelet learning unit 100 may generate various probability tables indicating the probability that the base wavelet function is selected as a probability.

The noise removing unit 200 searches for an n-gram table output by the wavelet learning unit 100 to select a base wavelet function optimal for the entire section and the dynamic section, performs thresholding, To remove noise. At this time, the noise removing unit 200 can remove the sums using various probability tables in addition to the n-gram table.

The operation of the noise removing unit 200 will now be described with reference to FIG.

Referring to FIG. 3, the noise removing unit 200 receives a noisy biomedical signal as shown in Equation (1) (S201).

In step S203, the noise removing unit 200 separates the noise bio-signal input in step S201 into dynamic sub-units, i.e., N sub-units, of the entire electrocardiogram signal T, as shown in equation (3).

The noise removing unit 200 performs wavelet transform and wavelet shrinkage by selecting an optimal basis wavelet function based on an n-gram table as shown in Table 3 generated by the wavelet learning unit 100 S205).

At this time, wavelet reduction is an operation for noise reduction, and it is possible to apply a thresholding technique, which is a wavelet reduction method proposed by Donoho.

Here, the thresholding technique includes a soft threholding technique and a hard thresholding technique.

The noise removing unit 200 performs a wavelet inverse transform process on the N bio-signals for which the wavelet reduction has been completed (S207), and outputs the noise-removed biomedical signal (S209).

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed exemplary embodiments, It belongs to the scope of right.

Claims (11)

A wavelet learning unit for receiving a noise-free biosignal and generating a probability table indicating probability of selecting a base wavelet function through wavelet training; and
A noise removing unit that receives a bio-signal including noise, removes the noise from the bio-signal including the noise through wavelet transformation and wavelet reduction using an optimal base wavelet function selected based on the probability table,
And outputs the biosignal signal to the living body. The method according to claim 1,
Wherein the wavelet learning unit comprises:
The method of claim 1, further comprising: dividing a total time interval of the noise-free biomedical signal into a dynamic segment period, applying a base wavelet function to each dynamic segment period, outputting a compensated signal through wavelet transform and inverse transform, A biomedical signal noise elimination apparatus for generating a probability table in which the mean square errors between biomedical signals are represented by a frequency at which a base wavelet function is selected and the probability is converted into a probability. 3. The method of claim 2,
Wherein the wavelet learning unit comprises:
And generates a matrix composed of the mean squared errors, and increases the frequency value to be selected as an optimal basis wavelet as the value of the matrix is closer to zero. 3. The method of claim 2,
The noise-
The entire time interval for the noise-containing biomedical signal is subdivided into a dynamic fragment period, and then wavelet transform and reduction and wavelet inverse transformation are performed by applying optimal base wavelet functions selected using the probability table for each dynamic fragment period And outputting the biomedical signal from which the noise has been removed. 5. The method of claim 4,
The noise-
A biosignal noise canceller for performing wavelet reduction by applying a thresholding technique. The method according to claim 1,
The probability table includes:
(N-gram) table. A method of removing a bio-signal noise of a bio-signal noise removing apparatus,
Generating a probability table that receives a noiseless biomedical signal and expresses a probability of selecting a base wavelet function through wavelet training;
Removing the noise from a bio-signal including noise through wavelet transformation and wavelet reduction using an optimal basis wavelet function selected based on the probability table
Wherein the bio signal noise elimination method comprises the steps of: 8. The method of claim 7,
Wherein the step of generating the probability table comprises:
Receiving the noiseless biomedical signal,
Subdividing the entire time interval for the noiseless bio-signal into a dynamic fragment period,
Performing wavelet transform and inverse transform by applying base wavelet functions to each dynamic segment period,
Generating a mean square error matrix including values indicating a degree of restoration between the compensated biosignal and the noiseless biomedical signal after the inverse transform,
Transforming the mean square error matrix into a table of frequency units in which the base wavelet function is selected, and
Generating a probability table by converting a frequency included in the frequency unit table into a probability value;
Wherein the bio signal noise elimination method comprises the steps of: 9. The method of claim 8,
Wherein the step of converting into the table of frequency units comprises:
Increasing the frequency 1 if the value indicating the degree of restoration is smaller than the sum of the minimum value and the standard deviation value, and
If the value indicating the degree of restoration is greater than the sum,
Wherein the bio signal noise elimination method comprises the steps of: 9. The method of claim 8,
The step of removing the noise includes:
Receiving a bio-signal including noise,
Subdividing the entire time interval of the bio-signal including the noise into a dynamic fragment period,
Performing wavelet transform and wavelet reduction by applying optimal base wavelet functions selected through the probability table for each dynamic segment period, and
Performing inverse transform processing on the biometric signal on which the wavelet transform and the wavelet reduction are performed to remove noise
Wherein the bio signal noise elimination method comprises the steps of: 11. The method of claim 10,
Wherein performing the wavelet reduction comprises:
A method of removing a biological signal noise by performing wavelet reduction by applying a thresholding technique.

Images