KR102178789B1 - Motor Imagery Classification Apparatus and Method thereof using Adaptive Noise Cancellation and Correlation coefficient Optimization - Google Patents

Abstract

)를 갖는 회전행렬을 이용하여 회전시켜 회전 뇌전도 신호들을 얻는 단계, (c) 상기 회전각도(

)별로 상기 회전 뇌전도 신호들 간에 상관계수들을 산출하는 단계, (d) 상기 회전 뇌전도 신호들 간에 상관계수들에 대한 피셔 비율(Fisher`s ratio)을 산출하는 단계, (e) 상기 산출된 회전 뇌전도 신호들 간에 상관계수들을 최적화시켜 최적 상관계수들을 얻는 단계, (f) 상기 최적 상관계수들 중에서 상기 피셔 비율이 높은 순서대로 n개의 최적 상관계수들을 특징벡터로서 추출하는 단계, 및 (g) 상기 특징벡터를 이용하여 분류기를 학습하는 단계를 포함한다. 여기서, n은 자연수임.A method of classifying a motor image is disclosed. The motor image classification method is a motor image classification method performed by a motor image classification device.(a) For each class to be classified, the noise of the EEG signals measured from a plurality of electrodes, respectively, is removed from the adaptive noise cancellation method. ) Applying an algorithm to remove and obtain preprocessed EEG signals, (b) rotation angle of the preprocessed EEG signals (

Obtaining rotational EEG signals by rotating using a rotation matrix having ), (c) the rotation angle (

) Calculating correlation coefficients between the rotational EEG signals, (d) calculating a Fisher`s ratio for correlation coefficients between the rotational EEG signals, (e) the calculated rotational EEG Optimizing correlation coefficients between signals to obtain optimal correlation coefficients, (f) extracting n optimal correlation coefficients as feature vectors in the order of the highest Fisher ratio among the optimal correlation coefficients, and (g) the feature And learning the classifier using the vector. Where n is a natural number.

Description

Motor Imagery Classification Apparatus and Method thereof using Adaptive Noise Cancellation and Correlation coefficient Optimization

The present invention relates to an apparatus and method for classifying a motor image using an EEG signal, and more specifically, to an apparatus and method for classifying a motor image using preprocessing through an active noise reduction technique and optimizing a correlation between EEG signals.

Brain-Computer Interface (BCI) is a technology that directly connects brain activity and computer without muscles, which is an intermediate medium, using neural signals.In particular, electrical signals generated by brain activity are transmitted to the subject's scalp. Electroencephalogram (EEG), which is measured through an electrode attached to the device, can be easily measured by a non-invasive method, and has the advantage of having a high time resolution capable of directly measuring brain activity without delay. Since the measured EEG signal is measured through the scalp, it has a low signal-to-noise ratio (SNR) unlike the method of measuring by connecting directly to the brain.

On the other hand, the EEG signal contains information on biological interference such as blinking of eyes, movement of the heart, and thinking about other intentions in addition to the brain activity to be measured. This is a very important process for low EEG signals. On the other hand, among the existing technologies, there is a blind source separation (BSS) technique that uses correlation. However, the BSS technique has a problem in that it is not possible to select a signal source inside the brain that responds to accurate interference and noise because it is often made empirically to determine the role of signal sources inside the brain. Accordingly, there is a need for a method capable of removing interference and noise without estimation of an internal signal source using only an EEG signal. In addition, there is a method of using a strong correlation to noise as a feature. However, since it does not change the given correlations, there is a limit to choosing the correlation that is most likely to be classified within it.

Korean Registered Patent No.1238780 Korean Registered Patent No. 1796768

It is an object of the present invention to effectively remove interference and noise of EEG signals using an active noise reduction algorithm, and at the same time, improve classification accuracy compared to conventional techniques by using it for classification of motor images, characterized by a correlation that has robust characteristics against noise. It is in order.

)를 갖는 회전행렬을 이용하여 회전시켜 회전 뇌전도 신호들을 얻는 단계, (c) 상기 회전각도(

)별로 상기 회전 뇌전도 신호들 간에 상관계수들을 산출하는 단계, (d) 상기 회전 뇌전도 신호들 간에 상관계수들에 대한 피셔 비율(Fisher`s ratio)을 산출하는 단계, (e) 상기 회전 뇌전도 신호들 간에 상관계수들을 최적화시켜 최적 상관계수들을 얻는 단계, (f) 상기 최적 상관계수들 중에서 상기 피셔 비율이 높은 순서대로 n개의 최적 상관계수들을 특징벡터로서 추출하는 단계, 및 (g) 상기 특징벡터를 이용하여 분류기를 학습하는 단계를 포함할 수 있다. 여기서, n은 자연수일 수 있다.A motor image classification method according to an embodiment of the present invention is a motor image classification method performed by a motor image classification device, wherein (a) noise of EEG signals measured from a plurality of electrodes for each class to be classified Removing by applying an adaptive noise cancellation algorithm and obtaining preprocessed EEG signals, (b) rotating the preprocessed EEG signals (

Obtaining rotational EEG signals by rotating using a rotation matrix having ), (c) the rotation angle (

) Calculating correlation coefficients between the rotational EEG signals, (d) calculating Fisher`s ratio for correlation coefficients between the rotational EEG signals, (e) the rotational EEG signals Optimizing correlation coefficients between the two to obtain optimal correlation coefficients, (f) extracting n optimal correlation coefficients as feature vectors in the order of the highest Fisher ratio among the optimal correlation coefficients, and (g) the feature vector It may include the step of learning the classifier using. Here, n may be a natural number.

)를 갖는 회전행렬을 이용하여 회전시켜 회전 뇌전도 신호들을 얻는 회전부, 상기 회전각도(

)별로 상기 회전 뇌전도 신호들 간에 상관계수들을 산출하고, 상기 회전 뇌전도 신호들 간에 상관계수들에 대한 피셔 비율(Fisher`s ratio)을 산출하는 산출부, 상기 회전 뇌전도 신호들 간에 상관계수들을 최적화시켜 최적 상관계수들을 얻고, 상기 최적 상관계수들 중에서 상기 피셔 비율이 높은 순서대로 n개의 최적 상관계수들을 특징벡터로서 추출하는 특징 추출부, 및 상기 특징벡터를 이용하여 분류기를 학습하는 분류부를 포함할 수 있다. 여기서, n은 자연수일 수 있다.The motor image classification apparatus according to an embodiment of the present invention removes noise of EEG signals measured from a plurality of electrodes for each class to be classified by applying an adaptive noise cancellation algorithm, and preprocessed EEG. A preprocessor for obtaining signals, the preprocessed EEG signals at a rotation angle (

A rotating part that obtains rotating EEG signals by rotating using a rotating matrix having ), the rotation angle (

) By calculating the correlation coefficients between the rotational EEG signals, a calculation unit that calculates a Fisher's ratio to the correlation coefficients between the rotational EEG signals, by optimizing the correlation coefficients between the rotational EEG signals A feature extraction unit that obtains optimal correlation coefficients and extracts n optimal correlation coefficients as a feature vector in the order of the highest Fisher ratio among the optimal correlation coefficients, and a classification unit that learns a classifier using the feature vector. have. Here, n may be a natural number.

According to an embodiment of the present invention, interference and noise of an EEG signal can be effectively removed.

In addition, it is possible to improve the classification accuracy by optimizing the correlation between EEG signals and using a correlation with a higher possibility of classification for classification of motor images.

1 is a diagram for describing a blind source separation (BSS) technique.
2 is a block diagram of an apparatus for classifying a motion image according to an embodiment of the present invention.
3 is a flowchart of a method for classifying a motion image according to an embodiment of the present invention.
4 is a flowchart of a pre-processing step included in FIG. 3.

Specific structural or functional descriptions of the embodiments according to the concept of the present invention disclosed in this specification are exemplified only for the purpose of describing the embodiments according to the concept of the present invention, and embodiments according to the concept of the present invention They may be implemented in various forms and are not limited to the embodiments described herein.

Since the embodiments according to the concept of the present invention can apply various changes and have various forms, the embodiments will be illustrated in the drawings and described in detail herein. However, this is not intended to limit the embodiments according to the concept of the present invention to specific disclosed forms, and includes changes, equivalents, or substitutes included in the spirit and scope of the present invention.

Terms such as first or second may be used to describe various elements, but the elements should not be limited by the terms. The above terms are only for the purpose of distinguishing one component from other components, for example, without departing from the scope of rights according to the concept of the present invention, the first component may be named as the second component, Similarly, the second component may also be referred to as a first component.

When a component is referred to as being "connected" or "connected" to another component, it is understood that it may be directly connected or connected to the other component, but other components may exist in the middle. Should be. On the other hand, when a component is referred to as being "directly connected" or "directly connected" to another component, it should be understood that there is no other component in the middle. Expressions describing the relationship between components, for example, "between" and "just between" or "directly adjacent to" should be interpreted as well.

The terms used in the present specification are only used to describe specific embodiments and are not intended to limit the present invention. Singular expressions include plural expressions unless the context clearly indicates otherwise. In the present specification, terms such as "comprise" or "have" are intended to designate that the specified features, numbers, steps, actions, components, parts, or combinations thereof exist, but one or more other features or numbers, It is to be understood that the presence or addition of steps, actions, components, parts, or combinations thereof, does not preclude the possibility of preliminary exclusion.

Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by a person of ordinary skill in the art to which the present invention belongs. Terms as defined in a commonly used dictionary should be interpreted as having a meaning consistent with the meaning in the context of the related technology, and should not be interpreted as an ideal or excessively formal meaning unless explicitly defined in this specification. Does not. Hereinafter, embodiments will be described in detail with reference to the accompanying drawings.

1 is a diagram for describing a blind source separation (BSS) technique.

Referring to FIG. 1, the BSS technique is a technique that has appeared to solve the point in which an EEG signal is mixed with various intentions of an internal signal source. More specifically, the BSS technique selectively removes internal signal sources corresponding to interference such as blinking or other intentions by inversely estimating signals of internal brain signal sources using EEG signals measured by electrodes. Accordingly, it is possible to construct an EEG signal specialized for the intention of the subject by estimating and removing signals of unnecessary internal signal sources in the classification of motor images.

However, this BSS technique has a problem in that it is difficult to select a signal source inside the brain that responds to accurate interference and noise because the judgment on the role of the signal sources inside the brain is made empirically. In addition, since the EEG signal measured through the electrode attached to the scalp of the subject is used, the SNR is low.

On the other hand, in order to solve the problem caused by the low SNR, there is a connectivity method using the correlation of EEG signals between two electrodes. Correlation is known to be robust against noise as it can remove noise that occurs independently of each other.

Existing connectivity methods using correlation focus on selecting the correlation that has the most effective information for intention recognition from among the correlations between the given connectivity, that is, the brain electrode signals. As one of these connectivity methods, using the t-statistic method, which is a statistical verification method using the given correlations, the degree to which the correlation is distinguishable for two intentions is quantified and the correlation with the highest value is selected. There is a technique. However, since it does not change the given correlations, there is a limit to choosing the correlation that is most likely to be classified within it.

Accordingly, the present invention focuses on recognition of intentions, pre-processing the EEG signal so that the classification probability is highest, and changing the correlation through a rotation matrix to obtain higher classification accuracy than the above-described conventional methods. Let's do it.

2 is a block diagram of an apparatus for classifying a motion image according to an embodiment of the present invention.

2, the motion image classification apparatus 10 according to an embodiment of the present invention includes a preprocessor 100, a rotation unit 200, a calculation unit 300, a feature extraction unit 400, and a classification unit 500. ).

The preprocessor 100 preprocesses the EEG signals measured for each class to be classified from a plurality of electrodes attached to the subject using an adaptive noise cancellation algorithm (hereinafter, ANC algorithm), and preprocesses EEG signals. Get the signals Here, the EEG signal measured from the plurality of electrodes attached to the subject is an electrical signal generated according to a motor image to be classified, and may be expressed as a matrix having a size of C x T, for example. At this time, C is the number of electrodes of the EEG signal attached to the subject, that is, the number of channels, and T is the number of samples per time (here, C and T are natural numbers).

The ANC algorithm used in preprocessing is an algorithm that cancels noise or interference signals by using the principle of wave superposition. The ANC algorithm continuously changes (or adapts) the system so that the signal processing is optimal under a specific criterion, and to this end, an adaptive filter with a function of changing the characteristics of the system is used.

Such an adaptive filter may include various techniques such as a least squares average (LMS) algorithm, a filtered-X LMS (FXLMS) algorithm, or a filtered-error LMS (FELMS) algorithm to minimize an error signal. . Meanwhile, the adaptive filter used in the present invention may preferably be an LMS algorithm having a small amount of calculation per step and stably converging.

Hereinafter, a process in which the preprocessor 100 preprocesses EEG signals using an ANC algorithm will be described in detail.

The EEG signals measured from the plurality of electrodes are signals in which the original signal and the noise signal are mixed, and can be expressed as Equation 1 below.

Here, x is the EEG signal, s is the original signal, and n is the noise signal.

The noise signal n is a reference signal (or reference signal) in the ANC algorithm and is a signal to be estimated and removed. In the ANC algorithm, the selection of the reference signal is an important factor because the initial setting has a high influence on the performance of the algorithm.

To this end, the preprocessor 100 may further include a setting unit (not shown).

The setting unit (not shown) selects reference signals based on the correlation coefficient. More specifically, the setting unit (not shown) first calculates correlation coefficients between the measured EEG signals. The correlation coefficient is a unit representing the degree of a linear relationship between two variables, such as Spearman correlation coefficient, Cronbach's alpha, and Pearson correlation coefficient. Hereinafter, the term “correlation coefficient” used in the present specification should be understood to refer to a term having the above-described meaning.

As an example, in the case of calculating the Pearson correlation coefficient, the setting unit (not shown) is used between the EEG signals measured by Equation 2 below (more specifically, measured from any two electrodes among a plurality of electrodes). The correlation coefficient between EEG signals) is calculated.

는 i번째 전극과 j번째 전극에 대한 피어슨 상관계수,

는 i번째 전극과 j번째 전극의 뇌전도 신호들 간에 공분산,

는 i번째 전극과 j번째 전극의 뇌전도 신호들의 표준편차의 곱이다.here,

Is the Pearson correlation coefficient for the ith electrode and the jth electrode,

Is the covariance between the EEG signals of the ith electrode and the jth electrode,

Is the product of the standard deviation of the EEG signals of the ith electrode and the jth electrode.

When the correlation coefficients are calculated between all the electrodes attached to the subject, the setting unit (not shown) selects the EEG signals having correlation coefficients between EEG signals less than a preset threshold among the correlation coefficients between the calculated EEG signals as a reference signal. Choose with That is, if the correlation coefficient between the EEG signals is low, it is determined as an interference or noise signal irrelevant to classification, and this is set as a reference signal. Here, the threshold value for setting the reference signal may be set differently according to the type of motion image to be classified or the subject.

Accordingly, the present invention can further improve the performance of the ANC algorithm by setting EEG signals having a low correlation coefficient as a reference signal through a setting unit (not shown), rather than empirically setting noise like conventional techniques. .

When the reference signals are selected, the preprocessor 100 estimates an accurate reference signal using the ANC algorithm and removes noise. When adaptive filtering is performed on M reference signals (M is a natural number) of discrete signal sections, the filtered result is as Equation 3 below.

Here, y(t) is the filtering result value, n(t) is the reference signal, and w(t) is the filter coefficient.

Therefore, the result value of the ANC algorithm is shown in Equation 4 below.

Here, e is the result of the ANC algorithm.

The ANC algorithm continuously updates the coefficient w of the filter until it estimates the correct reference signal n according to the result e. Various algorithms for the above-described adaptive filter can be used for updating.

As an example, in the case of using the LMS algorithm, the update of the filter coefficient w may be performed by Equation 5 below.

Here, α is the size of the step, and the algorithm can be set to have an appropriate adaptation speed and stability.

The LMS algorithm updates the filter coefficient according to Equation 5 above, but updates the ANC algorithm so that the result value of the ANC algorithm (ie, the error signal) is the minimum squared mean value of e.

That is, the preprocessor 100 obtains preprocessed EEG signals by accurately estimating and removing noise while updating the adaptive filter for EEG signals and reference signals.

)를 갖는 회전행렬을 이용하여 회전시켜 회전 뇌전도 신호들을 얻는다. 회전행렬은 임의의 신호를 원점을 중심으로 회전시키며, 회전행렬에 의하여 회전된 회전 뇌전도 신호들은 아래의 수학식 6과 같이 나타낼 수 있다.The rotation unit 200 rotates the EEG signals preprocessed through the preprocessor 100 at a rotation angle (

Using a rotation matrix having ), rotation EEG signals are obtained. The rotation matrix rotates an arbitrary signal around the origin, and the rotational EEG signals rotated by the rotation matrix can be expressed as Equation 6 below.

는 회전 뇌전도 신호,

는 뇌전도 신호이다. 한편, 회전각도(

)는 기 설정된 각도 구간을 일정 크기의 단위 각도로 쪼개었을 때의 각각의 경계에 해당하는 각도일 수 있다. 이에 따라, 후술할 산출부(300)에서 산출되는 상관계수들 및 피셔 비율은 상기 쪼개진 회전각도(

)별로 산출된다.here,

Is a rotating EEG signal,

Is an electroencephalogram signal. Meanwhile, the rotation angle (

) May be an angle corresponding to each boundary when a preset angle section is divided into a unit angle of a predetermined size. Accordingly, the correlation coefficients and the Fisher ratio calculated by the calculation unit 300 to be described later are the split rotation angle (

) Is calculated by

Unlike conventional techniques using fixed correlation coefficients for EEG signals by changing the correlation coefficients according to the rotation matrix, the present invention can also use the correlation coefficient of the rotation EEG signals.

)별로 산출하고, 상기 산출된 회전 뇌전도 신호들 간에 상관계수들에 대한 피셔 비율(Fisher`s ratio)을 산출한다. 회전 뇌전도 신호들 간에 상관계수들을 산출하는 것은 앞서 상술한 상관계수 산출에서 뇌전도 신호 대신 회전 뇌전도 신호를 기반으로 산출되는 것이므로, 이에 대한 상세한 설명은 생략하기로 한다.The calculation unit 300 calculates the correlation coefficients between the rotational EEG signals obtained through the rotation unit 200 at a rotation angle (

), and a Fisher`s ratio for correlation coefficients between the calculated rotational EEG signals. Since the calculation of the correlation coefficients between the rotational EEG signals is calculated based on the rotational EEG signal instead of the EEG signal in the above-described correlation coefficient calculation, a detailed description thereof will be omitted.

Fisher's ratio is used when evaluating the classification ability of a feature for classification in a linear discrimination technique for reducing the number of dimensions, and can be expressed as Equation 7 below.

)별로 회전 뇌전도 신호들 간에 상관계수들에 대한 피셔 비율을 산출할 수 있다.Here, F is the Fisher ratio, m ₁ and m ₂ are the averages of the correlation coefficients for classes ₁ and ₂ to be classified, and v ₁ and v ₂ are the variances of the correlation coefficients for classes ₁ and ₂ respectively. Meanwhile, the average and variance of the correlation coefficients are calculated using the EEG signals measured for each cycle for each combination of electrodes. Using Equation 7, the calculation unit 300 is the rotation angle of the rotation matrix (

), it is possible to calculate the Fisher ratio for the correlation coefficients between the rotating EEG signals.

The classification ability of correlation coefficients to be used as features may be evaluated by using the Fisher ratio calculated through the calculation unit 300 as described above. In other words, it can be seen that the higher the Fisher ratio, the higher the classifiability of the correlation coefficient.

)별로 산출하고, 산출된 상관계수들에 대한 피셔 비율을 회전각도(

)별로 산출할 수 있다.As a result, the correlation coefficient of the two electrodes selected from among the plurality of electrodes attached to the subject through the calculation unit 300 of the present invention is determined by the rotation angle (

), and calculate the Fisher ratio for the calculated correlation coefficients by rotation angle (

) Can be calculated.

The feature extraction unit 400 optimizes the correlation coefficients between the rotational EEG signals calculated through the calculation unit 300 to obtain optimal correlation coefficients, and features n optimal correlation coefficients in the order of the highest Fisher ratio among the optimum correlation coefficients. Extract as a vector.

)별로 산출된 상관계수들에 대한 피셔 비율들 중에서 가장 높은 피셔 비율일 때의 회전각도(

)와 그때의 상관계수를 최적 상관계수로서 추출한다. 이러한 과정을 모든 전극들의 조합들에 대하여 수행하여 상관계수들을 최적화시키고 최적 상관계수들을 얻는다.First, the feature extraction unit 400 has a rotation angle for each combination of specific electrodes (

Rotation angle at the highest Fisher ratio among the Fisher ratios to the correlation coefficients calculated by) (

) And the correlation coefficient at that time are extracted as the optimal correlation coefficient. This process is performed for all combinations of electrodes to optimize correlation coefficients and obtain optimal correlation coefficients.

Next, among the extracted optimal correlation coefficients, n correlation coefficients are extracted as feature vectors in the order of the highest Fisher ratio (where n is a natural number).

The classification unit 500 learns a classifier using the feature vector extracted by the feature extraction unit 400. The training data for learning are EEG signals measured from a plurality of electrodes, and the classifier is trained by featuring optimal correlation coefficients extracted by the feature extraction unit 400 for the training data. In this case, the classifier may be based on a Support Vector Machine (SVM).

3 is a flowchart of a method for classifying a motion image according to an embodiment of the present invention. Hereinafter, detailed descriptions of portions overlapping with the previously described portions will be omitted.

3, a method for classifying a motion image according to an embodiment of the present invention includes a pre-processing step (S610), a rotation step (S620), a correlation coefficient calculation step (S630), a Fisher ratio calculation step (S640), and an optimization step ( S650), a feature vector extraction step (S660), and a learning step (S670).

In step S610, noise of EEG signals measured from a plurality of electrodes for each class to be classified is removed by applying an ANC algorithm, and preprocessed EEG signals are obtained.

Referring to FIG. 4, which is a flowchart of the pre-processing step included in FIG. 3, step S610 may further include steps S611 to S613 for pre-processing.

First, in step S611, correlation coefficients are calculated between the measured EEG signals.

Next, in step S612, among the correlation coefficients between the EEG signals calculated through step S611, EEG signals having correlation coefficients between EEG signals less than a preset threshold are selected as reference signals in the ANC algorithm.

Finally, in step S613, preprocessed EEG signals are obtained while updating the adaptive filter for EEG signals and reference signals. Here, the update may be performed until the squared average value is minimized.

)를 갖는 회전행렬을 이용하여 회전시켜 회전 뇌전도 신호들을 얻는다. S620 단계는 기 설정된 각도 구간을 일정 크기의 단위 각도로 쪼개었을 때의 각각의 경계에 해당하는 모든 회전각도(

)에 대하여 수행될 수 있다.Step S620 is the rotation angle of the EEG signals preprocessed through step S610 (

Using a rotation matrix having ), rotation EEG signals are obtained. In step S620, all rotation angles corresponding to each boundary when the preset angle section is divided into a unit angle of a certain size (

) Can be performed.

)별로 회전 뇌전도 신호들 간에 상관계수들을 산출한다. S611 단계와 달리, S630 단계에서는 뇌전도 신호가 아닌 회전 뇌전도 신호들 간에 상관계수들을 산출한다.Step S630 is the rotation angle (

) Calculate correlation coefficients between the rotating EEG signals. Unlike in step S611, in step S630, correlation coefficients are calculated between rotating EEG signals, not EEG signals.

In step S640, a Fisher ratio for correlation coefficients between the rotating EEG signals calculated through step S630 is calculated. The Fisher ratio is calculated by Equation 7.

)별로 산출된 상관계수들에 대한 피셔 비율들 중에서 가장 높은 피셔 비율일 때의 회전각도(

)와 그때의 상관계수를 최적 상관계수로서 추출하는 과정을 모든 전극들의 조합들에 대하여 수행함으로써 최적 상관계수들을 얻는다.In step S650, optimal correlation coefficients are obtained by optimizing correlation coefficients between the rotating EEG signals calculated through step S630. More specifically, step S650 is the rotation angle for each of the combinations of specific electrodes (

Rotation angle at the highest Fisher ratio among the Fisher ratios to the correlation coefficients calculated by) (

) And the correlation coefficient at that time as the optimum correlation coefficient are performed for all combinations of electrodes to obtain optimum correlation coefficients.

Meanwhile, steps S620 to S650 described above may be performed for all combinations of two electrodes selected from a plurality of electrodes.

In step S660, n optimal correlation coefficients are extracted as feature vectors in the order of the highest Fisher ratio among the optimal correlation coefficients.

In step S670, the classifier is trained using the feature vector extracted through step S660. Here, the classifier may be an SVM-based classifier.

<Example>

Hereinafter, an exemplary embodiment of the apparatus 10 and method for classifying a motion image of the present invention and the performance of classifying a motion image according to the same will be compared with other techniques. According to an embodiment, the EEG signal data used for classification of motor images are BCI competition Ⅱ datasets IV and Ib.

First, dataset IV is data that the subject predicts and inputs future behavior according to the criteria set by himself. The dataset IV data was collected through a total of 416 motor image signal measurements from 28 electrodes attached to the subject, 316 as training signals and 100 as test signals. The classifier used the LS-SVM classifier, which is most commonly used for BCI classification. Meanwhile, the threshold for selecting the reference signal was set to 0.7. In this case, the classification accuracy between the present invention and the existing techniques is shown in Table 1 below.

Classification accuracy[%] The present invention 88 Connectivity technique 84 Winning technique 84

Referring to Table 1, it can be seen that the present invention exhibits a classification accuracy of 88 [%], whereas the conventional techniques exhibit a classification accuracy of 84 [%].

Next, dataset Ib is the cognitive image data of subjects with amyotrophic lateral sclerosis (ALS). The dataset Ib data was collected through a total of 300 motor image signal measurements from 6 electrodes attached to the subject, 200 were used as training signals and 100 were used as test signals. The classifier and threshold were set the same as for dataset IV. In this case, the classification accuracy between the present invention and the existing techniques is shown in Table 2 below.

Classification accuracy[%] The present invention 62.78 Connectivity technique 57.22 Winning technique 54.4

Referring to Table 2, it can be seen that the present invention exhibits a classification accuracy of 62.78 [%], whereas the conventional techniques show classification accuracy of 57.22 [%] and 54.4 [%], respectively.

As described above, the motion image classification apparatus 10 and method of the present invention can obtain a strong correlation to noise by using active noise control, and the correlation is optimized using a rotation matrix and improved by using this as a feature vector. It can have classification accuracy.

The apparatus described above may be implemented as a hardware component, a software component, and/or a combination of a hardware component and a software component. For example, the devices and components described in the embodiments include, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable array (FPA), It can be implemented using one or more general purpose computers or special purpose computers, such as a programmable logic unit (PLU), a microprocessor, or any other device capable of executing and responding to instructions. The processing device may execute an operating system (OS) and one or more software applications executed on the operating system. In addition, the processing device may access, store, manipulate, process, and generate data in response to the execution of software. For the convenience of understanding, although it is sometimes described that one processing device is used, one of ordinary skill in the art, the processing device is a plurality of processing elements and/or a plurality of types of processing elements. It can be seen that it may include. For example, the processing device may include a plurality of processors or one processor and one controller. In addition, other processing configurations are possible, such as a parallel processor.

The software may include a computer program, code, instructions, or a combination of one or more of these, configuring the processing unit to behave as desired or processed independently or collectively. You can command the device. Software and/or data may be interpreted by a processing device or to provide instructions or data to a processing device, of any type of machine, component, physical device, virtual equipment, computer storage medium or device. , Or may be permanently or temporarily embodyed in a transmitted signal wave. The software may be distributed over networked computer systems and stored or executed in a distributed manner. Software and data may be stored on one or more computer-readable recording media.

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, and the like alone or in combination. The program instructions recorded on the medium may be specially designed and configured for the embodiment, or may be known and usable to those skilled in computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floptical disks. -A hardware device specially configured to store and execute program instructions such as magneto-optical media, and ROM, RAM, flash memory, and the like. Examples of the program instructions include not only machine language codes such as those produced by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like. The hardware device described above may be configured to operate as one or more software modules to perform the operation of the embodiment, and vice versa.

As described above, although the embodiments have been described by the limited drawings, various modifications and variations are possible from the above description to those of ordinary skill in the art. For example, the described techniques are performed in a different order from the described method, and/or components such as a system, structure, device, circuit, etc. described are combined or combined in a form different from the described method, or other components Alternatively, even if substituted or substituted by an equivalent, an appropriate result can be achieved.

Therefore, other implementations, other embodiments, and claims and equivalents fall within the scope of the claims to be described later.

Claims (10)

As a method for classifying a moving image performed by a moving image classification device,
(a) removing noise of EEG signals measured from a plurality of electrodes for each class to be classified by applying an adaptive noise cancellation algorithm and obtaining preprocessed EEG signals;
(b) rotation angle of the preprocessed EEG signals (

Obtaining rotational EEG signals by rotating using a rotation matrix having );
(c) the rotation angle (

) Calculating correlation coefficients between the rotational EEG signals;
(d) calculating a Fisher`s ratio for correlation coefficients between the rotating EEG signals;
(e) obtaining optimal correlation coefficients by optimizing correlation coefficients between the rotating EEG signals;
(f) extracting n optimal correlation coefficients as feature vectors in the order of the highest Fisher ratio among the optimal correlation coefficients; And
(g) learning a classifier using the feature vector,
The (e) step is the rotation angle (

Extracting a correlation coefficient at the highest Fisher ratio among the correlation coefficients between the rotational EEG signals calculated for each) as an optimal correlation coefficient,
How to classify motion images.
Where n is a natural number.
The method of claim 1,
Step (a)
(a-1) calculating correlation coefficients between the EEG signals;
(a-2) selecting EEG signals having correlation coefficients between EEG signals less than a preset threshold among the correlation coefficients between the EEG signals as reference signals; And
(a-3) A motor image classification method comprising the step of obtaining the pre-processed EEG signals while updating an adaptive filter for the EEG signals and reference signals.
The method of claim 1,
The correlation coefficient is a Pearson correlation coefficient.
The method of claim 1,
The Fisher ratio is a method of classifying motion images defined by Equation 1 below.
[Equation 1]

Where F is the Fisher ratio, m ₁ and m ₂ are the averages of the correlation coefficients for class 1,2, respectively, and v ₁ and v ₂ are the variances of the correlation coefficients for class 1,2, respectively.
delete The method of claim 1,
Steps (b) to (e) are performed on all combinations of two electrodes selected from the plurality of electrodes.
The method of claim 1,
The classifier of step (g) is a motion image classification method based on SVM (Support Vector Machine).
A preprocessor for removing noise of EEG signals measured from a plurality of electrodes for each class to be classified by applying an adaptive noise cancellation algorithm and obtaining preprocessed EEG signals;
The preprocessed EEG signals are rotated at an angle (

) By using a rotation matrix having a rotation unit to obtain rotation EEG signals;
The rotation angle (

) A calculation unit that calculates correlation coefficients between the rotational EEG signals, and calculates a Fisher`s ratio with respect to the correlation coefficients between the rotational EEG signals;
A feature extraction unit for optimizing correlation coefficients between the rotational EEG signals to obtain optimum correlation coefficients, and extracting n optimum correlation coefficients as feature vectors in the order of the highest Fisher ratio among the optimum correlation coefficients; And
A classification unit for learning a classifier using the feature vector,
The feature extraction unit is the rotation angle (

Extracting a correlation coefficient at the highest Fisher ratio among the correlation coefficients between the rotational EEG signals calculated for each) as an optimal correlation coefficient,
Movement image classification device.
Where n is a natural number.
The method of claim 8,
The pretreatment unit
Further comprising a setting unit for calculating correlation coefficients between the EEG signals and selecting EEG signals having correlation coefficients between EEG signals less than a preset threshold among the correlation coefficients between the EEG signals as reference signals,
A motor image classification apparatus for obtaining the preprocessed EEG signals while updating an adaptive filter for the EEG signals and reference signals.
The method of claim 8,
The Fisher ratio is a motion image classification device defined by Equation 1 below.
[Equation 1]

Where F is the Fisher ratio, m ₁ and m ₂ are the averages of the correlation coefficients for class 1,2, respectively, and v ₁ and v ₂ are the variances of the correlation coefficients for class 1,2, respectively.

Images