KR101619973B1 - Motor imagery eeg classification using spectrum analysis and vector quantization - Google Patents

Abstract

The present invention relates to a system and a discrimination method for extracting characteristics of a time-frequency domain through spectral analysis of an electroencephalogram (EEG) signal and classifying signals by vector quantization, The method comprising the steps of: receiving an electroencephalogram (EEG) signal of a user from an electroencephalogram receiving unit; filtering the electroencephalogram signal received by the electroencephalogram filtering unit with a frequency band of a specific region; Performing a vector quantization using a code book dictionary with respect to the extracted feature vector, and selecting a class of the feature vector from the feature classifier, In the step of extracting the feature vector, the extracted feature vector is converted into a time- It characterized in that it comprises the features womb.

Description

TECHNICAL FIELD [0001] The present invention relates to a motion estimation method and a discrimination system for motion imagery using spectral analysis and vector quantization,

More particularly, the present invention relates to a system and method for identifying a motion imaginary EEG, and more particularly, to a method and apparatus for extracting characteristics of a time-frequency domain from an electroencephalogram (EEG) signal through spectral analysis, And to a system and a discrimination method capable of discriminating the motion imaginary brain conduction.

In general, HCI is an abbreviation of Human Computer Interaction. It deals with the interaction between human and computer. It deals with the process of designing and evaluating the operating system so that human and computer can interact easily and easily.

Traditional HCI can be viewed as three components: individual, computer, and interaction. It is a field in which a person can develop a system that can easily and conveniently perform his / her tasks using a machine called a computer.

BCI (Brain Computer Interface) means communication between a human brain and a computer. BCI can be measured using the EEG signal generated between the nervous system and the cranial nervous system.

BCI provides a new interface means to those who can not use part of the body because it provides the intention of the person through the brain signal without movement of the body, and it can improve the quality of life.

When control is attempted using BCI, a specific pattern of EEG signal is generated according to human intention, and meaningful information can be extracted from this signal using the signal processing algorithm of BCI system.

In addition, a technology for selecting a meaningful few number of optimal channels from a multi-channel EEG signal through a CFE (Channel-Frequency-ERSP) map using Event-Related Spectral Perturbation (ERSP) is disclosed.

In addition, as a non-homogeneous spatial filter optimization method for each time-frequency space in consideration of non-stationary characteristics of the EEG signal with time, a spatial filter using CSP (Common Spatial Pattern) A technique for obtaining time-frequency segments is disclosed.

Also, a technique for an Invariant Common Spatio-Spectral Pattern (iCSSP) in which a CSP spatial filter is improved is also disclosed.

However, conventional techniques have a disadvantage in that a complicated signal processing process is required to discriminate the EEG.

- Seokhong Il, Lee Sung Hwan, "EEG Channel Selection for Motion Imagery Classification," Korea Information Science Society 2010 Vol. 37, No. 2 (C), pp.195-198, 2010 - Kwon Tae Hwan and Lee Sung Hwan, "Motion Estimation EEG Signal Classification Based on Incompatible Spatial Filter Optimization," Korean Information Science Society, Vol. 38, No. 1, pp. 69-47, 2011 - Jho Hyun, An Min Kyu, Ahn Sang Tae, and Jeong Sang Chan, "A Study on the Performance Comparison of EEG Feature Extraction Algorithm for Brain-Computer Interface Based on EEG", Journal of KISS: Software and Applications Volume 38, Issue 11, pp.599-605, 2011

Disclosure of Invention Technical Problem [8] The present invention has been conceived to solve the problems described above, and it is an object of the present invention to provide an apparatus and method for classifying an EEG signal through classification of time- frequency domain features and vector quantization through spectral analysis, System and a discrimination method using the same.

However, the object of the present invention is not limited to the above-mentioned objects, and other objects not mentioned can be clearly understood by those skilled in the art from the following description.

In order to achieve the above object, a method of discriminating a motion imaginary brain image using spectral analysis and vector quantization according to an embodiment of the present invention comprises the steps of: receiving an electroencephalogram (EEG) signal of a user from an electroencephalogram receiver; Extracting a feature vector from the EEG signal filtered by the feature vector extracting unit; and using a codebook dictionary for the feature vector extracted by the vector quantization unit, And a step of selecting a class of the quantized feature vector, wherein in the step of extracting the feature vector, the extracted feature vector has a feature of time-space type .

In addition, the method of discriminating motion imaginary EEG using spectral analysis and vector quantization according to the present invention is characterized in that, in the filtering step, the EEG filtering unit filters the EEG signal so that only signals in the 8 to 30 Hz band pass.

In addition, the motion imaginary brain image discrimination method using spectrum analysis and vector quantization according to the present invention is characterized in that the brain image filtering section includes a Butterworth filter.

In addition, in the method of discriminating motion imaginary EEG using spectral analysis and vector quantization according to the present invention, in the step of extracting the feature vector, the feature vector extracting unit extracts the feature vector using a STFT (Short-Time Fourier Transform) .

The method of discriminating motion imaginary EEG using spectral analysis and vector quantization according to the present invention is characterized in that in the vector quantization step, the vector quantization unit replaces each feature vector with the closest center vector and performs vector quantization.

In addition, the method of discriminating motion imaginary EEG using spectral analysis and vector quantization according to the present invention is characterized in that the vector quantization unit selects a center vector using an Euclidean distance method including the following equation.

In addition, the method of discriminating motion imaginary EEG using spectral analysis and vector quantization according to the present invention is characterized in that the feature classifier selects the class using the following equation.

Here, vi represents the index value of the codebook dictionary corresponding to the i-th vector.

In addition, the method of discriminating motion imaginary EEG using spectral analysis and vector quantization according to the present invention comprises the steps of: extracting the feature vectors and then generating a certain number of representative feature vectors by clustering the extracted feature vectors; And the codebook generator generates the codebook dictionary using the representative feature vector, thereby forming the codebook dictionary used in the vector quantization step.

In addition, the method of discriminating the kinetic imaginary EEG using spectral analysis and vector quantization according to the present invention is characterized in that the feature vectors are clustered using the Linde-Buzo-Gray (LBG) algorithm of the clustering part.

A motion imaginary brain image discrimination system using spectral analysis and vector quantization according to an embodiment of the present invention includes an electroencephalogram (EEG) signal receiving unit for receiving an electroencephalogram (EEG) signal, an electroencephalogram A clustering unit for clustering the extracted feature vectors and generating a predetermined number of representative feature vectors; a clustering unit for clustering the extracted feature vectors to generate a predetermined number of representative feature vectors; A codebook generator for generating a codebook dictionary using the generated representative feature vectors; a vector quantizer for performing vector quantization using the codebook dictionary for the feature vector extracted by the feature vector extractor; And a feature classifying unit for selecting a class of the quantized feature vector, It characterized in that it comprises.

According to the system for discriminating motion imaginary EEG using spectral analysis and vector quantization according to the present invention and the discrimination method using the same, time-frequency domain characteristics are extracted through spectrum analysis using a short-time Fourier transform (STFT) algorithm and LBG The Linde-Buzo-Gray algorithm can be applied to vector quantization.

In addition, according to the present invention, it is possible to provide a new interface means to people who can not use a part of the body without movement of the body by accurately discriminating the motion imaginary brain conduction signal.

FIG. 1 is a block diagram schematically showing a motion imaginary brain image discrimination system using spectral analysis and vector quantization according to an embodiment of the present invention.
FIG. 2 and FIG. 3 are flowcharts illustrating a method of identifying a motion imaginary brain image using spectral analysis and vector quantization according to an embodiment of the present invention.
4 is an exemplary diagram illustrating a method for determining the probability of a class of a feature vector according to the present invention.
5 is an exemplary diagram illustrating a timing structure of an electroencephalogram signal according to the present invention.
6 is an exemplary diagram illustrating exemplary performance results according to the size of a codebook according to the present invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, a detailed description of preferred embodiments of the present invention will be given with reference to the accompanying drawings. In the following description of the present invention, detailed description of known functions and configurations incorporated herein will be omitted when it may make the subject matter of the present invention rather unclear.

Embodiments in accordance with the concepts of the present invention can make various changes and have various forms, so that specific embodiments are illustrated in the drawings and described in detail in this specification or application. It is to be understood, however, that it is not intended to limit the embodiments according to the concepts of the present invention to the particular forms of disclosure, but includes all modifications, equivalents, and alternatives falling within the spirit and scope of the invention.

It is to be understood that when an element is referred to as being "connected" or "connected" to another element, it may be directly connected or connected to the other element, . On the other hand, when an element is referred to as being "directly connected" or "directly connected" to another element, it should be understood that there are no other elements in between. Other expressions that describe the relationship between components, such as "between" and "between" or "neighboring to" and "directly adjacent to" should be interpreted as well.

The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The singular expressions include plural expressions unless the context clearly dictates otherwise. In this specification, the terms "comprises ", or" having ", or the like, specify that there is a stated feature, number, step, operation, , Steps, operations, components, parts, or combinations thereof, as a matter of principle.

FIG. 1 is a block diagram schematically showing a motion imaginary brain image discrimination system using spectral analysis and vector quantization according to an embodiment of the present invention. Referring to FIG.

Referring to the drawings, a motion imaginary brain image discrimination system 100 using spectral analysis and vector quantization according to the present invention includes an electroencephalogram receiving unit 110, an electroencephalogram filtering unit 120, a feature vector extracting unit 130, a vector quantizing unit 140, a feature classifying unit 150, a clustering unit 160, and a codebook generating unit 170.

The electroencephalogram (EEG) signal measured by a user or a plurality of users is not specifically shown but can be received by wire or wireless.

The EEG filtering unit 120 includes a band-pass filter, for example, a Butterworth filter. The EEG signal received through the EEG receiving unit 110 is supplied to a frequency band of a specific region, specifically, 8 to 30 Hz It is possible to filter such that only the signal of the band is passed.

The feature vector extraction unit 130 may extract a feature vector from the filtered electrocardiogram signal through the filtering unit 120. [ Generally, when extracting features from the electroencephalogram signal, Gaussian filtering, Laplacian filtering, Fourier transform, wavelet transform, and the like can be used.

However, in the present invention, a feature vector may be extracted using a STFT (Short-Time Fourier Transform) method so as to include a time-space type feature in an extracted feature vector.

The vector quantization unit 140 performs vector quantization using a codebook dictionary configured for the feature vector extracted by the feature vector extraction unit 130. The feature classification unit 150 classifies the class of the electro- The class for the quantized feature vector can be selected.

In addition, the clustering unit 160 generates a certain number of representative feature vectors obtained by clustering a plurality of feature vectors extracted through the feature vector extracting unit 130 using an LBG (Linde-Buzo-Gray) algorithm And the codebook generator 170 can generate a codebook dictionary used in quantization in the vector quantization unit 140 using a predetermined number of representative feature vectors.

FIG. 2 and FIG. 3 are flowcharts showing a method of discriminating a motion imaginary EEG using spectral analysis and vector quantization according to the present invention. 1 to 2, a method for discriminating the motion imaginary EEG of the present invention will be described.

2, when the electroencephalogram (EEG) signal of the user is received by the electroencephalogram receiving unit 110 (S101), the electroencephalogram filtering unit 120 performs filtering on the received electroencephalogram signal in a frequency band of a specific region (S102).

In general, when performing motion imagery, the EEG signal includes Event-related Desynchronization (ERD) / Event-related Synchronization (ERS) patterns, Wave region.

In order to extract only the information of the alpha wave and the beta wave region, the EEG filtering unit 120 may filter the EEG signal so that only signals in the 8 to 30 Hz band can be passed. At this time, the electroencephalogram filtering unit 120 may be constituted by, for example, a band-pass filter using a Butterworth filter.

Next, the feature vector extracting unit 130 extracts a feature vector including a time-space type feature from the EEG signal filtered through the EEG filtering unit 120 using a STFT (Short-Time Fourier Transform) method (S103).

The STFT can perform the Fourier transform after truncating the entire signal using a window function in small time units.

In this Fourier transform, for example, a Hamming window is used as a window function, and a window size is set to 128, which is a data size of about 1 second. Since the information on both ends is lost when using the window function, an additional sequence is constructed by setting an overlap interval of 0.5 second, and a characteristic of 8-30 Hz is used.

Since the experiment data of 9 seconds is overlapped every 0.5 second, the spectral feature composed of 17 sequence data can be extracted, spectrum is obtained for each channel of C3 and C4, and a spectrum consisting of 17 vectors of 46 dimensions is constructed In addition, since the signal strength varies greatly, the MAX normalization is performed and the Fourier transform is performed by normalizing the data with a value between 0 and 1.

Thereafter, the vector quantization unit 140 performs vector quantization using the codebook dictionary with respect to the feature vector extracted through the feature vector extraction unit 130 (S104). A codebook dictionary used for vector quantization will be described later with reference to FIG.

In the vector quantization process, the vector quantization unit 140 performs vector quantization by replacing each feature vector with the nearest center vector, and uses the Euclidean distance method as a method for selecting the closest center vector. And can be expressed as Equation 1 below.

[Formula 1]

For the quantized feature vector, the feature classifier 150 selects a class for the quantized feature vector to obtain the class of the electroencephalogram signal (S105). At this time, the feature classifying unit 150 calculates the final left and right probabilities by multiplying the left and right probabilities corresponding to all the sequences by using the following Equation 2, .

[Formula 2]

Here, vi represents the index value of the codebook dictionary corresponding to the i-th vector.

FIG. 3 is a diagram for explaining generation of a codebook dictionary in the motion imaginary brain image discrimination method of the present invention. The process of receiving the EEG signal and extracting the feature vector described in steps S101 to S103 of FIG. The detailed description thereof will be omitted.

Referring to FIGS. 1 to 3, after the feature vector extraction (S103), the clustering unit 160 clusters a plurality of feature vectors of various types extracted through the feature vector extracting unit 130, (S106).

At this time, the clustering unit 160 may use an LBG (Linde-Buzo-Gray) algorithm. This LBG algorithm can be applied to various error measurement methods, and particularly, it can be applied to voice, image, and the like.

The codebook generator 170 may generate a codebook dictionary used in quantization in the vector quantization unit 140 using a predetermined number of representative feature vectors.

In this case, the codebook generator 170 may use the belonging vectors belonging to a certain number of representative feature vectors, that is, a plurality of classes of feature vectors. For example, referring to FIG. 4, the probability of the left or right of the representative feature vector can be calculated.

In the left part of FIG. 4, five vectors (black dots) are grouped, and the classes of the vectors are composed of 1 or 2, and each of the vectors consists of 2 and 3 vectors.

When calculating this as a ratio, the left probability is 40% and the right probability is 60%. In the right side of FIG. 4, classes 1 and 2 are composed of four vectors and one vector, respectively, so that the left probability is 80% and the right probability is 20%.

Therefore, the codebook generator 170 generates a codebook dictionary by setting the left and right probability values of the representative feature vectors using the values.

As described above, in the present invention, features of the time-frequency domain are extracted through spectral analysis of an electroencephalogram (EEG) signal, and signals are classified through a vector quantization process to discriminate the motion imaginary electroencephalogram.

[Experimental Example]

In order to confirm the performance of the present invention, the data provided by the Graz University of Technology was used as the motor imagery experimental data of data set III of the 2003 BCI competition II.

FIG. 5 is an exemplary diagram illustrating a timing structure of an electroencephalogram signal, and FIG. 6 is an exemplary diagram illustrating exemplary performance results according to a codebook size.

The operation imagination consists of two classes, each representing left and right, consisting of a total of 280 training data and 140 test data. As shown in FIG. 5, one experimental data shows that three signals of C3, C4, and CZ for the two-second wait signal, one second of the cross display signal, and the motion imaginary signal for the left and right arrow display for six seconds Data are provided, but in this experiment, two channels of C3 and C4 were used.

Experiments were conducted by generating a codebook using learning EEG and quantizing the test data using the generated codebook.

FIG. 6 is a graph showing the result of discrimination experiment in which the size of the codebook is applied from 8 to 1024. Depending on the size of the codebook, the common characteristics of the entire group can be enhanced or the unique characteristics of each of the signals can be enhanced.

The total number of vectors extracted from the experimental data used in the experiment is 2380, and the size of codebook 1024 is larger than that of experimental data, so that the noise or abnormal signals can be strengthened. However, the overall performance average is about 74% And it showed the highest performance of 78.57% at 8 and 32 sizes.

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 embodiments, but, on the contrary, is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims. will be. Accordingly, the true scope of the present invention should be determined by the technical idea of the appended claims.

100: Exercise imagination EEG discrimination system
110: an electroencephalogram receiving unit 120: an electroencephalogram filtering unit
130: Feature vector extraction unit 140: Vector quantization unit
150: feature classification unit 160: clustering unit
170: codebook generator

Claims (15)

A method for discriminating an electrocardiographic brain,
Receiving a user's electroencephalogram (EEG) signal;
Filtering the electroencephalogram signal received by the electroencephalogram filtering unit with a frequency band of a specific region;
Extracting a feature vector from an EEG signal filtered by the feature vector extraction unit;
Performing vector quantization using a codebook dictionary for the feature vector from which the vector quantization unit has been extracted; And
Selecting a class of the feature quantization feature quantization feature vector,
In the step of extracting the feature vector,
Wherein the extracted feature vector includes features of a time-space type.
The method according to claim 1,
In the filtering step,
Wherein the EEG filtering unit filters the EEG signal so that only a signal in a band of 8 to 30 Hz is passed through the spectrum analyzer and the vector quantization.
3. The method of claim 2,
Wherein the brain-power filtering unit includes a Butterworth filter. 2. The method of claim 1, wherein the brain-power filtering unit includes a Butterworth filter.
The method according to claim 1,
In the step of extracting the feature vector,
Wherein the feature vector extracting unit extracts the feature vector using a short-time Fourier transform (STFT) method.
The method according to claim 1,
Wherein in the vector quantization step, the vector quantization unit performs vector quantization by replacing each feature vector with the closest center vector, and the vector quantization unit performs vector quantization using the spectral analysis and the vector quantization.
6. The method of claim 5,
Wherein the vector quantization unit selects a center vector using an Euclidean distance method including the following equation: < EMI ID = 17.0 >

The method according to claim 1,
Wherein the feature classifying unit selects the class using the following equation: < RTI ID = 0.0 >#< / RTI >

(Where vi represents the index value of the codebook dictionary corresponding to the i-th vector)
The method according to claim 1,
Wherein the codebook dictionary used in the vector quantization step comprises:
After extracting each of the feature vectors,
Clustering feature vectors extracted by the clustering unit to generate a certain number of representative feature vectors; And
And generating a codebook dictionary by using a representative feature vector of the codebook generator. The method of discriminating motion imaginary brain diagram using spectrum analysis and vector quantization.
9. The method of claim 8,
Wherein the clustering unit clusters the feature vectors using an LBG (Linde-Buzo-Gray) algorithm.
An electroencephalogram (EEG) signal receiving unit for receiving an electroencephalogram (EEG) signal;
An electroencephalogram filtering unit for filtering the received electroencephalogram signal with a frequency band of a specific region;
A feature vector extractor for extracting a feature vector including a time-space type feature from the filtered EEG signal;
A clustering unit for clustering the extracted feature vectors to generate a certain number of representative feature vectors;
A codebook generator for generating a codebook dictionary using the generated representative feature vectors;
A vector quantization unit that performs vector quantization using the codebook dictionary for the feature vector extracted by the feature vector extraction unit; And
And a feature classifying unit that selects a class of the quantized feature vector. The system for discriminating a motion imaginary brain image using spectral analysis and vector quantization.
11. The method of claim 10,
Wherein the EEG filtering unit includes a Butterworth filter and filters the EEG signal so that only a signal in a band of 8 to 30 Hz is passed through the spectrum analyzer and the vector quantization.
11. The method of claim 10,
Wherein the feature vector extractor extracts the feature vector using a STFT (Short-Time Fourier Transform) method.
11. The method of claim 10,
Wherein the vector quantization unit performs vector quantization by replacing each feature vector with a nearest center vector, and then performing vector quantization on the motion vector.
14. The method of claim 13,
Wherein the vector quantization unit selects a center vector using an Euclidean distance method including the following equation: < EMI ID = 17.0 >

11. The method of claim 10,
Wherein the feature classifying unit selects a class of the quantized feature vector using the following equation: < EMI ID = 17.0 >

(Where vi represents the index value of the codebook dictionary corresponding to the i-th vector)

Images