阅读说明：本技术 一种区分脑电眨眼伪迹与额极痫样放电的智能分类方法 (Intelligent classification method for distinguishing electroencephalogram blink artifact and frontal epilepsy-like discharge ) 是由 曹九稳 王建辉 崔小南 郑润泽 蒋铁甲 高峰 于 2021-08-19 设计创作，主要内容包括：本发明公开了一种区分脑电眨眼伪迹与额极痫样放电的智能分类方法,首先对脑电EEG信号进行滤波处理与信号切割；然后对处理后的脑电EEG信号进行平滑非线性能量算子(SNEO)信号变换与变分模态提取(VME)信号变换得到SNEO数据集和VME数据集；然后对三个数据集对应的信号进行20维特征提取；再通过K-means算法进行二分类无监督聚类,构建无监督分类模型；最后通过建立的无监督分类模型实现眨眼伪迹与痫样放电信号的分类。本发明既克服包含痫样放电信号背景下眨眼伪迹检测精度低的困难,又可以解决现有模型忽略痫样放电的问题,还能够实现眨眼伪迹与额极痫样放电的精准自动化分类。(The invention discloses an intelligent classification method for distinguishing electroencephalogram blink artifact and frontal epilepsy-like discharge, which comprises the steps of firstly carrying out filtering processing and signal cutting on electroencephalogram EEG signals; then, carrying out Smooth Nonlinear Energy Operator (SNEO) signal transformation and Variational Mode Extraction (VME) signal transformation on the processed EEG signals to obtain an SNEO data set and a VME data set; then, 20-dimensional feature extraction is carried out on signals corresponding to the three data sets; carrying out two-classification unsupervised clustering through a K-means algorithm to construct an unsupervised classification model; and finally, classifying the blink artifact and the seizure discharge signal through the established unsupervised classification model. The method not only overcomes the difficulty of low detection precision of the blinking artifact under the background containing the epileptic-like discharge signal, but also can solve the problem that the epileptic-like discharge is neglected by the existing model, and can realize accurate automatic classification of the blinking artifact and the frontal epileptic-like discharge.)

1. An intelligent classification method for distinguishing electroencephalogram blink artifact and frontal epilepsy-like discharge is characterized by comprising the following steps:

step 1, carrying out filtering processing and signal cutting on an EEG signal to obtain a sample data set;

step 2, carrying out Smooth Nonlinear Energy Operator (SNEO) signal transformation and Variational Mode Extraction (VME) signal transformation on the EEG signals processed in the step 1 to obtain an SNEO data set and a VME data set;

step 3, 20-dimensional feature extraction is carried out on signals corresponding to the three data sets;

step 4, carrying out two-classification unsupervised clustering through a K-means algorithm to construct an unsupervised classification model;

and 5, classifying the blink artifact and the seizure discharge signal through the established unsupervised classification model.

2. The intelligent classification method for distinguishing the electroencephalogram blink artifact from the frontal epilepsy-like discharge according to claim 1, characterized in that the specific steps of the step 1 are as follows:

filtering original EEG signals of 21 channels and 1000 Hz; each channel respectively passes through a 50Hz notch filter and a 0.5-30Hz band-pass filter to obtain a filtering signal; because the 1 st channel (Fp1 channel) and the 2 nd channel (Fp2 channel) are mainly influenced by the blink artifact and the epilepsy-like discharge signal, the subsequent electroencephalogram signal only uses the first two channels; because the duration time of the blink artifact potential signal and the duration time of the frontal epilepsy-like discharge signal are both within 1s, the filtering data are cut by adopting a sliding window method, the window size is set to be 1s, and the step length is 0.5 s; and obtaining cutting signals, wherein all the cutting signals form a complete sample data set, and the number of the signals of the sample data set is assumed to be T.

3. The intelligent classification method for distinguishing the electroencephalogram blink artifact from the frontal epilepsy-like discharge according to claim 2, characterized in that the step 2 comprises the following specific steps:

assume that each signal in the sample data set is S_2*1000，S_i(i ═ 1, 2) denotes all data of the ith channel, and N denotes a data length;

2-1 calculating the ith channel S of all signals of the sample data set by the following formula_iCorresponding SNEO signal of (1), denoted as ES_i：

ES_i(t)＝ω(t)*(S_i(t)*S_i(t)-S_i(t-1)*S_i(t+1)) (1)

In which ω (t) denotes a window function, ES_i(t) represents ES_iThe t-th data of (1); then, all SNEO signals form a complete SNEO data set, and the number of the signals in the SNEO data set is T;

2-2 calculating the ith channel S of all signals of the sample data set by the following formula_iCorresponding VME signal, denoted VS_i：

Where a represents the compaction factor, ω_dRepresenting an approximation of the center frequency of the desired mode, m representing an iteration coefficient, λ representing a rise coefficient, VS_i(t) represents VS_iT data of (VS)_i(0) Is 0; and then all VME signals form a complete VME data set, and the number of the signals in the VME data set is T.

4. The intelligent classification method for distinguishing the electroencephalogram blink artifact from the frontal epileptic-like discharge according to claim 3, characterized in that the specific steps of the step 3 are as follows:

3-1, calculating the multi-dimensional multi-channel characteristics of the electroencephalogram sample, wherein the total 20-dimensional characteristics mainly comprise:

1) calculating monotonically increasing fractional slope maxima MSMI1 and MSMI2 for channel 1(Fp1 channel) and channel 2(Fp2 channel), respectively, and monotonically increasing fractional slope counts SCMI1 and SCMI2 that exceed a given threshold for each signal in the sample data set;

2) calculating monotonically increasing fractional slope maxima MSSN1 and MSSN2, respectively, monotonically increasing fractional slope counts SCSN1 and SCSN2 that exceed a given threshold for channel 1(Fp1 channel) and channel 2(Fp2 channel) for each signal in the SNEO signal dataset;

3) calculating first difference absolute averages AAF1 and AAF2, first difference absolute maximums MAF1 and MAF2, second difference absolute averages AAS1 and AAS2, second difference absolute maximums MAS1 and MAS2, monotonically increasing partial slope averages ASMI and ASMI2, and monotonically increasing partial slope maximums MSVM1 and MSVM2, respectively, for channel 1(Fp1 channel) and channel 2(Fp2 channel) of each signal in the VME data set;

3-2 calculating the slope g of the monotonically increasing portion of the ith channel of the signal in the sample data set by the following formula:

scv represents a signal monotonically increasing fractional count value;

3-3 utilizing step 3-2 to calculate the monotonically increasing partial slope sequences of channel 1 and channel 2 of the sample data set signal as SMIO1 and SMIO2, respectively, the monotonically increasing partial slope sequences of channel 1 and channel 2 of the SNEO data set signal as SMIS1 and SMIS2, respectively, and the monotonically increasing partial slope sequences of channel 1 and channel 2 of the VME data set signal as SMIV1 and SMIV2, respectively;

3-4 calculating the monotonically increasing fractional slope maxima MSMI1 and MSMI2 of the sample data set signal by the following formula:

MSMI1＝max(SMIO1)，MSMI2＝max(SMIO2) (4)

where max () denotes taking the maximum value;

3-5 calculate the monotonically increasing fractional slope counts SCMI1 and SCMI2 of the sample data set signal that exceed a given threshold by:

SCMI1＝count(SMIO1＞k1)，SCMI2＝count(SMIO2＞k2) (5)

wherein k1 and k2 represent the setting threshold of the 1 st channel and the 2 nd channel of the original data set signal respectively, and count () represents the counting value;

3-6 monotonically increasing partial slope maxima MSSN1 and MSSN2 of the SNEO dataset signal are calculated by the following formula:

MSSN1＝max(SMIS1)，MSSN2＝max(SMIS2)

3-7 monotonically increasing partial slope counts SCSN1 and SCSN2 of the SNEO dataset signal above a given threshold are calculated by:

SCSN1＝count(SMIS1＞k3)，SCSN2＝count(SMIS2＞k4) (7)

wherein k3 and k4 represent the set thresholds of the 1 st channel and the 2 nd channel of the SNEO data set signal, respectively;

3-8 calculating the first difference absolute mean values of the VME data set signals AAF1 and AAF2 by:

3-9 calculate the first difference absolute maximum values MAF1 and MAF2 for the VME data set signals by:

MAF1＝max(|ES₁(i+1)-ES₁(i)|) (10)

MAF2＝max(|ES₂(i+1)-ES₂(i)|) (11)

wherein i1, 2., N-1;

3-10 calculating the second difference absolute mean values of the VME data set signals AAS1 and AAS2 by:

3-11 calculate the second absolute maximum of difference MAS1 and MAS2 for the VME data set signal by the following equations:

MAS1＝max(|ES₁(i+2)-ES₁(i)|) (14)

MAS2＝max(|ES₂(i+2)-ES₂(i)|) (15)

wherein i ═ 1, 2., N-2;

3-12 calculate the monotonically increasing partial slope averages ASMI1 and ASMI2 for the VME data set signal by the following equations:

where VL1 represents the length of the SMIV1 sequence, VL2 represents the length of the SMIV2 sequence, SMIV1(i) represents the ith element of the SMIV1 sequence, and SMIV2(i) represents the ith element of the SMIV2 sequence;

3-13 calculate monotonically increasing partial slope maxima MSVM1 and MSVM2 for the VME data set signal by the following formula:

MSVM1＝max(SMIV1)，MSVM2＝max(SMIV2) (18)

the total number of samples of the three data sets 3-14 is T, for signals with the length of N, 4-dimensional features are extracted from the sample data sets, 4-dimensional features are extracted from the SNEO data sets, and 12-dimensional features are extracted from the VME data sets, so that 20-dimensional feature extraction is extracted from the steps 3-2 to 3-13 to form a feature vector FVs of T x 20.

5. The intelligent classification method for distinguishing the electroencephalogram blink artifact from the frontal epileptic-like discharge according to claim 4, characterized in that the specific steps of the step 4 are as follows:

4-1, performing two-classification unsupervised clustering by using the feature vector FVs obtained in the step 3 and combining a K-means algorithm to obtain an unsupervised classification model.

Technical Field

The invention belongs to the field of electroencephalogram signal processing and intelligent medical treatment, and relates to a classification method based on a K-means unsupervised clustering algorithm and fusing multi-class signal multi-dimensional characteristic representation, aiming at a strong-noise electroencephalogram signal containing frontal epilepsy-like discharge. In the fields of epilepsia electroencephalogram analysis and artifact filtering, effective distinguishing of frontal epileptic-like discharge and blink artifacts is of great importance. The invention relates to an automatic classification method of epilepsy-like discharge signals and blink artifact signals based on EEG data under the epilepsy-like discharge background.

Background

Electroencephalograms (EEG) are the general reaction of electrophysiological activities of brain nerve cells on the surface of the cerebral cortex or scalp, contain rich brain activity information, and are widely applied to the fields of brain-computer interfaces (BCI), psychology, auxiliary diagnosis of diseases of the central nervous system of the brain (such as cerebral apoplexy, epilepsy, encephalitis and metabolic encephalopathy), and the like. The blink artifact is unavoidable in the electroencephalogram acquisition process, so that the detection of the electroencephalogram artifact is a crucial link in subsequent electroencephalogram analysis and signal analysis. However, in the auxiliary detection and analysis research of epilepsy, the waveform of a epilepsy-like discharge signal in the background of an epileptic is very similar to an electroencephalogram artifact caused by blinking, and the occurrence positions of the epilepsy-like discharge signal and the electroencephalogram artifact are also highly consistent and are concentrated in an electroencephalogram forehead channel, which often generates great interference on the detection of the blink artifact signal. At present, the key problem that the high similarity between the blink artifact and the frontal epilepsy-like discharge is usually ignored in the conventional blink artifact detection method, only the recognition of the blink artifact is concerned, and the classification of the epilepsy-like discharge signal and the blink artifact signal is ignored, so that the conventional blink artifact detection model in the electroencephalogram of the frontal epilepsy patient fails. Aiming at the problem, the invention provides a classification method for the fusion multi-class signal multi-dimensional characteristic representation of the strong noise electroencephalogram signals containing the frontal epilepsy-like discharge, and realizes the automatic and accurate classification of the blink artifact and the epilepsy-like discharge signals based on the K-means algorithm.

Disclosure of Invention

Aiming at the defects in the prior art, the invention provides an intelligent classification method for distinguishing electroencephalogram blink artifact and frontal epilepsy-like discharge.

Aiming at the characteristics that electroencephalogram signals of frontal lobe epilepsy patients contain a large number of epilepsy-like discharge signals, have high similarity with blink artifacts and are difficult to distinguish, the invention provides a multi-dimensional feature representation method which can be realized under a few channels by utilizing various signal transformation methods, and constructs an accurate classification model of the blink artifacts and the frontal epilepsy-like discharge signals by combining a K-means algorithm.

The technical scheme of the invention mainly comprises the following steps:

step 1, carrying out filtering processing and signal cutting on the EEG signal.

And 2, performing Smooth Nonlinear Energy Operator (SNEO) signal transformation and Variational Mode Extraction (VME) signal transformation on the EEG signals processed in the step 1 to obtain an SNEO data set and a VME data set.

And 3, performing 20-dimensional feature extraction on signals corresponding to the three data sets.

And 4, carrying out two-classification unsupervised clustering through a K-means algorithm to construct an unsupervised classification model.

And 5, classifying the blink artifact and the seizure discharge signal through the established unsupervised classification model.

The step 1 comprises the following specific steps:

and filtering the 21-channel 1000Hz original EEG signals. Each channel respectively passes through a 50Hz notch filter and a 0.5-30Hz band-pass filter to obtain a filtering signal; since the blink artifact and seizure-like discharge signals mainly affect the 1 st channel (Fp1 channel) and the 2 nd channel (Fp2 channel), the subsequent brain electrical signals only use the first two channels. Because the duration time of the blink artifact potential signal and the duration time of the frontal epilepsy-like discharge signal are within 1s, the filtering data are cut by adopting a sliding window method, the window size is set to be 1s, and the step length is 0.5 s. And obtaining cutting signals, wherein all the cutting signals form a complete sample data set, and the number of the signals of the sample data set is assumed to be T.

The step 2 comprises the following specific steps:

assume that each signal in the sample data set is S_2*1000，S_i(i ═ 1, 2) denotes all data of the ith channel, and N denotes a data length.

2-1 calculating the ith channel S of all signals of the sample data set by the following formula_iCorresponding SNEO signal of (1), denoted as ES_i：

ES_i(t)＝ω(t)*(S_i(t)*S_i(t)-S_i(t-1)*S_i(t+1))

In which ω (t) denotes a window function, ES_i(t) represents ES_iThe t-th data of (1). All SNEO signals are then combined into a complete SNEO data set, and the number of signals in the SNEO data set is T.

2-2 calculating the ith channel S of all signals of the sample data set by the following formula_iCorresponding VME signal, denoted VS_i：

Where a represents the compaction factor, ω_dRepresenting an approximation of the center frequency of the desired mode, m representing an iteration coefficient, λ representing a rise coefficient, VS_i(t) represents VS_iT data of (VS)_i(0) Is 0. And then all VME signals form a complete VME data set, and the number of the signals in the VME data set is T.

The step 3 comprises the following specific steps:

3-1, calculating the multi-dimensional multi-channel characteristics of the electroencephalogram sample, wherein the method mainly comprises the following steps:

1) computing monotonically increasing fractional slope maxima MSMI1(maximum value of the slope of the monotonic increasing partial slope of channel 1) and MSMI2(maximum value of the slope of channel 2), SCMI1(slope of the slope of channel 1) and SCMI2(slope of the monotonic increasing partial slope of channel 2) respectively for each signal in the sample dataset, exceeding a given threshold

2) Calculating monotonically increasing partial slope maxima MSSN1(maximum value of the slope of the monoclonal organizing part of SNEO of channel 1) and MSSN2(maximum value of the slope of the monoclonal organizing part of SNEO of channel 2), and SCSN1(slope of the monoclonal organizing part of SNEO of channel 1) and SCSN2(slope of the monoclonal organizing part of SNEO of channel 2) for each signal in the SNEO signal dataset, respectively, exceeding a given threshold

3) Calculating the first absolute average of difference value AAF1(average of the absolute value of the first difference of channel 1) and channel 2(Fp2 channel), AAF2(average of the absolute value of channel 2), MAF 25 (maximum of the absolute value of the first difference of channel 1) and MAF2(maximum of the absolute value of the first difference of channel 2), AAS 5632 (average of the absolute value of the first difference of channel 1) and AAF1(average of the absolute value of the second difference of channel 1 and AAS 68 (average of the absolute value of the second difference of channel 26) for each signal in the VME data set channel 2), monotonically increasing fractional slope averages ASMI1(average value of the slope of the microbiological interception part of channel 1) and ASMI2(average value of the slope of the microbiological interception part of channel 2), monotonically increasing fractional slope maxima MSVM1(maximum value of the slope of the microbiological interception part of VME of channel 1) and MSVM2(maximum value of the slope of the microbiological interception part of VME of channel 2)

The above are 20 dimensional features.

3-2 calculating the slope g of the monotonically increasing portion of the ith channel of the signal in the original data set by:

wherein scv represents a signal monotonically increasing fractional count value.

3-3 utilizes 3-2 to calculate the monotonically increasing partial slope sequences for channel 1 and channel 2 of the sample data set signal as SMIO1 and SMIO2, respectively, for channel 1 and channel 2 of the SNEO data set signal as SMIS1 and SMIS2, respectively, and for channel 1 and channel 2 of the VME data set signal as SMIV1 and SMIV2, respectively.

3-4 calculating the monotonically increasing fractional slope maxima MSMI1 and MSMI2 of the sample data set signal by the following formula:

MSMI1＝max(SMIO1)，MSMI2＝max(SMIO2)

where max () denotes taking the maximum value.

3-5 calculate the monotonically increasing fractional slope counts SCMI1 and SCMI2 of the sample data set signal that exceed a given threshold by:

SCMI1＝count(SMIO1＞k1)，SCMI2＝count(SMIO2＞k2)

where k1 and k2 represent the set thresholds for the 1 st and 2 nd channels of the original data set signal, respectively, and count () represents the count value.

3-6 monotonically increasing partial slope maxima MSSN1 and MSSN2 of the SNEO dataset signal are calculated by the following formula:

MSSN1＝max(SMIS1)，MSSN2＝max(SMIS2)

3-7 monotonically increasing partial slope counts SCSN1 and SCSN2 of the SNEO dataset signal above a given threshold are calculated by:

SCSN1＝count(SMIS1＞k3)，SCSN2＝count(SMIS2＞k4)

where k3 and k4 represent the set thresholds for the 1 st and 2 nd channels of the SNEO dataset signal, respectively.

3-8 calculating the first difference absolute mean values of the VME data set signals AAF1 and AAF2 by:

3-9 calculate the first difference absolute maximum values MAF1 and MAF2 for the VME data set signals by:

MAF1＝max(|ES₁(i+1)-ES₁(i)|)

MAF2＝max(|ES₂(i+1)-ES₂(i)|)

wherein i is1, 2.

3-10 calculating the second difference absolute mean values of the VME data set signals AAS1 and AAS2 by:

3-11 calculate the second absolute maximum of difference MAS1 and MAS2 for the VME data set signal by the following equations:

MAS1＝max(|ES₁(i+2)-ES₁(i)|)

MAS2＝max(|ES₂(i+2)-ES₂(i)|)

wherein i is1, 2.

3-12 calculate the monotonically increasing partial slope averages ASMI1 and ASMI2 for the VME data set signal by the following equations:

where VL1 represents the length of the SMIV1 sequence, VL2 represents the length of the SMIV2 sequence, SMIV1(i) represents the ith element of the SMIV1 sequence, and SMIV2(i) represents the ith element of the SMIV2 sequence.

3-13 calculate monotonically increasing partial slope maxima MSVM1 and MSVM2 for the VME data set signal by the following formula:

MSVM1＝max(SMIV1)，MSVM2＝max(SMIV2)

the total number of samples of the three data sets 3-14 is T, and for signals with the length of N, 4-dimensional features are extracted from the original data set, 4-dimensional features are extracted from the SNEO data set, and 12-dimensional features are extracted from the VME data set, so that 20-dimensional feature extraction is extracted from the steps 3-2 to 3-13 to form a feature vector FVs of T × 20.

The step 4 comprises the following specific steps:

4-1, performing two-classification unsupervised clustering by using the feature vector FVs obtained in the step 3 and combining a K-means algorithm to obtain an unsupervised classification model.

The invention has the following beneficial effects

Aiming at the problems that the waveform of an epileptic discharge signal of an epileptic is very similar to an electroencephalogram artifact caused by blink and the occurrence positions of the epileptic discharge signal and the electroencephalogram artifact caused by blink are highly consistent, and the current situation that most blink artifact detection models ignore epileptic discharge signals is faced, the invention provides a classification model which is combined with a K-means algorithm and is used for fusing multi-signal multi-dimensional characteristic representation of a high-noise electroencephalogram signal containing frontal-pole epileptic discharge, so that the difficulty that the blink artifact detection precision is low under the background containing the epileptic discharge signals is overcome, the problem that the epileptic discharge is ignored by the existing model can be solved, and the accurate automatic classification of the blink artifact and the frontal-pole epileptic discharge can be realized.

Drawings

FIG. 1 is a schematic block diagram of the method of the present invention;

FIG. 2 is a flow chart of a method of the present invention;

FIG. 3 is a graph comparing the raw signal, the SNEO signal and the VME signal for blink artifact and seizure signals according to the present invention;

Detailed Description

The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

As shown in fig. 1, fig. 2 and fig. 3, the implementation steps of the classification method of blink artifact and frontal epileptic discharge signals based on multi-signal multi-dimensional feature representation have been described in detail in the summary of the invention, that is, the technical scheme of the invention mainly includes the following steps:

step 1, carrying out filtering processing and signal cutting on the EEG signal.

And 2, performing Smooth Nonlinear Energy Operator (SNEO) signal transformation and Variational Mode Extraction (VME) signal transformation on the EEG signals processed in the step 1 to obtain an SNEO data set and a VME data set.

And 3, performing 20-dimensional feature extraction on signals corresponding to the three data sets.

And 4, carrying out two-classification unsupervised clustering through a K-means algorithm to construct an unsupervised classification model.

And 5, classifying the blink artifact and the seizure discharge signal through the established unsupervised classification model.

The step 1 comprises the following specific steps:

and filtering the 21-channel 1000Hz original EEG signals. Each channel respectively passes through a 50Hz notch filter and a 0.5-30Hz band-pass filter to obtain a filtering signal; since the blink artifact and seizure-like discharge signals mainly affect the 1 st channel (Fp1 channel) and the 2 nd channel (Fp2 channel), the subsequent brain electrical signals only use the first two channels. Because the duration time of the blink artifact potential signal and the duration time of the frontal epilepsy-like discharge signal are within 1s, the filtering data are cut by adopting a sliding window method, the window size is set to be 1s, and the step length is 0.5 s. And obtaining cutting signals, wherein all the cutting signals form a complete sample data set, and the number of the signals of the sample data set is assumed to be T.

The step 2 comprises the following specific steps:

assume that each signal in the sample data set is S_2*1000，S_i(i ═ 1, 2) denotes all data of the ith channel, and N denotes a data length.

2-1 is represented by the following formulaComputing ith channel S of all signals of sample data set_iCorresponding SNEO signal of (1), denoted as ES_i：

ES_i(t)＝ω(t)*(S_i(t)*S_i(t)-S_i(t-1)*S_i(t+1))

In which ω (t) denotes a window function, ES_i(t) represents ES_iThe t-th data of (1). All SNEO signals are then combined into a complete SNEO data set, and the number of signals in the SNEO data set is T.

2-2 calculating the ith channel S of all signals of the sample data set by the following formula_iCorresponding VME signal, denoted VS_i：

Where a represents the compaction factor, ω_dRepresenting an approximation of the center frequency of the desired mode, m representing an iteration coefficient, λ representing a rise coefficient, VS_i(t) represents VS_iT data of (VS)_i(0) Is 0. And then all VME signals form a complete VME data set, and the number of the signals in the VME data set is T.

The step 3 comprises the following specific steps:

3-1, calculating the multi-dimensional multi-channel characteristics of the electroencephalogram sample, wherein the method mainly comprises the following steps:

1) computing monotonically increasing fractional slope maxima MSMI1(maximum value of the slope of the monotonic increasing partial slope of channel 1) and MSMI2(maximum value of the slope of channel 2), SCMI1(slope of the slope of channel 1) and SCMI2(slope of the monotonic increasing partial slope of channel 2) respectively for each signal in the sample dataset, exceeding a given threshold

2) Calculating monotonically increasing partial slope maxima MSSN1(maximum value of the slope of the monoclonal organizing part of SNEO of channel 1) and MSSN2(maximum value of the slope of the monoclonal organizing part of SNEO of channel 2), and SCSN1(slope of the monoclonal organizing part of SNEO of channel 1) and SCSN2(slope of the monoclonal organizing part of SNEO of channel 2) for each signal in the SNEO signal dataset, respectively, exceeding a given threshold

3) Calculating the first absolute average of difference value AAF1(average of the absolute value of the first difference of channel 1) and channel 2(Fp2 channel), AAF2(average of the absolute value of channel 2), MAF 25 (maximum of the absolute value of the first difference of channel 1) and MAF2(maximum of the absolute value of the first difference of channel 2), AAS 5632 (average of the absolute value of the first difference of channel 1) and AAF1(average of the absolute value of the second difference of channel 1 and AAS 68 (average of the absolute value of the second difference of channel 26) for each signal in the VME data set channel 2), monotonically increasing fractional slope averages ASMI1(average value of the slope of the microbiological interception part of channel 1) and ASMI2(average value of the slope of the microbiological interception part of channel 2), monotonically increasing fractional slope maxima MSVM1(maximum value of the slope of the microbiological interception part of VME of channel 1) and MSVM2(maximum value of the slope of the microbiological interception part of VME of channel 2)

The above are 20 dimensional features.

3-2 calculating the slope g of the monotonically increasing portion of the ith channel of the signal in the original data set by:

wherein scv represents a signal monotonically increasing fractional count value.

3-3 calculate the monotonically increasing partial slope sequences for channel 1 and channel 2 of the sample data set signal as SMIO1 and SMIO2, respectively, for channel 1 and channel 2 of the SNEO data set signal as SMIS1 and SMIS2, respectively, and for channel 1 and channel 2 of the VME data set signal as SMIV1 and SMIV2, respectively, using equations 3-2.

3-4 calculating the monotonically increasing fractional slope maxima MSMI1 and MSMI2 of the sample data set signal by the following formula:

MSMI1＝max(SMIO1)，MSMI2＝max(SMIO2)

where max () denotes taking the maximum value.

3-5 calculate the monotonically increasing fractional slope counts SCMI1 and SCMI2 of the sample data set signal that exceed a given threshold by:

SCMI1＝count(SMIO1＞k1)，SCMI2＝count(SMIO2＞k2)

where k1 and k2 represent the set thresholds for the 1 st and 2 nd channels of the original data set signal, respectively, and count () represents the count value.

3-6 monotonically increasing partial slope maxima MSSN1 and MSSN2 of the SNEO dataset signal are calculated by the following formula:

MSSN1＝max(SMIS1)，MSSN2＝max(SMIS2)

3-7 monotonically increasing partial slope counts SCSN1 and SCSN2 of the SNEO dataset signal above a given threshold are calculated by:

SCSN1＝count(SMIS1＞k3)，SCSN2＝count(SMIS2＞k4)

where k3 and k4 represent the set thresholds for the 1 st and 2 nd channels of the SNEO dataset signal, respectively.

3-8 calculating the first difference absolute mean values of the VME data set signals AAF1 and AAF2 by:

3-9 calculate the first difference absolute maximum values MAF1 and MAF2 for the VME data set signals by:

MAF1＝max(|ES₁(i+1)-ES₁(i)|)

MAF2＝max(|ES₂(i+1)-ES₂(i)|)

wherein i is1, 2.

3-10 calculating the second difference absolute mean values of the VME data set signals AAS1 and AAS2 by:

3-11 calculate the second absolute maximum of difference MAS1 and MAS2 for the VME data set signal by the following equations:

MAS1＝max(|ES₁(i+2)-ES₁(i)|)

MAS2＝max(|ES₂(i+2)-ES₂(i)|)

wherein i is1, 2.

3-12 calculate the monotonically increasing partial slope averages ASMI1 and ASMI2 for the VME data set signal by the following equations:

where VL1 represents the length of the SMIV1 sequence, VL2 represents the length of the SMIV2 sequence, SMIV1(i) represents the ith element of the SMIV1 sequence, and SMIV2(i) represents the ith element of the SMIV2 sequence.

3-13 calculate monotonically increasing partial slope maxima MSVM1 and MSVM2 for the VME data set signal by the following formula:

MSVM1＝max(SMIV1)，MSVM2＝max(SMIV2)

the total number of samples of the three data sets 3-14 is T, and for signals with the length of N, 4-dimensional features are extracted from the original data set, 4-dimensional features are extracted from the SNEO data set, and 12-dimensional features are extracted from the VME data set, so that 20-dimensional feature extraction is extracted from the steps 3-2 to 3-13 to form a feature vector FVs of T × 20.

The step 4 comprises the following specific steps:

4-1, performing two-classification unsupervised clustering by using the feature vector FVs obtained in the step 3 and combining a K-means algorithm to obtain an unsupervised classification model.

In order to achieve a better effect of accurately classifying the blink artifact and the frontal epileptic-like discharge signal, the following description is introduced from the aspects of parameter selection and design in practical application as a reference for the invention to be used in other applications:

when the method is used for processing EEG signals, the length of input data is set to be 1 second, and the data overlapping rate is set to be 50%. The 20-dimensional features extracted in the steps 3-2 to 3-13 comprise the most main information for distinguishing the blink artifact from the eclamptic discharge signals, and the parameter K is set to be 2 by adopting a K-means algorithm finally.

In order to truly test the classification effect of the invention on blink artifact and frontal epilepsy-like discharge, test experiments were performed on EEG data of real patients in children hospitals affiliated to the medical college at Zhejiang university:

the experimental data is 21 channels, the sampling frequency is 1000Hz, the data is divided into 5 different sick individuals, and the average data length is 20 minutes. The accuracy (accuracuracy) of the Wavelet-ICA-SVM algorithm proposed by ChongYehSai et al in 2018 on the data set is 71.46%, 65.20%, 60.16%, 66.34% and 61.54% respectively, the accuracy of the invention on different data sets is 99.68%, 98.81%, 84.70%, 98.30% and 88.48% respectively, the invention is improved by 29.05% compared with the average accuracy of a comparison algorithm, and the invention is fully proved to have excellent performance on real data.

The method can be used for solving the problem that the blinking artifact of the epileptic patient is difficult to recognize clinically under the background of the frontal epilepsy-like discharge, the feature vector is obtained through signal transformation and multi-dimensional feature representation, and the blinking artifact and the frontal epilepsy-like discharge signal can be accurately classified by combining a K-means clustering model.