CN111387975B - Electroencephalogram signal identification method based on machine learning - Google Patents

Electroencephalogram signal identification method based on machine learning Download PDF

Info

Publication number
CN111387975B
CN111387975B CN202010199754.XA CN202010199754A CN111387975B CN 111387975 B CN111387975 B CN 111387975B CN 202010199754 A CN202010199754 A CN 202010199754A CN 111387975 B CN111387975 B CN 111387975B
Authority
CN
China
Prior art keywords
electroencephalogram signal
electroencephalogram
signal
reference interval
sample data
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202010199754.XA
Other languages
Chinese (zh)
Other versions
CN111387975A (en
Inventor
唐玮
束云潇
郝敬宾
杨雅涵
刘送永
张梅梅
姜雨辰
王帅
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Xuzhou Health Research Institute Co ltd
China University of Mining and Technology CUMT
Original Assignee
Xuzhou Health Research Institute Co ltd
China University of Mining and Technology CUMT
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Xuzhou Health Research Institute Co ltd, China University of Mining and Technology CUMT filed Critical Xuzhou Health Research Institute Co ltd
Priority to CN202010199754.XA priority Critical patent/CN111387975B/en
Publication of CN111387975A publication Critical patent/CN111387975A/en
Application granted granted Critical
Publication of CN111387975B publication Critical patent/CN111387975B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
    • A61B5/316—Modalities, i.e. specific diagnostic methods
    • A61B5/369—Electroencephalography [EEG]
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7203—Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7225—Details of analog processing, e.g. isolation amplifier, gain or sensitivity adjustment, filtering, baseline or drift compensation
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235—Details of waveform analysis
    • A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
    • A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/02—Preprocessing
    • G06F2218/04—Denoising
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/15—Biometric patterns based on physiological signals, e.g. heartbeat, blood flow

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Molecular Biology (AREA)
  • Theoretical Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Signal Processing (AREA)
  • Public Health (AREA)
  • Animal Behavior & Ethology (AREA)
  • Surgery (AREA)
  • Psychiatry (AREA)
  • Veterinary Medicine (AREA)
  • Medical Informatics (AREA)
  • Evolutionary Computation (AREA)
  • Data Mining & Analysis (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Pathology (AREA)
  • Mathematical Physics (AREA)
  • Physiology (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computational Linguistics (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Fuzzy Systems (AREA)
  • Psychology (AREA)
  • Evolutionary Biology (AREA)
  • Power Engineering (AREA)
  • Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)

Abstract

The invention discloses an electroencephalogram signal identification method based on machine learning, which comprises the steps of constructing sample data comprising a plurality of electroencephalograms, extracting alpha wave rhythm of each electroencephalogram signal in the sample data, training a deep learning framework to obtain an initial model, inputting alpha wave rhythm of a target electroencephalogram signal into the initial model for retraining to obtain an electroencephalogram signal deep learning model, extracting characteristic parameters of each electroencephalogram signal in the sample data, and dividing the data set into a first reference interval, a second reference interval, a third reference interval and a fourth reference interval. The method comprises the steps of obtaining an electroencephalogram signal to be detected, extracting characteristic parameters of the electroencephalogram signal to be detected, obtaining parameters to be detected, identifying a reference interval where the parameters to be detected are located, and detecting a user state represented by the corresponding electroencephalogram signal to be detected according to the reference interval where the parameters to be detected are located, wherein the identified reference interval has high accuracy, and therefore the accuracy of the detected user state is improved.

Description

Electroencephalogram signal identification method based on machine learning
Technical Field
The invention relates to the technical field of electroencephalogram signal identification, in particular to an electroencephalogram signal identification method based on machine learning.
Background
Research shows that the rhythmicity of the electroencephalogram signals has obvious expression on mesoscopic and macroscopic scales. Different rhythmic brain electrical waves are thought to have different roles in brain function. More and more research has shown that due to the extreme complexity of the brain and the interactions between internal systems, there are complex changes in different rhythms even when the simplest brain functions act. Each brain wave has a corresponding different brain consciousness state, that is, different brain waves are needed under different consciousness states to complete the brain work best. If the brain fails to produce corresponding brain waves in a particular situation, there is a problem, for example, if the brain does not produce delta waves and theta waves when trying to sleep, insomnia may occur.
Researches show that different intelligence quotient people, different psychological stress people and different brain disease patients have different characteristics in parameters such as electroencephalogram signal alpha, beta, theta, delta wave rhythm, power spectrum and the like. By extracting the characteristic parameters and carrying out identification, classification, similarity estimation and the like on the electroencephalogram signals according to different components, various related problems are facilitated to be solved.
The patent with publication number CN110558977A provides a epileptic seizure electroencephalogram signal classification method based on machine learning fuzzy feature selection. Although the method can obtain higher epilepsia electroencephalogram signal classification accuracy through the SVM, the method has limitation on classification targets and is insufficient for solving the problem of identification and detection of other electroencephalogram signals. Therefore, the method cannot widely solve the problem of electroencephalogram signal identification and detection. The machine learning system belongs to the field of artificial intelligence and comprises three types of supervised learning, unsupervised learning and semi-supervised learning. In recent years, deep learning, which is equivalent to a deep neural network in terms of basic structure, is becoming a new field in machine learning research. The deep neural network can realize the approximation of complex functions by a multi-level feature learning method when the number of network layers is larger, so that the deep neural network can express the feature expression capability which the shallow neural network does not have.
In conclusion, based on the current situation that the electroencephalogram signal components and the associated events thereof and the number of people are large, the electroencephalogram signal features of the samples are extracted by using the deep learning method, so that the electroencephalogram signals can be identified and detected in a generalized manner. However, the electroencephalogram signal identification scheme has limitations, and accuracy of identification results is easily affected.
Disclosure of Invention
Aiming at the problems, the invention provides an electroencephalogram signal identification method based on machine learning.
In order to realize the aim of the invention, the invention provides an electroencephalogram signal identification method based on machine learning, which comprises the following steps:
s10, constructing sample data comprising a plurality of electroencephalograms, extracting alpha wave rhythms of the electroencephalograms in the sample data after denoising and normalizing the sample data, and integrating the alpha wave rhythms into an electroencephalogram deep learning frame for training to obtain an initial model;
s20, judging the rhythm of the target electroencephalogram signal by adopting the initial model, and if the judgment result is consistent with the reference rhythm of the target electroencephalogram signal, inputting the alpha-wave pitch of the target electroencephalogram signal into the initial model for retraining to obtain an electroencephalogram signal deep learning model;
s30, extracting characteristic parameters of each electroencephalogram signal in sample data by adopting characteristic vector method power spectrum estimation, and dividing a data set formed by the characteristic parameters into a first reference interval, a second reference interval, a third reference interval and a fourth reference interval;
s40, acquiring the electroencephalogram signal to be detected, extracting the characteristic parameters of the electroencephalogram signal to be detected, obtaining the parameter to be detected, and identifying the reference interval where the parameter to be detected is located.
In one embodiment, after constructing the sample data including a plurality of electroencephalogram signals, the method further includes:
denoising each electroencephalogram signal of the sample data by using filters with cut-off frequencies of 0.5Hz and 40 Hz;
and sequentially carrying out data standardization on the sample data subjected to the denoising treatment on all sampling points corresponding to a single electrode by using z-score standardization, and converting the sampling points corresponding to each electrode into data distribution with the mean value of 0 and the standard deviation of 1 so as to realize the standardization treatment on the sample data.
In one embodiment, extracting the α -wave pitch of each electroencephalogram signal in the sample data includes:
the method comprises the steps of constructing sine signals with frequencies of 8.5Hz, 9.5Hz, 10.5Hz, 11.5Hz and 12.5Hz and cosine signals with frequencies of 8.5Hz, 9.5Hz, 10.5Hz, 11.5Hz and 12.5Hz in a group respectively, using the two groups of signals as reference signals of alpha waves, enabling the two groups of signals and sample data to form input ends of an independent component analysis method, separating input signals of the input ends by a FastICA algorithm to separate alpha waves of all electroencephalogram signals in the sample data to obtain electroencephalogram signals D without the alpha waves, and determining alpha wave rhythms of all the electroencephalogram signals according to differences between all the electroencephalogram signals in the sample data and the corresponding electroencephalogram signals D.
In one embodiment, integrating the alpha-wave rhythm into a brain electrical signal deep learning framework for training, and obtaining the initial model comprises:
integrating alpha wave pitch into brain electric signal deep learning frame, converting alpha wave pitch into corresponding picture in brain electric signal deep learning frame to obtain alpha wave pitch picture, making brain electric signal alpha wave pitch picture data be recognized as array arranged according to pixels by computer, filtering the features of each possible position of a data picture by adding a weight coefficient to a convolutional layer of an electroencephalogram deep learning frame to obtain a filtered array, adding a nonlinear factor through an activation function, performing nonlinear mapping on the output result of the convolutional layer by adopting the nonlinear factor, pooling the non-linearly mapped matrix in a pooling layer to obtain a pooled matrix, performing weighted summation on the pooled matrix through a full-link layer, and identifying and classifying the obtained result, and determining an initial model according to the operating parameters of the electroencephalogram signal deep learning frame when the identification and classification precision reaches the set precision.
In one embodiment, extracting the feature parameters of each electroencephalogram signal in the sample data by using the power spectrum estimation of the feature vector method comprises:
and performing spectrum estimation calculation on each electroencephalogram signal by adopting an MUSIC spectrum estimation method to obtain the maximum value, the minimum value, the average value and the standard deviation of the power spectrum amplitude as characteristic parameters of each electroencephalogram signal.
In one embodiment, each brain electrical signal of the sample data is 10 minutes in length.
The electroencephalogram signal identification method based on machine learning constructs sample data comprising a plurality of electroencephalogram signals, after the sample data is denoised and normalized, extracting alpha wave rhythm of each electroencephalogram signal in the sample data, integrating the alpha wave rhythm into an electroencephalogram signal deep learning frame for training to obtain an initial model, judging the rhythm of a target electroencephalogram signal by adopting the initial model, if the judgment result is consistent with the reference rhythm of the target electroencephalogram signal, inputting the alpha wave pitch of the target electroencephalogram signal into the initial model for retraining to obtain an electroencephalogram signal deep learning model, extracting the characteristic parameters of each electroencephalogram signal in sample data by adopting characteristic vector method power spectrum estimation, and dividing a data set formed by the characteristic parameters into a first reference interval, a second reference interval, a third reference interval and a fourth reference interval. The method comprises the steps of obtaining an electroencephalogram signal to be detected, extracting characteristic parameters of the electroencephalogram signal to be detected, obtaining parameters to be detected, identifying a reference interval where the parameters to be detected are located, and detecting a user state represented by the corresponding electroencephalogram signal to be detected according to the reference interval where the parameters to be detected are located, wherein the identified reference interval where the parameters to be detected are located has high accuracy, and therefore accuracy of the detected user state is improved.
Drawings
FIG. 1 is a flow chart of an electroencephalogram signal identification method based on machine learning according to an embodiment;
FIG. 2 is a diagram of deep learning neural network propagation for one embodiment;
FIG. 3 is a flowchart illustrating a brain electrical signal recognition method based on machine learning according to another embodiment.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.
Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the application. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. It is explicitly and implicitly understood by one skilled in the art that the embodiments described herein can be combined with other embodiments.
Referring to fig. 1, fig. 1 is a schematic flow chart of an electroencephalogram signal identification method based on machine learning according to an embodiment, and includes the following steps:
s10, constructing sample data comprising a plurality of electroencephalogram signals, extracting alpha wave rhythms of all electroencephalogram signals in the sample data after denoising and normalizing the sample data, and integrating the alpha wave rhythms into an electroencephalogram signal deep learning framework for training to obtain an initial model.
In one embodiment, each electroencephalogram signal of the sample data is 10 minutes in length.
Specifically, the sample data may include 100 electroencephalograms, each of which has a length of 10 minutes. All electroencephalogram signals in the sample data have the same or similar characteristics. Denoising and standardizing each electroencephalogram signal in the sample data, selecting middle 300 seconds of data, extracting alpha wave rhythm by using an independent component analysis method, and integrating the obtained rhythm into an electroencephalogram signal deep learning framework for training to obtain an electroencephalogram signal deep learning model (initial model).
And S20, judging the rhythm of the target electroencephalogram signal by adopting the initial model, and if the judgment result is consistent with the reference rhythm of the target electroencephalogram signal, inputting the alpha wave rhythm of the target electroencephalogram signal into the initial model for retraining to obtain an electroencephalogram signal deep learning model.
The target brain electrical signal is the brain electrical signal with known rhythm (the known rhythm is the corresponding reference rhythm). The target brain electrical signal has the same or similar characteristics with the brain electrical signal in the sample data.
The steps are used for judging the 10-minute electroencephalogram signal rhythm of the recognition target (target electroencephalogram signal) based on the established electroencephalogram signal deep learning model (initial model). If the judgment result is consistent with the sample electroencephalogram signal, the alpha wave pitch of the identified target is sent to the electroencephalogram signal deep learning frame for retraining so as to further perfect the model and improve the accuracy of the model. If the result is different from the sample signal, it is discarded.
S30, extracting characteristic parameters of each electroencephalogram signal in the sample data by adopting characteristic vector method power spectrum estimation, and dividing a data set formed by the characteristic parameters into a first reference interval, a second reference interval, a third reference interval and a fourth reference interval.
And in the steps, the characteristic parameters of the electroencephalogram signals in the sample data are extracted by using power spectrum estimation of a characteristic vector method, and the characteristic parameter range is determined to be used as the basis for further classification of the target electroencephalogram signals. Dividing a data set formed by the characteristic parameters into four reference intervals: a first reference interval, a second reference interval, a third reference interval, and a fourth reference interval. The characteristic parameters of the related target electroencephalogram signals (such as electroencephalogram signals to be detected) can be distinguished by utilizing the four reference intervals, and distinguishing results are obtained.
Specifically, the first reference interval is at the center position in the space where the feature parameter vector is located. The second reference interval, the third reference interval and the fourth reference interval are arranged outwards in sequence. Wherein the outer side of the previous reference interval is the inner side of the next reference interval.
S40, acquiring the electroencephalogram signal to be detected, extracting the characteristic parameters of the electroencephalogram signal to be detected, obtaining the parameter to be detected, and identifying the reference interval where the parameter to be detected is located.
The electroencephalogram signal to be detected and each electroencephalogram signal in the sample data have the same or similar characteristics. The electroencephalogram signals to be detected can be acquired through a related acquisition device worn by a user.
Identifying the reference interval in which the parameter to be measured is located includes:
and comparing the parameter to be measured with the first reference interval, the second reference interval, the third reference interval and the fourth reference interval, and determining the reference interval with the highest similarity to the parameter to be measured as the reference interval where the parameter to be measured is located.
Optionally, the electroencephalogram signal to be detected and the target electroencephalogram signal can be the same electroencephalogram signal. At this time, the above steps perform more time-length electroencephalogram signal acquisition on the identification target consistent with the result sample in S20, and transmit the acquired data to the computer. Segmenting the electroencephalogram signal according to the time length of 10 minutes, and estimating and extracting the characteristic parameters of the electroencephalogram signal in each segment by adopting a characteristic vector method power spectrum. And comparing the extracted characteristic parameters with the first reference interval, the second reference interval, the third reference interval and the fourth reference interval by using a support vector machine, and judging the similarity degree of the recognition target and the sample according to the interval where the characteristic parameters are located.
In one embodiment, the first reference interval, the second reference interval, the third reference interval, and the fourth reference interval may respectively characterize a degree of depression of the user corresponding to the electroencephalogram signal to be detected, for example, the first reference interval characterizes normal mood, the second reference interval characterizes mild depression, the third reference interval characterizes moderate depression, and the fourth reference interval characterizes severe depression. At the moment, the range of the electroencephalogram characteristic parameters of the patient (corresponding user) can be judged by using the four reference intervals, and four judgment results of health, mild depression, moderate depression and severe depression are obtained. The four benchmarks are stored in respective computers.
In the practical application process, the confirmed patient wears the electroencephalogram signal acquisition device with the wireless data transmission function, and the acquired real-time data are transmitted to the computer. Segmenting the electroencephalogram signal according to the time length of 10 minutes, and estimating and extracting the characteristic parameters of the electroencephalogram signal in each segment by adopting a characteristic vector method power spectrum. And comparing the extracted characteristic parameters with the first reference interval, the second reference interval, the third reference interval and the fourth reference interval by using a support vector machine, and judging the incidence degree of the depression according to the range of the characteristic parameters. If the color is light, no early warning is given. If the degree is moderate, a low-level early warning signal is sent to the family members. If the degree of severity is severe, a high-level early warning signal is sent to the family members and the medical institutions appointed in advance.
In one example, the extracted characteristic parameters (characteristic parameters of the electroencephalogram signal to be detected) are compared with a first reference interval, a second reference interval, a third reference interval and a fourth reference interval, and the similarity between the corresponding target and the sample is judged and identified according to the interval where the parameter to be detected is located, specifically: dividing the interval in the middle of the total data set into a first reference interval, wherein the number of samples in the first reference interval is not more than 30%; dividing a section, which is outside the first reference section of the data set and has the sample number reaching 50% of the residual sample number, into a second reference section; the remaining part of 80% is divided into a third reference interval; the last part is divided into a fourth reference interval. And judging the similarity degree of the target signal and the sample signal according to the section of the electroencephalogram characteristic parameter of the patient. If the output result is in the third reference interval, the output result is similar to the basic characteristics; if the current time is in the second reference interval, outputting a result that most features are similar; and if the output result is in the first reference interval, outputting that almost all the characteristics are similar.
The electroencephalogram signal identification method based on machine learning constructs sample data comprising a plurality of electroencephalogram signals, after the sample data is denoised and normalized, extracting alpha wave rhythm of each electroencephalogram signal in the sample data, integrating the alpha wave rhythm into an electroencephalogram signal deep learning frame for training to obtain an initial model, judging the rhythm of a target electroencephalogram signal by adopting the initial model, if the judgment result is consistent with the reference rhythm of the target electroencephalogram signal, inputting the alpha wave pitch of the target electroencephalogram signal into the initial model for retraining to obtain an electroencephalogram signal deep learning model, extracting the characteristic parameters of each electroencephalogram signal in sample data by adopting characteristic vector method power spectrum estimation, and dividing a data set formed by the characteristic parameters into a first reference interval, a second reference interval, a third reference interval and a fourth reference interval. The method comprises the steps of obtaining an electroencephalogram signal to be detected, extracting characteristic parameters of the electroencephalogram signal to be detected to obtain a parameter to be detected, identifying a reference interval where the parameter to be detected is located, and detecting a user state represented by the corresponding electroencephalogram signal to be detected according to the reference interval where the parameter to be detected is located, wherein the identified reference interval where the parameter to be detected is located has high accuracy, and therefore the accuracy of the detected user state is improved.
In one embodiment, after constructing the sample data including a plurality of electroencephalogram signals, the method further includes:
denoising each electroencephalogram signal of the sample data by using filters with cut-off frequencies of 0.5Hz and 40 Hz;
and sequentially carrying out data normalization processing on the sample data after the de-noising processing on all sampling points corresponding to a single electrode by using z-score standardization, and converting the sampling points corresponding to each electrode into data distribution with the mean value of 0 and the standard deviation of 1 so as to realize the normalization processing of the sample data.
The denoising and normalization processing on the sample data can specifically comprise denoising the electroencephalogram signal by using filters with cut-off frequencies of 0.5Hz and 40Hz based on a FastICA method. Based on Matlab R2018a, data normalization processing is sequentially performed on all sampling points corresponding to a single electrode by using z-score standardization, and the sampling points corresponding to each electrode are converted into data distribution with the average value of 0 and the standard deviation of 1.
In one embodiment, extracting the α -wave pitch of each electroencephalogram signal in the sample data includes:
constructing sine signals with frequencies of 8.5Hz, 9.5Hz, 10.5Hz, 11.5Hz and 12.5Hz and cosine signals with frequencies of 8.5Hz, 9.5Hz, 10.5Hz, 11.5Hz and 12.5Hz, respectively, using the two groups of signals as reference signals of alpha waves, forming input ends of an independent component analysis method by using the two groups of signals and sample data, separating input signals of the input ends by using a FastICA algorithm to separate alpha waves of all electroencephalogram signals in the sample data to obtain electroencephalogram signals D without the alpha waves, and determining alpha wave rhythms of all the electroencephalogram signals according to differences between all the electroencephalogram signals in the sample data and the corresponding electroencephalogram signals D.
In the embodiment, a set of sine signal and cosine signal with frequencies of 8.5Hz, 9.5Hz, 10.5Hz, 11.5Hz and 12.5Hz is constructed, and 10 signals in total are used as reference signals of alpha wave. The 10 signals and the original electroencephalogram signal form an input end of an independent component analysis method, and then 11 input signals are separated by using a FastICA algorithm so as to separate alpha waves in the original electroencephalogram signal (each electroencephalogram signal in sample data) to obtain an electroencephalogram signal D without the alpha waves. The difference between the D signal and the original resting state electroencephalogram signal obtains an alpha wave component to ensure the accuracy of the determined alpha wave rhythm.
In one embodiment, integrating the alpha-wave rhythm into a brain electrical signal deep learning framework for training, and obtaining the initial model comprises:
integrating alpha wave pitch into brain electric signal deep learning frame, converting alpha wave pitch into corresponding picture in brain electric signal deep learning frame to obtain alpha wave pitch picture, making brain electric signal alpha wave pitch picture data be recognized as array arranged according to pixels by computer, filtering the characteristics of each possible position of a data picture by adding a weight coefficient to a convolutional layer of an electroencephalogram deep learning frame to obtain a filtered array, adding a nonlinear factor through an activation function, performing nonlinear mapping on the output result of the convolutional layer by adopting the nonlinear factor, pooling the non-linearly mapped matrix in a pooling layer to obtain a pooled matrix, performing weighted summation on the pooled matrix through a full-link layer, and identifying and classifying the obtained result, and determining an initial model according to the operating parameters of the electroencephalogram signal deep learning frame when the identification and classification precision reaches the set precision.
Specifically, in this embodiment, the obtained α -node law is integrated into a deep learning framework of electroencephalogram signals for depression to perform training, and the deep learning framework adopts Apache MXnet, which specifically includes:
1) data sample conversion format: and converting 100 parts of alpha wave law data obtained by an independent component analysis method into pictures, wherein the size of each picture is 128 × 128, and expanding each sample data by two times by using a linear interpolation method.
2) Selecting a model: and according to the sample condition, selecting a single-input single-output sequential model.
3) Constructing a network layer: including an input layer, an output layer, and a hidden layer. Each layer includes a common layer, a convolutional layer, a pooling layer, a fully-connected layer, and an activation function. Wherein the first, second, fourth and sixth layers are convolutional layers, the third, fifth, seventh, eighth and ninth layers are pooling layers, and the tenth and eleventh layers are all-connected layers.
4) Training: the electroencephalogram alpha wave pitch picture data are identified to be an array arranged according to pixels by a computer, and the characteristics of each possible position of the data picture are filtered at the convolutional layer by adding a weight coefficient to obtain a filtered array. And adding a nonlinear factor through the activation function, and performing nonlinear mapping on the convolution layer output result. And performing pooling treatment on the matrix subjected to the nonlinear mapping in a pooling layer to obtain a pooled matrix. And carrying out weighted summation on the obtained pooling matrix through the full-connection layer, and identifying and classifying the obtained result. The classification result is as follows: data smaller than 0 are not output, and data larger than 0 are output as original values.
Further, constructing the network layer includes:
1) storing the processed EEG signal alpha wave pitch picture sample data in a computer;
2) forward propagation: forward propagation refers to the process of data transmission from X to the neural network, through the various hidden layers, with eventual loss. The input value of the second layer is the output value of the first layer, and the forward propagation process is to calculate Z first[1]Then calculate A[1]Then calculate Z[2]Then calculate A[2]，......A[L]。
3) And (3) back propagation: the back propagation is mainly carried out in the optimization process of the neural network, a total loss function is calculated at the L end, and then forward feedback is carried out layer by layer according to a gradient decreasing formula to form a back propagation mechanism.
First calculating the last layer, i.e. the second hidden layerParameters and corresponding values of the layers: dZ[L]、dw[L]、db[L]I.e. the chain derivation method commonly used in neural networks, the value of each layer is calculated forward layer by layer based on the total loss L.
4) Model optimization: using Adam's algorithm, the first and second order momentum initiatives were set to 0, with decay rates of 0.99 and 0.999. Penalties are introduced using L2 regularization to penalize all trainable parameters in the model, reducing the impact of overfitting on the model.
5) And obtaining a trained electroencephalogram signal deep learning model (initial model) which is used as a basis for identifying and judging the target electroencephalogram signal and storing the model in a computer.
The electroencephalogram signal deep learning model (initial model) judges the electroencephalogram signal of the identification target, wherein the result I is that the identification signal and the sample signal belong to similar signals, and the result II is that the identification signal and the sample signal are not similar signals.
In one embodiment, extracting the feature parameters of each electroencephalogram signal in the sample data by using the power spectrum estimation of the feature vector method comprises:
and performing spectrum estimation calculation on each electroencephalogram signal by adopting an MUSIC spectrum estimation method to obtain the maximum value, the minimum value, the average value and the standard deviation of the amplitude of the power spectrum as characteristic parameters of each electroencephalogram signal.
In this embodiment, a feature vector method power spectrum estimation is used to extract electroencephalogram characteristic parameters, specifically, a MUSIC spectrum estimation method in a feature vector method power spectrum is used to perform spectrum estimation calculation on acquired electroencephalograms, so as to obtain a maximum value, a minimum value, an average value and a standard deviation of power spectrum amplitudes as characteristic parameters of classified signals, and the formed characteristic vectors are used as input of a support vector machine.
Further, the MUSIC spectrum estimation method specifically includes:
1) the covariance matrix of the array matrix is R ═ E [ XX [ ]H]＝AE[SSH]AH+δ2I＝ARSAH+δ2I；ARSAHIs a signal portion;
2) performing characteristic decomposition on R to obtain R ═ USVSUS H+UNVNUN H(ii) a Formula antecedent USThe consequent U is a signal subspace grown for large eigenvalues corresponding to the eigenvectorsNA noise subspace spanned by the corresponding feature vectors of the small noise feature values;
3) the spectrum estimation formula of the MUSIC algorithm is as follows
Figure BDA0002418949200000091
Figure BDA0002418949200000092
Is a noise subspace feature vector matrix, aH(θ) is the steering vector in the signal subspace.
Further, the support vector machine may specifically use Matlab Toolbox. And taking the characteristic parameters as input vectors of the classifier, selecting proper kernel functions and parameters to construct a high latitude space, and obtaining an optimal classification surface, so that the input characteristic vectors can be linearly classified. The output result is as follows: the result one is a signal with similar components, the result two is a substantially similar signal, the result three is a signal of the same kind, and the result four is a signal of the same kind.
In an embodiment, a propagation diagram of a deep learning neural network may be shown in fig. 2, and a flowchart of the electroencephalogram signal identification method based on machine learning may also be shown in fig. 3; the corresponding electroencephalogram signal identification and monitoring system based on machine learning comprises an electroencephalogram signal deep learning model, identification based on a certain component, further classification of an identified target signal and the like.
The specific implementation steps comprise:
the method comprises the following steps: taking the example that a certain component is the rhythm of the electroencephalogram signal alpha wave (the actual operation process can select any electroencephalogram signal component which can be converted into a signal image): the method comprises the steps of selecting a sample electroencephalogram signal as a sample (sample data) for identification and detection, wherein the number of the samples is 100, and the length is 10 minutes. Each sample has the same or similar characteristics. Denoising and standardizing the electroencephalogram signals in the sample data, selecting intermediate 300 seconds of data, extracting alpha wave rhythm by using an independent component analysis method, and integrating the obtained rhythm into an electroencephalogram signal deep learning framework for training to obtain an electroencephalogram signal deep learning model.
And step two, judging the 10-minute electroencephalogram signal rhythm of the recognition target based on the established electroencephalogram signal deep learning model. If the judgment result is that the rhythm of the electroencephalogram signal is consistent with the rhythm of the sample electroencephalogram signal, the alpha wave pitch of the identified target is sent to the electroencephalogram signal deep learning frame for retraining so as to further perfect the model and improve the accuracy of the model. If the result is a different rhythm from the sample signal, it is discarded as unused. New samples can be continuously added in the actual operation process, and the identification accuracy is improved along with the increase of the sample library
And thirdly, extracting the characteristic parameters of the sample electroencephalogram signals by using power spectrum estimation of a characteristic vector method, and determining the range of the characteristic parameters as the basis for further classification of the target electroencephalogram signals. The data set formed by the characteristic parameters is divided into four reference intervals, namely a first reference interval, a second reference interval, a third reference interval and a fourth reference interval. The characteristic parameters of the target electroencephalogram signals can be distinguished by using four references to obtain a distinguishing result. The four reference intervals are stored in the computer.
And step four, carrying out electroencephalogram signal acquisition for more time lengths on the identification target consistent with the result sample in the step two, and transmitting the acquired data to a computer. Segmenting the electroencephalogram signal according to the time length of 10 minutes, and estimating and extracting the characteristic parameters of the electroencephalogram signal in each segment by adopting a characteristic vector method power spectrum. And comparing the extracted characteristic parameters with the first reference interval, the second reference interval, the third reference interval and the fourth reference interval by using a support vector machine, and judging the similarity degree of the recognition target and the sample according to the interval where the characteristic parameters are located. If the output result is in the third reference interval, the output result is similar in basic characteristics (basically similar signals); if the signal is in the interval II, the output result is that most features are similar (like signals); if the signal is in the interval one, the output result is that almost all the characteristics are similar (the same signal).
The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.
It should be noted that the terms "first \ second \ third" referred to in the embodiments of the present application merely distinguish similar objects, and do not represent a specific ordering for the objects, and it should be understood that "first \ second \ third" may exchange a specific order or sequence when allowed. It should be understood that "first \ second \ third" distinct objects may be interchanged under appropriate circumstances such that the embodiments of the application described herein may be implemented in an order other than those illustrated or described herein.
The terms "comprising" and "having" and any variations thereof in the embodiments of the present application are intended to cover non-exclusive inclusions. For example, a process, method, apparatus, product, or device that comprises a list of steps or modules is not limited to the listed steps or modules but may alternatively include other steps or modules not listed or inherent to such process, method, product, or device.
The above-mentioned embodiments only express several embodiments of the present application, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the concept of the present application, which falls within the scope of protection of the present application. Therefore, the protection scope of the present patent application shall be subject to the appended claims.

Claims (6)

1. A machine learning-based electroencephalogram signal identification method is characterized by comprising the following steps:
s10, constructing sample data comprising a plurality of electroencephalograms, extracting alpha wave rhythms of the electroencephalograms in the sample data after denoising and normalizing the sample data, and integrating the alpha wave rhythms into an electroencephalogram deep learning frame for training to obtain an initial model;
s20, judging the rhythm of the target electroencephalogram signal by adopting an initial model, and if the judgment result is that the identification signal and the sample signal belong to similar signals, inputting the alpha-wave pitch of the target electroencephalogram signal into the initial model for retraining to obtain an electroencephalogram signal deep learning model;
s30, extracting characteristic parameters of each electroencephalogram signal in sample data by adopting characteristic vector method power spectrum estimation, and dividing a data set formed by the characteristic parameters into a first reference interval, a second reference interval, a third reference interval and a fourth reference interval; the first reference interval is a signal with similar components, the second reference interval is a basically similar signal, the third reference interval is a signal of the same kind, and the fourth reference interval is a signal of the same kind;
and S40, carrying out electroencephalogram signal acquisition for more time duration on the identification signal consistent with the sample signal, transmitting the acquired data to a computer, segmenting the data, extracting electroencephalogram signal characteristic parameters of each segment, comparing the electroencephalogram signal characteristic parameters with a first reference interval, a second reference interval, a third reference interval and a fourth reference interval, and judging the similarity degree of the identification target and the sample according to the interval where the characteristic parameters are located.
2. The machine learning-based electroencephalogram signal identification method according to claim 1, further comprising, after constructing sample data including a plurality of electroencephalogram signals:
denoising each electroencephalogram signal of the sample data by using filters with cut-off frequencies of 0.5Hz and 40 Hz;
and sequentially carrying out data normalization processing on the sample data after the de-noising processing on all sampling points corresponding to a single electrode by using z-score standardization, and converting the sampling points corresponding to each electrode into data distribution with the mean value of 0 and the standard deviation of 1 so as to realize the normalization processing of the sample data.
3. The machine learning-based electroencephalogram signal identification method according to claim 1, wherein extracting the alpha-wave pitch of each electroencephalogram signal in the sample data comprises:
the method comprises the steps of constructing sine signals with frequencies of 8.5Hz, 9.5Hz, 10.5Hz, 11.5Hz and 12.5Hz and cosine signals with frequencies of 8.5Hz, 9.5Hz, 10.5Hz, 11.5Hz and 12.5Hz in a group respectively, using the two groups of signals as reference signals of alpha waves, enabling the two groups of signals and sample data to form input ends of an independent component analysis method, separating input signals of the input ends by a FastICA algorithm to separate alpha waves of all electroencephalogram signals in the sample data to obtain electroencephalogram signals D without the alpha waves, and determining alpha wave rhythms of all the electroencephalogram signals according to differences between all the electroencephalogram signals in the sample data and the corresponding electroencephalogram signals D.
4. The machine learning-based electroencephalogram signal identification method according to claim 1, wherein the step of integrating the alpha-wave pitch into an electroencephalogram signal deep learning framework for training to obtain an initial model comprises the following steps:
integrating alpha wave pitch into brain electric signal deep learning frame, converting alpha wave pitch into corresponding picture in brain electric signal deep learning frame to obtain alpha wave pitch picture, making brain electric signal alpha wave pitch picture data be recognized as array arranged according to pixels by computer, filtering the characteristics of each possible position of a data picture by adding a weight coefficient to a convolutional layer of an electroencephalogram deep learning frame to obtain a filtered array, adding a nonlinear factor through an activation function, performing nonlinear mapping on the output result of the convolutional layer by adopting the nonlinear factor, pooling the non-linearly mapped matrix in a pooling layer to obtain a pooled matrix, performing weighted summation on the pooled matrix through a full-link layer, and identifying and classifying the obtained result, and determining an initial model according to the operating parameters of the electroencephalogram signal deep learning frame when the identification and classification precision reaches the set precision.
5. The machine learning-based electroencephalogram signal identification method according to claim 1, wherein extracting the characteristic parameters of each electroencephalogram signal in sample data by adopting a characteristic vector method power spectrum estimation comprises:
and performing spectrum estimation calculation on each electroencephalogram signal by adopting an MUSIC spectrum estimation method to obtain the maximum value, the minimum value, the average value and the standard deviation of the power spectrum amplitude as characteristic parameters of each electroencephalogram signal.
6. The machine learning-based electroencephalogram signal identification method of claim 1, wherein the length of each electroencephalogram signal of sample data is 10 minutes.
CN202010199754.XA 2020-03-20 2020-03-20 Electroencephalogram signal identification method based on machine learning Active CN111387975B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010199754.XA CN111387975B (en) 2020-03-20 2020-03-20 Electroencephalogram signal identification method based on machine learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010199754.XA CN111387975B (en) 2020-03-20 2020-03-20 Electroencephalogram signal identification method based on machine learning

Publications (2)

Publication Number Publication Date
CN111387975A CN111387975A (en) 2020-07-10
CN111387975B true CN111387975B (en) 2022-06-17

Family

ID=71410943

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010199754.XA Active CN111387975B (en) 2020-03-20 2020-03-20 Electroencephalogram signal identification method based on machine learning

Country Status (1)

Country Link
CN (1) CN111387975B (en)

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111956241A (en) * 2020-08-14 2020-11-20 北京脑陆科技有限公司 Psychological stress detection method based on EEG signal
CN112257658B (en) * 2020-11-11 2023-10-10 微医云(杭州)控股有限公司 Electroencephalogram signal processing method and device, electronic equipment and storage medium
CN113499087B (en) * 2021-06-08 2023-06-06 宁波大学 Electroencephalogram signal feature extraction method based on ITD modal parameter identification
CN113576493A (en) * 2021-08-23 2021-11-02 安徽七度生命科学集团有限公司 User state identification method for health physiotherapy cabin

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106821376A (en) * 2017-03-28 2017-06-13 南京医科大学 A kind of epileptic attack early warning system and method based on deep learning algorithm
CN107049239A (en) * 2016-12-28 2017-08-18 苏州国科康成医疗科技有限公司 Epileptic electroencephalogram (eeg) feature extracting method based on wearable device
CN107196809A (en) * 2017-07-07 2017-09-22 南京邮电大学 Identity identifying method and Verification System based on brain electrical feature
CN108776788A (en) * 2018-06-05 2018-11-09 电子科技大学 A kind of recognition methods based on brain wave
CN109359610A (en) * 2018-10-26 2019-02-19 齐鲁工业大学 Construct method and system, the data characteristics classification method of CNN-GB model
CN109948427A (en) * 2019-01-24 2019-06-28 齐鲁工业大学 A kind of idea recognition methods based on long memory models in short-term
CN110013248A (en) * 2018-01-08 2019-07-16 上海交通大学 Brain electricity tensor mode identification technology and brain-machine interaction rehabilitation system
CN110840432A (en) * 2019-12-02 2020-02-28 苏州大学 Multichannel electroencephalogram epilepsy automatic detection device based on one-dimensional CNN-LSTM

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107049239A (en) * 2016-12-28 2017-08-18 苏州国科康成医疗科技有限公司 Epileptic electroencephalogram (eeg) feature extracting method based on wearable device
CN106821376A (en) * 2017-03-28 2017-06-13 南京医科大学 A kind of epileptic attack early warning system and method based on deep learning algorithm
CN107196809A (en) * 2017-07-07 2017-09-22 南京邮电大学 Identity identifying method and Verification System based on brain electrical feature
CN110013248A (en) * 2018-01-08 2019-07-16 上海交通大学 Brain electricity tensor mode identification technology and brain-machine interaction rehabilitation system
CN108776788A (en) * 2018-06-05 2018-11-09 电子科技大学 A kind of recognition methods based on brain wave
CN109359610A (en) * 2018-10-26 2019-02-19 齐鲁工业大学 Construct method and system, the data characteristics classification method of CNN-GB model
CN109948427A (en) * 2019-01-24 2019-06-28 齐鲁工业大学 A kind of idea recognition methods based on long memory models in short-term
CN110840432A (en) * 2019-12-02 2020-02-28 苏州大学 Multichannel electroencephalogram epilepsy automatic detection device based on one-dimensional CNN-LSTM

Also Published As

Publication number Publication date
CN111387975A (en) 2020-07-10

Similar Documents

Publication Publication Date Title
CN111387975B (en) Electroencephalogram signal identification method based on machine learning
Yalcin et al. Epilepsy diagnosis using artificial neural network learned by PSO
CN114052735B (en) Deep field self-adaption-based electroencephalogram emotion recognition method and system
Sharmila et al. Wavelet-based feature extraction for classification of epileptic seizure EEG signal
Ashokkumar et al. RETRACTED: Implementation of deep neural networks for classifying electroencephalogram signal using fractional S‐transform for epileptic seizure detection
CN111310570B (en) Electroencephalogram signal emotion recognition method and system based on VMD and WPD
KR102141185B1 (en) A system of detecting epileptic seizure waveform based on coefficient in multi-frequency bands from electroencephalogram signals, using feature extraction method with probabilistic model and machine learning
CN112766355B (en) Electroencephalogram signal emotion recognition method under label noise
CN110367967A (en) A kind of pocket lightweight human brain condition detection method based on data fusion
Malviya et al. A novel technique for stress detection from EEG signal using hybrid deep learning model
Mohammadi et al. Discrimination of depression levels using machine learning methods on EEG signals
CN112932501B (en) Method for automatically identifying insomnia based on one-dimensional convolutional neural network
CN114595725B (en) Electroencephalogram signal classification method based on addition network and supervised contrast learning
Putra et al. EEG-based emotion classification using wavelet decomposition and K-nearest neighbor
Al-dabag et al. EEG motor movement classification based on cross-correlation with effective channel
Bhandari et al. Emotion recognition and classification using EEG: A review
Guan et al. A non-contact paraparesis detection technique based on 1D-CNN
Pan et al. Recognition of human inner emotion based on two-stage FCA-ReliefF feature optimization
Vrbancic et al. Automatic detection of heartbeats in heart sound signals using deep convolutional neural networks
Kanna et al. Smart Detection and Removal of Artifacts in Cognitive Signals Using Biomedical Signal Intelligence Applications
Bhanumathi et al. Feedback Artificial Shuffled Shepherd Optimization‐Based Deep Maxout Network for Human Emotion Recognition Using EEG Signals
Yuan et al. Automatic seizure detection using logarithmic Euclidean-Gaussian mixture models (LE-GMMs) and improved deep forest learning
Chalaki et al. Epileptic seizure classification using ConvLSTM deep classifier and rotation short-time Fourier Transform
CN117786497A (en) Brain electricity emotion recognition method and system based on colistin predation optimization algorithm
Wankhade et al. IKKN predictor: An EEG signal based emotion recognition for HCI

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant