Nothing Special   »   [go: up one dir, main page]

CN114861702A - Noise identification method for EEG data by using deep neural network - Google Patents

Noise identification method for EEG data by using deep neural network Download PDF

Info

Publication number
CN114861702A
CN114861702A CN202210307716.0A CN202210307716A CN114861702A CN 114861702 A CN114861702 A CN 114861702A CN 202210307716 A CN202210307716 A CN 202210307716A CN 114861702 A CN114861702 A CN 114861702A
Authority
CN
China
Prior art keywords
convolution
layer
filters
step length
eeg 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.)
Pending
Application number
CN202210307716.0A
Other languages
Chinese (zh)
Inventor
邹凌
杨亮宇
周天彤
李文杰
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.)
Changzhou University
Original Assignee
Changzhou University
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 Changzhou University filed Critical Changzhou University
Priority to CN202210307716.0A priority Critical patent/CN114861702A/en
Publication of CN114861702A publication Critical patent/CN114861702A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • 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
    • 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
    • A61B5/25—Bioelectric electrodes therefor
    • A61B5/251—Means for maintaining electrode contact with the body
    • A61B5/256—Wearable electrodes, e.g. having straps or bands
    • 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]
    • A61B5/377—Electroencephalography [EEG] using evoked responses
    • 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
    • A61B5/7207—Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal of noise induced by motion artifacts
    • 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
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G06F18/2134—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on separation criteria, e.g. independent component analysis
    • 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
    • G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • G06F18/24133—Distances to prototypes
    • G06F18/24137—Distances to cluster centroïds
    • G06F18/2414—Smoothing the distance, e.g. radial basis function networks [RBFN]
    • 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
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/047—Probabilistic or stochastic 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
    • 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
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Data Mining & Analysis (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Biophysics (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Biomedical Technology (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Surgery (AREA)
  • Public Health (AREA)
  • Mathematical Physics (AREA)
  • Veterinary Medicine (AREA)
  • Animal Behavior & Ethology (AREA)
  • Medical Informatics (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Pathology (AREA)
  • Signal Processing (AREA)
  • Psychiatry (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computational Linguistics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Probability & Statistics with Applications (AREA)
  • Physiology (AREA)
  • Psychology (AREA)
  • Fuzzy Systems (AREA)
  • Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)

Abstract

The invention relates to the technical field of neural network algorithms, in particular to a noise identification method for EEG data by using a deep neural network, which comprises S1, collecting EEG data induced during the combined action of a tested stroke joint through an EEG cap; s2, performing down-sampling, coarse filtering and independent component decomposition on the data in the data set; s3, cutting the IC topographic map; s4, sending the IC topographic map into a two-dimensional CNN convolution network, sending the IC activation into a one-dimensional convolution network, and sending the output values of the two convolution networks into a DNN convolution network; and S5, classifying the output value of the DNN convolutional network through a softmax classifier, and performing noise elimination on the EEG data. According to the invention, the deep neural network is built by utilizing the one-dimensional CNN convolution, the two-dimensional CNN convolution and the DDN convolution, so that noise components can be effectively classified, the sensitivity to EEG noise collected in real time is high, and the noise components can be effectively removed.

Description

Noise identification method for EEG data by using deep neural network
Technical Field
The invention relates to the technical field of neural network algorithms, in particular to a noise identification method for EEG data by utilizing a deep neural network.
Background
Motor Imagery (MI) refers to the use of the brain to subjectively visualize movement of limbs or muscles without objective limb or muscle movement, thereby stimulating the corresponding brain region. At present, MI is mainly applied to training and treating certain limb dysfunction diseases such as rehabilitation after stroke and muscular atrophy.
The currently common MI paradigm is to imagine the movements of left and right limbs, but the patients after stroke are mostly unilateral dyskinesia. Therefore, the classic MI paradigm is somewhat unsatisfactory for the training and treatment of the affected limb of a patient by a physician. Meanwhile, in order to obtain an Electroencephalogram (EEG) in a good MI paradigm, an experienced technician is required to perform operations such as filtering, independent component analysis, re-referencing, etc. on the original signal, but limb rehabilitation requires feature extraction and classification of the EEG in real time for controlling assistive devices such as exoskeletons.
An automatic denoising algorithm for the MI paradigm for single-limb dyskinesia after stroke and EEG signals is thus generated. Considering that the movement function of the patient after the stroke has a large obstacle, the action required to be imagined in the paradigm should be difficult to enter. With the rise of deep learning, there are many research results for applying neural networks to denoising of EEG, however, the improvement of signal quality cannot be maximized only from the time domain consideration of EEG, and meanwhile, the selection of signal length also leads to different results.
Disclosure of Invention
Aiming at the defects of the existing algorithm, the invention designs an MI paradigm suitable for single limb dyskinesia after apoplexy, which mainly carries out individual training and linkage training on joints of the shoulder, elbow, wrist and finger of the right upper limb of a patient; and a noise identification method for EEG data by utilizing a deep neural network is established by utilizing one-dimensional convolution, two-dimensional convolution and dense connection convolution, and noise elimination is carried out on the EEG data in a window dividing mode.
The normal form design is according to the demand of unilateral dyskinesia patient behind the cerebral apoplexy, designs a MI normal form that is applicable to unilateral dyskinesia behind the apoplexy, and the auxiliary disease carries out the limbs training of different degrees, and the recovered branch of academic or vocational study doctor carries out upper limbs movement function aassessment to the apoplexy patient through brunstrom hemiplegia functional evaluation method, Fugl-Meyer method, upper field sensitivity method, selects suitable normal form by the nimble degree of joint of score diagnosis disease after the aassessment.
EEG data processing comprises down sampling, filtering, re-referencing and ICA analysis, wherein a part needing human intervention is changed into automatic data processing through a deep neural network, and the human intervention is that an experienced technician is required to examine the ICA result during ICA analysis, noise considered by the technician is judged, and the influence caused by subjective consciousness is reduced through a deep learning method.
The technical scheme adopted by the invention is as follows: a noise identification method for EEG data by using a deep neural network comprises the following steps:
s1, collecting EEG data induced by the combined action of four joints of a shoulder, an elbow, a wrist and a finger in a tested stroke state through an EEG cap, and dividing the action into single joint training and joint comprehensive training;
further, the single joint training is to stimulate a brain movement function area to generate EEG data by seven actions of stimulating sources of shoulder flexion 30 degrees, shoulder extension 30 degrees, elbow flexion 90 degrees, elbow flexion maximum, wrist left rotation, wrist right rotation and finger fist grasping; the multi-joint comprehensive training stimulates a brain movement function area to generate EEG data by eight actions of stimulating sources of forearm pronation, forearm supination, upper limb forward flexion of 90 degrees, upper limb backward extension upward lifting and touching waist, upper limb abduction of 90 degrees, upper limb forward flexion of 180 degrees, shoulder touching experiment and finger nose experiment;
s2, performing down-sampling, coarse filtering and ICA independent component decomposition on data in the data set to obtain IC activation and IC topographic map of a plurality of independent components;
s3, cutting the IC topographic map, dividing the IC activation into certain windows according to certain time, and setting the overlapping time of adjacent windows;
s4, constructing a deep neural network consisting of one-dimensional CNN convolution, two-dimensional CNN convolution and DDN convolution, sending the IC topographic map into the two-dimensional CNN convolution network, activating the IC and sending the IC topographic map into the one-dimensional convolution network, sending output values of the two convolution networks into the DNN convolution network, and setting a hyper-parameter;
further, the one-dimensional CNN convolution includes: a first convolution layer, wherein the width of an inner core is 20, the number of filters is 5, and the step length is 2; a second convolution layer, the width of the inner core is 6, the number of the filters is 16, and the step length is 2; a third convolution layer, the width of the kernel is 15, the number of the filters is 34, and the step length is 1; a fourth convolution layer, the width of the kernel is 11, the number of the filters is 21, and the step length is 1; a fifth convolution layer, the width of the kernel is 6, the number of the filters is 45, and the step length is 1; a sixth convolution layer, the kernel width is 7, the number of filters is 30, and the step length is 1; the first, second, third, fifth and sixth convolution layers are connected with the pooling layer with the kernel width of 4 and the step length of 2; the activation functions are all PreLU functions; finally, accessing a Flatten layer for dimension reduction to obtain a characteristic column vector activated by the IC;
further, the two-dimensional CNN convolution includes: the first convolution layer of the IC map, the size of an inner core is 20 x 20, the number of filters is 40, and the step length is 2 x 2; the second convolution layer of the IC map has the kernel size of 5 × 5, the number of filters is 50, and the step length is 2 × 2; the size of an inner core of the third convolution layer of the IC map is 6 x 6, the number of filters is 14, and the step length is 1 x 1; the fourth convolution layer of the IC map has the kernel size of 9 × 9, the number of filters of 35 and the step size of 2 × 2; the fifth convolution layer of the IC map has the kernel size of 12 × 12, 21 filters and the step length of 1 × 1; the sixth convolution layer of the IC map has the size of 16 × 16 inner cores, the number of filters is 45, and the step length is 2 × 2; the seventh convolution layer of the IC map, the size of the inner core is 8 × 8, the number of the filters is 30, and the step length is 2 × 2; the second, third, fifth and seventh convolution layers of the IC map are connected with the pooling layers with the size of 2 x 2 and the step length of 1 x 1; the activation functions are all PreLU functions; and finally, accessing a Flatten layer for dimension reduction to obtain a characteristic column vector of the IC topographic map.
Further, the DDN convolution comprises a first fully connected layer, a second fully connected layer, and a third fully connected, the first fully connected output dimension being 135, the second fully connected output dimension being 210, and the third fully connected output dimension being 28;
s5, classifying the output value of the DNN convolutional network through a softmax classifier, screening and eliminating noise of EEG data according to the classification result, calculating the signal-to-noise ratio and the root-mean-square difference of the EEG data with the noise eliminated, and evaluating the denoising effect.
The classification results are eye movement noise, electrocardio noise, head movement noise and myoelectricity noise;
and (3) taking the EEG data as a detection window according to the time length of 5000ms, taking 4 types of noise classified by a softmax classifier as detection indexes, carrying out step detection on the EEG data according to the detection window, screening out a time window containing noise components, and removing the time window.
The invention has the beneficial effects that:
1. the designed single-limb motor imagery paradigm can assist patients after stroke in limb rehabilitation training, and is beneficial to functional recovery of brain motor areas; meanwhile, different normal forms can be selected according to different post-stroke rehabilitation processes to form a hierarchical rehabilitation training system;
2. the deep neural network is built by utilizing the one-dimensional CNN convolution, the two-dimensional CNN convolution and the DDN convolution, so that noise components can be effectively classified, the sensitivity to EEG noise collected in real time is high, and the noise components can be effectively removed;
3. the method is used for assisting cerebral apoplexy patients with different degrees to carry out rehabilitation training, and obtaining real-time noiseless EEG data of the patients during the rehabilitation training.
Drawings
FIG. 1 is a method of noise identification of EEG data using a deep neural network in accordance with the present invention;
FIG. 2 is a schematic time flow diagram of two single-limb motor imagination paradigms designed by the present invention;
FIG. 3 is a block diagram of a data communication and format conversion module;
FIG. 4 is a flow chart of an embodiment of the present invention.
Detailed Description
The invention will be further described with reference to the accompanying drawings and examples, which are simplified schematic drawings and illustrate only the basic structure of the invention in a schematic manner, and therefore only show the structures relevant to the invention.
The paradigm design part of the invention is developed and finished by using JavaScript language in Windows environment, and the deep neural network part uses Pycharm software and GPU computational power unit with Keras 2.4.0 and Tensorflow 2.4.1 deep learning environment.
As shown in fig. 2, the tested hand is naturally placed on the thigh, the eyes are 1 m away from the screen, and the main test selects single joint training or multi-joint comprehensive training according to the function status of the tested limb;
training a single joint: the stimulus sources are divided into seven actions of shoulder flexion 30 degrees, shoulder extension 30 degrees, elbow flexion 90 degrees, elbow flexion maximum, wrist left rotation, wrist right rotation and finger fist making, 2000ms of prompt is provided before the stimulus sources appear, each stimulus source randomly appears 4000ms, 2000ms of rest enters the next 2000ms of prompt after the stimulus sources are finished, each action of each round of training randomly appears 20 times, the total number is 140 times, and the whole rehabilitation training paradigm is 5 rounds in total.
Multi-joint comprehensive training: for single joint training paradigm, the multi-joint comprehensive paradigm combines four joints, has formed 8 compound actions such as pointing the nose experiment, touching the shoulder and raising. After the 2000ms prompt, the multi-joint action guide GIF map appears 6000ms as the stimulus source (two times of circular playing), and the same 2000ms rest is generated after the end. In contrast, due to the complexity of the training movements, each movement occurs randomly 10 times for each training round, for a total of 80 times, and the whole rehabilitation training paradigm has 5 rounds.
As shown in fig. 3, the communication module in the data communication and format conversion module is connected with the signal output end of the electroencephalogram cap and the input end of the data analysis module; the communication module transmits the data acquired by the electroencephalogram cap hardware to the data analysis module through a socket protocol, so that the online processing of the data is realized; the format conversion is mainly to convert the original data of 64 or 128 leads into data files of the format of raw, mff or bdf, so that the subsequent reading and storing are convenient.
As shown in fig. 1, a method for recognizing noise in EEG data by using a deep neural network includes the following main steps:
s1, collecting EEG data induced by the combined action of the four joints of the shoulder, elbow, wrist and finger under the state of the tested apoplexy through an EEG cap,
s2, firstly, down-sampling collected data to 250Hz, then, carrying out coarse filtering on the data stream at 0.1-50 Hz, and then, carrying out Independent Component Analysis (ICA), wherein the number of Independent Components (ICs) is 8-24 according to different lead numbers, and 12 ICs are selected according to 64 lead numbers in the invention to obtain the activation and topographic map of each Component;
s3, cutting the topographic map of 12 ICs to 160 pixels by 3 colors;
the 12 IC activations are divided into a plurality of trails (trial times) according to the time of 8000ms, two adjacent trails are overlapped for 2000ms, and when the data length is less than 8000ms, the two adjacent trails are automatically overlapped forwards for 8000 ms.
S4, selecting hyper-parameters required for constructing a Neural network, and constructing a deep Neural network including a one-dimensional Convolutional Neural Network (CNN), a two-dimensional CNN, and a Dense Convolutional network (densneet), specifically:
the two-dimensional CNN structure used to extract features from IC topography is:
layer 1: the convolution layer has the kernel size of 20 × 20, the number of filters is 40, the step size is 2 × 2, and the activation function is a PreLU and is used for improving the overfitting of the model;
layer 2: the convolution Layer, the size of the kernel is 5 × 5, the number of the filters is 50, the step size is 2 × 2, the activation function is also set as PreLU, a maximum pooling Layer is added after Layer2, the size is 2 × 2, and the step size is 1 × 1;
layer 3: the convolution Layer with PreLU has kernel size of 6 × 6, filter number of 14, step size of 1 × 1, and maximum pooling Layer with the same size as Layer2 is added after the convolution Layer;
layer 4: the convolution layer with PreLU has the kernel size of 9 × 9, the number of filters of 35 and the step size of 2 × 2;
layer 5: the convolution Layer with PreLU has kernel size of 12 × 12, filter number of 21, step size of 1 × 1, and maximum pooling Layer, and size identical to Layer 3;
layer 6: the convolution layer with PreLU has the kernel size of 16 × 16, the number of filters of 45 and the step size of 2 × 2;
layer 7: the convolution layer with PreLU has inner core size of 8 × 8, filter number of 30, step length of 2 × 2, and one maximum pooling layer with pooling size of 2 × 2 and step length of 2 × 2;
layer 8: after the IC topographic map passes through Layer7, reducing the dimension of the result to obtain a characteristic column vector of the IC topographic map;
the one-dimensional CNN structure used for extracting features for IC time activation is:
layer 1: the convolutional layer, the activation function is PreLU, the kernel width is 20, the number of filters is 5, the step length is 2, and then a maximum pooling layer with the kernel width of 4 and the step length of 2 is connected;
layer 2: the convolution Layer with PreLU has the kernel width of 6, the number of filters of 16 and the step length of 2, and is followed by the maximum pooling Layer which is the same as Layer 1;
layer 3: the convolution Layer with PreLU has the kernel width of 15, the number of filters of 34 and the step length of 1, and is followed by the maximum pooling Layer which is the same as Layer 1;
layer 4: the convolution layer with PreLU has a kernel width of 11, a filter number of 21 and a step length of 1;
layer 5: the convolution Layer with PreLU has the kernel width of 6, the number of filters of 45 and the step length of 1, and is followed by the maximum pooling Layer which is the same as Layer 1;
layer 6: the convolutional Layer with PreLU has a kernel width of 7, the number of filters of 30, a step size of 1, and is followed by the same maximum pooling Layer as Layer 1.
Layer 7: and in a Flatten Layer, after the time activation of the IC passes through a Layer6 of the one-dimensional CNN, reducing the dimension of the result to obtain a characteristic column vector of the time activation of the IC.
Combining the features of the IC topographic map and the time activated features to obtain a new feature vector; inputting the data into a Dense layer, wherein a Dense network is specifically composed as follows:
layer 1: a full connection layer, which further processes the merged feature vectors and outputs dimension 135;
layer 2: fully connected layer, output dimension 210;
layer 3: and (4) outputting a result by a full connection layer and a softmax after outputting the dimension 28, wherein the output dimension is 4.
S5, classifying the output value of the DNN convolutional network through a softmax classifier, screening and eliminating noise of EEG data according to the classification result, calculating the signal-to-noise ratio and the root-mean-square difference of the EEG data with the noise eliminated, and evaluating the denoising effect.
And (3) testing the denoising effect: after EEG data is subjected to noise screening and elimination of a deep neural network, a proper index is selected to evaluate the denoising effect of the EEG data; the denoising effect is evaluated mainly by considering the occupation ratio of noise in the whole signal and the deviation before and after denoising;
the noise removal effect of the invention is evaluated by adopting the SNR and the RMSD, and the formula is as follows:
Figure BDA0003566786750000081
Figure BDA0003566786750000082
in the formula, x eeg [n]Representing the effective component, x, in the de-noised brain wave noise [n]Representing the noise components in the de-noised brain wave,
Figure BDA0003566786750000091
representing the power of the effective components in the denoised electroencephalogram,
Figure BDA0003566786750000092
power, V, representing noise components in denoised brain electrical eeg Representing effective components in denoised brain electricityEffective value of pressure, V noise Representing the effective voltage value of noise components in the denoised electroencephalogram; EEG (electroencephalogram) i Representing the de-noised signal, eeg i Representing the electroencephalogram signals before denoising;
the SNR represents the ratio of effective components to noise components in the denoised signal, and the larger the ratio, the more effective components are contained; the RMSE represents the deviation between the denoised signal and the original signal, and the smaller the deviation is, the most effective components are reserved.
As shown in fig. 4, the experimental procedure of this embodiment is as follows:
1. the main test is that the electroencephalogram cap is tried on, and a proper single-limb motor imagery normal form is selected according to the limb motor function condition after the stroke is tested;
2. after training is started, the electroencephalogram cap acquires real-time EEG data in the experimental process of a tested object, the data communication module transmits the data from hardware to software, and the format conversion module converts the data into a format which can be processed by the software;
3. transmitting the data after format conversion to EEG data analysis, performing noise screening on the data by a deep neural network after down-sampling, filtering and ICA decomposition, and finally performing data elimination on the screened noise;
4. and the data after noise elimination is output and stored in real time by a noise-free EEG output module.
In light of the foregoing description of the preferred embodiment of the present invention, many modifications and variations will be apparent to those skilled in the art without departing from the spirit and scope of the invention. The technical scope of the present invention is not limited to the content of the specification, and must be determined according to the scope of the claims.

Claims (5)

1. A method for noise identification of EEG data using a deep neural network, comprising the steps of:
s1, collecting EEG data induced by the combined action of four joints of a shoulder, an elbow, a wrist and a finger in a tested stroke state through an EEG cap, and dividing the action into single joint training and joint comprehensive training;
s2, carrying out down-sampling, coarse filtering and ICA independent component decomposition on the EEG data to obtain IC activation and IC topographic map of a plurality of independent components;
s3, cutting the IC topographic map, dividing the IC activation into a plurality of trails according to a certain time, and setting the overlapping time of adjacent windows;
s4, constructing a deep neural network consisting of one-dimensional CNN convolution, two-dimensional CNN convolution and DDN convolution, sending the IC topographic map into the two-dimensional CNN convolution network, activating the IC and sending the IC topographic map into the one-dimensional convolution network, sending output values of the two convolution networks into the DNN convolution network, and setting a hyper-parameter;
s5, classifying the output value of the DNN convolutional network through a softmax classifier, screening and eliminating noise of EEG data according to the classification result, calculating the signal-to-noise ratio and the root-mean-square difference of the EEG data with the noise eliminated, and evaluating the denoising effect.
2. The method for noise recognition of EEG data using deep neural network as claimed in claim 1, wherein said training of single joint is to stimulate brain motor function area to generate EEG data by seven actions of shoulder flexion 30 °, shoulder extension 30 °, elbow flexion 90 °, elbow flexion maximum, wrist left-hand, wrist right-hand and finger fist grasping;
the multi-joint comprehensive training stimulates a brain movement function area to generate EEG data by eight actions of stimulating the stimulation source of forearm pronation, forearm supination, upper limb forward flexion of 90 degrees, upper limb backward extension upward lifting and touching waist, upper limb abduction of 90 degrees, upper limb forward flexion of 180 degrees, shoulder touching experiment and finger nose experiment.
3. The method of noise recognition on EEG data using a deep neural network according to claim 1, wherein: the one-dimensional CNN convolution comprises: a first convolution layer, wherein the width of an inner core is 20, the number of filters is 5, and the step length is 2; a second convolution layer, the width of the inner core is 6, the number of the filters is 16, and the step length is 2; a third convolution layer, the width of the kernel is 15, the number of the filters is 34, and the step length is 1; a fourth convolution layer, the width of the kernel is 11, the number of the filters is 21, and the step length is 1; a fifth convolution layer, the width of the kernel is 6, the number of the filters is 45, and the step length is 1; a sixth convolution layer, the kernel width is 7, the number of filters is 30, and the step length is 1; the first, second, third, fifth and sixth convolution layers are connected with the pooling layer with the kernel width of 4 and the step length of 2; the activation functions are all PreLU functions; and finally, accessing a Flatten layer for dimension reduction to obtain the IC activated characteristic column vector.
4. The method of noise identification of EEG data using a deep neural network of claim 1, wherein said two-dimensional CNN convolution comprises: the first convolution layer of the IC map, the size of an inner core is 20 x 20, the number of filters is 40, and the step length is 2 x 2; the second convolution layer of the IC map has the kernel size of 5 × 5, the number of filters is 50, and the step length is 2 × 2; the size of an inner core of the third convolution layer of the IC map is 6 x 6, the number of filters is 14, and the step length is 1 x 1; the fourth convolution layer of the IC map has the kernel size of 9 × 9, the number of filters of 35 and the step size of 2 × 2; the fifth convolution layer of the IC map has the kernel size of 12 × 12, 21 filters and the step length of 1 × 1; the sixth convolution layer of the IC map has the size of 16 × 16 inner cores, the number of filters is 45, and the step length is 2 × 2; the seventh convolution layer of the IC map, the size of the inner core is 8 × 8, the number of the filters is 30, and the step length is 2 × 2; connecting the second, third, fifth and seventh convolution layers of the IC map with the pooling layers with the size of 2 x 2 and the step length of 1 x 1; the activation functions are all PreLU functions; and finally, accessing a Flatten layer for dimension reduction to obtain a characteristic column vector of the IC topographic map.
5. The method of noise identification of EEG data with a deep neural network according to claim 1, characterized in that: the DDN convolution includes a first fully-connected layer, a second fully-connected layer, and a third fully-connected, the first fully-connected output dimension being 135, the second fully-connected output dimension being 210, and the third fully-connected output dimension being 28.
CN202210307716.0A 2022-03-27 2022-03-27 Noise identification method for EEG data by using deep neural network Pending CN114861702A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210307716.0A CN114861702A (en) 2022-03-27 2022-03-27 Noise identification method for EEG data by using deep neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210307716.0A CN114861702A (en) 2022-03-27 2022-03-27 Noise identification method for EEG data by using deep neural network

Publications (1)

Publication Number Publication Date
CN114861702A true CN114861702A (en) 2022-08-05

Family

ID=82630070

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210307716.0A Pending CN114861702A (en) 2022-03-27 2022-03-27 Noise identification method for EEG data by using deep neural network

Country Status (1)

Country Link
CN (1) CN114861702A (en)

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110015538A1 (en) * 2009-07-17 2011-01-20 Matthews Jr Thomas Virgil System and method for analyzing electroencephalography data
CN109784242A (en) * 2018-12-31 2019-05-21 陕西师范大学 EEG Noise Cancellation based on one-dimensional residual error convolutional neural networks
CN110164472A (en) * 2019-04-19 2019-08-23 天津大学 Noise classification method based on convolutional neural networks
CN113208622A (en) * 2021-04-07 2021-08-06 北京脑陆科技有限公司 Electroencephalogram EEG signal denoising method and system based on deep neural network technology
CN113723171A (en) * 2021-06-16 2021-11-30 陕西师范大学 Electroencephalogram signal denoising method based on residual error generation countermeasure network
CN113842115A (en) * 2021-09-27 2021-12-28 长春工程学院 Improved EEG signal feature extraction method

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110015538A1 (en) * 2009-07-17 2011-01-20 Matthews Jr Thomas Virgil System and method for analyzing electroencephalography data
CN109784242A (en) * 2018-12-31 2019-05-21 陕西师范大学 EEG Noise Cancellation based on one-dimensional residual error convolutional neural networks
CN110164472A (en) * 2019-04-19 2019-08-23 天津大学 Noise classification method based on convolutional neural networks
CN113208622A (en) * 2021-04-07 2021-08-06 北京脑陆科技有限公司 Electroencephalogram EEG signal denoising method and system based on deep neural network technology
CN113723171A (en) * 2021-06-16 2021-11-30 陕西师范大学 Electroencephalogram signal denoising method based on residual error generation countermeasure network
CN113842115A (en) * 2021-09-27 2021-12-28 长春工程学院 Improved EEG signal feature extraction method

Similar Documents

Publication Publication Date Title
CN110765920B (en) Motor imagery classification method based on convolutional neural network
Ibrahimy et al. Design and optimization of Levenberg-Marquardt based neural network classifier for EMG signals to identify hand motions
CN110555468A (en) Electroencephalogram signal identification method and system combining recursion graph and CNN
CN111860410A (en) Myoelectric gesture recognition method based on multi-feature fusion CNN
Zhao et al. Deep CNN model based on serial-parallel structure optimization for four-class motor imagery EEG classification
CN113180692B (en) Electroencephalogram signal classification and identification method based on feature fusion and attention mechanism
CN113558639A (en) Motor intention brain muscle network analysis method based on glange causal relation and graph theory
Kisa et al. EMG based hand gesture classification using empirical mode decomposition time-series and deep learning
Abibullaev et al. A brute-force CNN model selection for accurate classification of sensorimotor rhythms in BCIs
Ghonchi et al. Spatio-temporal deep learning for EEG-fNIRS brain computer interface
Mirzabagherian et al. Temporal-spatial convolutional residual network for decoding attempted movement related EEG signals of subjects with spinal cord injury
CN117908665A (en) Control method of disabled person rehabilitation equipment based on brain-computer interface
CN114082169A (en) Disabled hand soft body rehabilitation robot motor imagery identification method based on electroencephalogram signals
CN118094317A (en) Motor imagery electroencephalogram signal classification system based on TimesNet and convolutional neural network
CN110321856B (en) Time-frequency multi-scale divergence CSP brain-computer interface method and device
CN115985463B (en) Real-time muscle fatigue prediction method and system based on wearable equipment
CN114861702A (en) Noise identification method for EEG data by using deep neural network
CN115154828A (en) Brain function remodeling method, system and equipment based on brain-computer interface technology
CN115024735A (en) Cerebral apoplexy patient rehabilitation method and system based on movement intention recognition model
Hamzah et al. EEG signal classification to detect left and right command using artificial neural network (ANN)
Tang et al. Enhancing EEG and sEMG Fusion Decoding Using a Multi-Scale Parallel Convolutional Network With Attention Mechanism
Azam et al. Classification Of Complicated Upper Limb Movements From Pre-movement EEG Signals Using STFT And Spectral Characteristics
CN117556307A (en) Motor imagery gesture recognition method based on electroencephalogram myoelectricity multi-mode fusion
Ghasemi et al. A new method for features normalization in motor imagery few-shot learning using resting-state
CN116010848A (en) Electroencephalogram data amplification method and system based on generation of countermeasure network model

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