CN110432899A - Electroencephalogram signal identification method based on depth stacking support matrix machine - Google Patents
Electroencephalogram signal identification method based on depth stacking support matrix machine Download PDFInfo
- Publication number
- CN110432899A CN110432899A CN201910664430.6A CN201910664430A CN110432899A CN 110432899 A CN110432899 A CN 110432899A CN 201910664430 A CN201910664430 A CN 201910664430A CN 110432899 A CN110432899 A CN 110432899A
- Authority
- CN
- China
- Prior art keywords
- eeg signal
- layer
- feature
- eeg
- support matrix
- 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.)
- Granted
Links
- 239000011159 matrix material Substances 0.000 title claims abstract description 44
- 238000000034 method Methods 0.000 title claims abstract description 20
- 238000012549 training Methods 0.000 claims abstract description 11
- 238000013145 classification model Methods 0.000 claims abstract description 6
- 238000005070 sampling Methods 0.000 claims description 6
- 238000000605 extraction Methods 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 abstract 1
- 238000010586 diagram Methods 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 2
- 230000003416 augmentation Effects 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000009472 formulation Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
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
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B2576/00—Medical imaging apparatus involving image processing or analysis
- A61B2576/02—Medical imaging apparatus involving image processing or analysis specially adapted for a particular organ or body part
- A61B2576/026—Medical imaging apparatus involving image processing or analysis specially adapted for a particular organ or body part for the brain
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Signal Processing (AREA)
- Physics & Mathematics (AREA)
- Pathology (AREA)
- Public Health (AREA)
- Biomedical Technology (AREA)
- Heart & Thoracic Surgery (AREA)
- Medical Informatics (AREA)
- Molecular Biology (AREA)
- Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- General Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Veterinary Medicine (AREA)
- Psychiatry (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physiology (AREA)
- Psychology (AREA)
- Power Engineering (AREA)
- Evolutionary Computation (AREA)
- Fuzzy Systems (AREA)
- Mathematical Physics (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The invention provides an electroencephalogram signal identification method based on a depth stack support matrix machine, which comprises the following steps: firstly, preprocessing an Electroencephalogram (EEG) signal and extracting characteristics; training a first layer Support Matrix Machine (SMM) by using the extracted original EEG signal features as input to obtain predicted output of the first layer; projecting the first layer of prediction output to an original EEG feature space by using matrix random projection, superposing the first layer of prediction output with the original EEG signal feature to obtain a second layer of EEG signal feature, and training a second layer of SMM by using the second layer of EEG signal feature as input to obtain the prediction output of the second layer; deeper EEG signal features are obtained in this manner and SMM is trained until accuracy converges to obtain the final classification model. The invention can accurately judge different types of EEG signals and ensure the safe and reliable operation of the BCI system based on the EEG.
Description
Technical field
The present invention relates to a kind of EEG signal identification methods that support matrix machine is stacked based on depth, belong to EEG signals knowledge
Other field.
Background technique
Electroencephalogram (Electroencephalogram, EEG) imaging method has temporal resolution high, and acquisition is simple, note
The features such as recording apparatus is noninvasive is widely used in recording the dynamic moving process of brain.By machine learning method EEG data with
Limb motion establishes reasonable mapping relations between being intended to, and realizes that the motion intention of patient in rehabilitation accurately detects, is to work as
One of the major issue of preceding research work urgent need to resolve.Current most of traditional machine learning algorithms require that input feature vector must
It must be vector form, and EEG signal feature is the form of matrix.It is defeated that matrix is directly changed into vector by conventional machines learning algorithm
Enter classifier, destroy the structural information in EEG signal feature, can classification results be generated with certain influence.In response to this problem,
Researcher proposes the number of support matrix machine (Support Matrix Machines, SMM) for direct processing array form
According to utilizing the structural information in matrix form feature by being introduced into kernel normal form, improve the recognition accuracy of EEG signal.But
It still falls within shallow-layer learning method, since the structure of single hidden layer causes its expression ability limited, in the classification problem that processing is complicated
When generalization ability it is poor.Therefore it proposes that a kind of depth stacks support matrix machine, in conjunction with support matrix machine and stacks generalization theory, protect
The structural information in EEG signal feature is stayed, while enhancing model to indicate ability, realizes accurately identifying for EEG signal.
Summary of the invention
The present invention provides a kind of EEG signal identification method that support matrix machine is stacked based on depth, using SMM as base
This module retains the structural information in EEG signal feature, in combination with generalization theory is stacked, introduces the conduct of matrix accidental projection
Core stack element helps to open raw EEG data manifold, has stronger expression energy using the prediction output of front layer module
Power, to realize accurately identifying for EEG signal.
The present invention uses following technical scheme to solve above-mentioned technical problem:
A kind of EEG signal identification method being stacked support matrix machine based on depth, is included the following steps:
Step 1: acquisition EEG signal;
Step 2: EEG signal being pre-processed, to eliminate noise and artefact, then feature is carried out using CSP algorithm and mentions
It takes, thus obtains raw EEG signal featureWhereinIndicate i-th of EEG signal feature, d1, d2Respectively
Represent recording electrode port number and time sampling points, yi∈ { 1, -1 } is corresponding class label;
Step 3: it is defeated to obtain first layer prediction using raw EEG signal feature as input training first layer SMM disaggregated model
Out.
Step 4: using matrix accidental projection by first layer prediction output project to original EEG feature space and with it is original
EEG signal feature is superimposed to obtain second layer EEG signal feature, obtains the pre- of the second layer as training second layer SMM is inputted
Survey output.
Step 5: generating every layer of EEG signal feature in this manner, every layer of SMM of training simultaneously obtains every layer of prediction output.
Step 6: repeating step 5 until precision convergence, obtain final classification model.
Preferably, the depth stacks support matrix machine using SMM as basic module, in combination with the extensive reason of stacking
By introducing matrix accidental projection helps to open raw EEG data as core stack element using the prediction output of front layer module
Manifold has stronger expression ability.
Preferably, the depth stacks support matrix machine using SMM as basic stackable unit, objective function are as follows:
Wherein W and b is Optimal Separating Hyperplane parameter, and tr () represents the mark of matrix, | | | |*For nuclear norm, C, τ are constraint
Lose the penalty coefficient of item and nuclear norm.
Preferably, the depth, which stacks support matrix power traction and enters matrix accidental projection as core stack element, constructs every layer
EEG signal feature, constructive formula are as follows:Wherein pl, mAnd ql,mIt is used for l layers
O is exported in projection m layers of front predictionmProjection matrix, the element in matrix obtains from standardized normal distribution N (0,1) sampling.
The utility model has the advantages that
1, using SMM as basic stackable unit, the EEG feature that can directly input matrix form is trained prediction,
Retain the structural information in EEG signal, improves the recognition accuracy of EEG signal.Currently based on the tradition of vector input form
Machine learning algorithm is easily lost the structural information in EEG signal;
2, this method introduces matrix accidental projection as core stack element and constructs every layer of EEG signal feature, utilizes front layer
The prediction output of module helps to open raw EEG data manifold, has a stronger expression ability, and traditional shallow-layer learning method by
Cause its expression ability limited in the structure of single hidden layer, generalization ability is poor in processing complicated classification problem.
Detailed description of the invention
Fig. 1 is the work flow diagram for stacking the EEG signal identification method of support matrix machine in the present invention based on depth;
Fig. 2 is matrix accidental projection technology schematic diagram in the present invention.
Specific embodiment
Further explanation is done to the present invention below with reference to example.
The main implementing procedure of the present invention is as follows, and dependence diagram is shown in Fig. 1.
Step 1: acquisition EEG signal;
Step 2: EEG signal being pre-processed, including bandpass filtering and artifact removal, to eliminate noise and artefact.So
Feature extraction is carried out using CSP algorithm afterwards, thus obtains raw EEG signal featureWhereinIt indicates
I-th of EEG signal feature, d1, d2Respectively represent recording electrode port number and time sampling points, yi∈ { 1, -1 } is corresponding
Class label;
Step 3: obtaining first layer disaggregated model f using raw EEG signal feature as training first layer SMM is inputted1=sgn
(tr(WTX)+b), wherein W and b is Optimal Separating Hyperplane parameter.The method for solving of W and b is as follows:
1, following objective function is constructed:
Wherein tr (WTW the mark of matrix) is represented, Z is that waiting for W is worth, | | Z | |*For nuclear norm, C, τ be constraint loss item with
And the penalty coefficient of nuclear norm.
The corresponding augmentation Lagrangian formulation of above formula are as follows:
WhereinQ (Z)=τ | | Z | |*, β is penalty factor,For Lagrange multiplier, | | | |FFor Frobenius norm.
2, it updates to solve using alternating direction multipliers method alternating iteration and obtains hyperplane parameter W and b, the specific method is as follows:
1) Z is initialized0=0, M0=0, β > 0, iterative characteristic value c1=0, m1=1, controlling elements μ ∈ (0,1).
2) pass through following formula undated parameter
Mt+1=Mt-β(Wt+1-Zt+1) (5c)
Wherein the solution of formula (5a) isThe solution of formula (5b) isAndWhereinIt is accorded with for singular value threshold operation, i.e.,Wherein{u}+=max (0, u).
3) the t times iterative characteristic value is calculatedIf ct< μ ct-1, thenOtherwise mt+1=1, Zt+1=
Zt-1, Mt+1=Mt-1, ct=μ-1ct-1。
4) 2)~3 t=t+1 is repeated) until convergence obtains hyperplane parameter W and b.
Step 4: being obtained predicting output accordingly by first layer disaggregated model, then utilize matrix accidental projection by first layer
Prediction output projects to original EEG feature space and is superimposed to obtain second layer EEG signal feature with raw EEG signal feature, that is,The prediction output of the second layer is obtained as training second layer SMM is inputted.
Step 5: generating every layer of EEG signal feature in this manner.Specifically, l layers of EEG signal latent structure formula
Are as follows:Wherein pl,mAnd qL, mIt is defeated for projecting the prediction of m layers of front for l layers
O outmProjection matrix, element in matrix is from N (0,1) sampling obtains.Training SMM is obtained in newly-generated EEG signal feature
L layers of Optimal Separating Hyperplane model fl=sgn (tr (Wl TX)+bl)。
Step 6: repeating step 5 until precision convergence, obtain final classification model.
The above are technical em- bodiments of the invention and technical characterstic, it is only used to illustrate the technical scheme of the present invention rather than limits
System.However those skilled in the art are still potentially based on teachings of the present invention and disclosure and make to technical solution of the present invention
Modification and equivalent replacement.Therefore, protection scope of the present invention should be not limited to the revealed content of embodiment, and should include various
Cover without departing substantially from substitution and amendment of the invention, and by the claims book.
Claims (4)
1. a kind of EEG signal identification method for stacking support matrix machine based on depth, which is characterized in that steps are as follows:
Step 1: acquisition EEG signal;
Step 2: EEG signal being pre-processed, to eliminate noise and artefact, feature extraction then is carried out to EEG signal, thus
Obtain n raw EEG signal featureWhereinIndicate i-th of EEG signal feature, d1,d2It respectively represents
Recording electrode port number and time sampling points, yi∈ { 1, -1 } is corresponding class label;
Step 3: using raw EEG signal feature as inputting training first layer SMM model and obtaining the prediction output of first layer, counting
Nicety of grading is calculated, final classification model is obtained if precision convergence, otherwise continues step 4;
Step 4: first layer prediction output being projected into original EEG feature space using matrix accidental projection and is believed with original EEG
Number feature is superimposed to obtain second layer EEG signal feature, as input training second layer SMM obtain the second layer prediction it is defeated
Out;
Step 5: repeating step 4 and generate subsequent layers EEG signal feature, each layer SMM of training and the prediction output for obtaining each layer, directly
It is restrained to precision, obtains final classification model;
Step 6: EEG signal to be sorted being inputted into final classification model, obtains its class label.
2. a kind of EEG signal identification method for stacking support matrix machine based on depth according to claim 1, feature
It is, in step 2, feature extraction is carried out to EEG signal using cospace mode.
3. a kind of EEG signal identification method for stacking support matrix machine based on depth according to claim 1, feature
It is, the depth stacks support matrix machine using SMM as basic stackable unit, objective function are as follows:
Wherein W and b is Optimal Separating Hyperplane parameter, and tr () represents the mark of matrix, | | | |*For nuclear norm, C, τ are constraint loss
The penalty coefficient of item and nuclear norm, ξiFor slack variable.
4. a kind of EEG signal identification method for stacking support matrix machine based on depth according to claim 1, feature
It is, the depth stacking support matrix power traction enters matrix accidental projection as core stack element and constructs every layer of EEG signal spy
Sign, constructive formula are as follows:Wherein pl,mAnd ql,mFor l layers for projecting front
M layers of prediction export omProjection matrix, the element in matrix obtains from standardized normal distribution N (0,1) sampling.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910664430.6A CN110432899B (en) | 2019-07-23 | 2019-07-23 | Electroencephalogram signal identification method based on depth stacking support matrix machine |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910664430.6A CN110432899B (en) | 2019-07-23 | 2019-07-23 | Electroencephalogram signal identification method based on depth stacking support matrix machine |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110432899A true CN110432899A (en) | 2019-11-12 |
CN110432899B CN110432899B (en) | 2020-07-03 |
Family
ID=68431108
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910664430.6A Active CN110432899B (en) | 2019-07-23 | 2019-07-23 | Electroencephalogram signal identification method based on depth stacking support matrix machine |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110432899B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114021608A (en) * | 2021-11-17 | 2022-02-08 | 南京工业大学 | Electroencephalogram recognition method fusing structure information between brain regions |
CN117338313A (en) * | 2023-09-15 | 2024-01-05 | 武汉纺织大学 | Multi-dimensional characteristic electroencephalogram signal identification method based on stacking integration technology |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101219048A (en) * | 2008-01-25 | 2008-07-16 | 北京工业大学 | Method for extracting brain electrical character of imagine movement of single side podosoma |
CN102542283A (en) * | 2010-12-31 | 2012-07-04 | 北京工业大学 | Optimal electrode assembly automatic selecting method of brain-machine interface |
CN102789316A (en) * | 2012-07-12 | 2012-11-21 | 上海海事大学 | Control method for movement of two-dimensional cursor of brain machine interface based on motor imageries |
-
2019
- 2019-07-23 CN CN201910664430.6A patent/CN110432899B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101219048A (en) * | 2008-01-25 | 2008-07-16 | 北京工业大学 | Method for extracting brain electrical character of imagine movement of single side podosoma |
CN102542283A (en) * | 2010-12-31 | 2012-07-04 | 北京工业大学 | Optimal electrode assembly automatic selecting method of brain-machine interface |
CN102789316A (en) * | 2012-07-12 | 2012-11-21 | 上海海事大学 | Control method for movement of two-dimensional cursor of brain machine interface based on motor imageries |
Non-Patent Citations (3)
Title |
---|
IMRAN RAZZAK ET AL: "Multiclass Support Matrix Machines by Maximizing the Inter-Class Margin for Single Trial EEG Classification", 《IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING》 * |
QINGQING ZHENG ET AL: "Multiclass support matrix machine for single trial EEG classification", 《NEUROCOMPUTING》 * |
QINGQING ZHENG ET AL: "Sparse Support Matrix Machines", 《PATTERN RECOGNITION》 * |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114021608A (en) * | 2021-11-17 | 2022-02-08 | 南京工业大学 | Electroencephalogram recognition method fusing structure information between brain regions |
CN117338313A (en) * | 2023-09-15 | 2024-01-05 | 武汉纺织大学 | Multi-dimensional characteristic electroencephalogram signal identification method based on stacking integration technology |
CN117338313B (en) * | 2023-09-15 | 2024-05-07 | 武汉纺织大学 | Multi-dimensional characteristic electroencephalogram signal identification method based on stacking integration technology |
Also Published As
Publication number | Publication date |
---|---|
CN110432899B (en) | 2020-07-03 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Wang et al. | Large-scale isolated gesture recognition using convolutional neural networks | |
CN106934042B (en) | Knowledge graph representation system and implementation method thereof | |
Hou et al. | Fruit recognition based on convolution neural network | |
CN105447473B (en) | A kind of any attitude facial expression recognizing method based on PCANet-CNN | |
Wan et al. | Regularization of neural networks using dropconnect | |
Osherov et al. | Increasing cnn robustness to occlusions by reducing filter support | |
CN105138998B (en) | Pedestrian based on the adaptive sub-space learning algorithm in visual angle recognition methods and system again | |
CN106203356B (en) | A kind of face identification method based on convolutional network feature extraction | |
Zhang et al. | Picking neural activations for fine-grained recognition | |
CN114332568A (en) | Training method, system, equipment and storage medium of domain adaptive image classification network | |
CN106127197A (en) | A kind of saliency object detection method based on notable tag sorting | |
CN109344856B (en) | Offline signature identification method based on multilayer discriminant feature learning | |
CN110432899A (en) | Electroencephalogram signal identification method based on depth stacking support matrix machine | |
CN117079098A (en) | Space small target detection method based on position coding | |
Yang et al. | Deep learning-based framework for Autism functional MRI image classification | |
Wang et al. | Differential geometry boosts convolutional neural networks for object detection | |
CN107832713A (en) | A kind of human posture recognition method based on OptiTrack | |
CN107423697A (en) | Activity recognition method based on non-linear fusion depth 3D convolution description | |
CN117036862B (en) | Image generation method based on Gaussian mixture variation self-encoder | |
CN109886160A (en) | It is a kind of it is non-limiting under the conditions of face identification method | |
Li et al. | Hybrid rgb-d object recognition using convolutional neural network and fisher vector | |
Daramola et al. | Efficient face recognition system using artificial neural network | |
Liu et al. | Adaptive Texture and Spectrum Clue Mining for Generalizable Face Forgery Detection | |
Zhang et al. | Lightweight PM-YOLO network model for moving object recognition on the distribution network side | |
Han et al. | Video action recognition based on deeper convolution networks with pair-wise frame motion concatenation |
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 |