CN106022256B - A kind of parameter optimization method of brain machine interface system decision model - Google Patents
A kind of parameter optimization method of brain machine interface system decision model Download PDFInfo
- Publication number
- CN106022256B CN106022256B CN201610329708.0A CN201610329708A CN106022256B CN 106022256 B CN106022256 B CN 106022256B CN 201610329708 A CN201610329708 A CN 201610329708A CN 106022256 B CN106022256 B CN 106022256B
- Authority
- CN
- China
- Prior art keywords
- classification
- data
- time
- sample
- window
- 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
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F3/00—Input arrangements for transferring data to be processed into a form capable of being handled by the computer; Output arrangements for transferring data from processing unit to output unit, e.g. interface arrangements
- G06F3/01—Input arrangements or combined input and output arrangements for interaction between user and computer
- G06F3/011—Arrangements for interaction with the human body, e.g. for user immersion in virtual reality
- G06F3/015—Input arrangements based on nervous system activity detection, e.g. brain waves [EEG] detection, electromyograms [EMG] detection, electrodermal response detection
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Human Computer Interaction (AREA)
- Neurosurgery (AREA)
- Neurology (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Dermatology (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Signal Processing (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The present invention relates to brain-computer interface technical fields, a kind of parameter optimization method of brain machine interface system decision model, the following steps are included: (1) acquires EEG signals training data and is pre-processed, (2) linear space integration single probe method data classification and evidence accumulation initiating process positioning.The present invention uses linear space integration single detection method, Classification and Identification is carried out to two kinds of experimental conditions in training data, by the accuracy rate of identification with the trend of the variation of time during single experiment, temporal positioning is carried out to the evidence accumulation in decision model, to carry out parameter optimization to sequential decision model, compared with previous sequential decision model, the invention is by positioning the evidence accumulation in decision model, eliminate invalid classifications information accumulation process, only effective classification information in evidence accumulation is accumulated, improve the real-time of the online brain machine interface system based on the decision model.
Description
Technical field
The present invention relates to a kind of parameter optimization methods of brain machine interface system decision model, belong to brain-computer interface technology neck
Domain.
Background technique
Brain machine interface system is a kind of communication system without relying on muscle control, it is intended to for disabled patient or movement damage
Hurt patient and a kind of new communications conduit is provided.The core of brain-computer interface technology just passes through the classification processing to EEG signals, identification
Different conscious activity states.The EEG signals recorded from scalp are very faint, and signal-to-noise ratio is very low, therefore different pattern-recognitions
Method is applied to brain machine interface system to extract brain electrical feature information, and training classifier to reduce classification error to the maximum extent
Rate.For this purpose, researcher conducts extensive research brain electricity classification method, such as support vector machines, artificial neural network etc..This
A little methods achieve preferable classifying quality in its applicable field, but only divide the brain electrical feature of special time period
Class does not account for the temporal information in assorting process.EEG signals have non-stationary, the differentiation of each period brain electrical feature
It spends and different, therefore these classifiers can not weigh the relationship between classification accuracy and decision-making time well.For reality
When the training set Limited Number that can acquire of brain machine interface system, the classifier of better performances how is gone out with a small amount of sample training, i.e.,
" small sample problem " is the challenge that brain machine interface system faces.Therefore, Recent study personnel gradually go to attention
It can continuously reflect the dynamic cataloging method that user is intended to.
" classification " in brain-computer interface field generally means that decision or selecting response.Decision is one kind of biological nervous system
Higher cognitive process, being unique in that for Information procession can stimulate transient state rapid make to adjudicate to influence behavior.By
Often there is uncertainty in the information that decision process is felt, therefore statistical theory is the strong of formalized description uncertain information
Broad theory tool.Sequential analysis theory provides a flexible mathematical model to decision process.Sequential analysis technique study is determined
When question and answer on politics is inscribed, not preparatory fixed sample amount is gradually sampled, needed for average sample size it is minimum, advantageously account for
" small sample problem " of brain machine interface system, therefore gradually attracted the attention of brain machine interface system researcher.
In sequential decision model, decision process is exactly the letter that since noisy information reach accumulation boundary accumulating starting point
Accumulation is ceased, when information accumulation reaches some boundary, then it is assumed that decision is completed.Therefore, sequential likelihood ratio test classification method
The classification information that every section of feature is calculated using probability ratio test method is accumulated classification information since data starting point, gradually taken
Sample judges when accumulating decision variable and reaching a certain threshold value, stops accumulation.This method is able to reflect decision process to branch
The tradeoff between the adaptation reaction and speed of decision and accuracy that selection evidence is made is held, therefore is suitable for online brain machine and connects
Port system.The tired of brain electrical feature information is just carried out when however, nowadays sequential likelihood ratio test classification method being all since experiment
Product, a length of whole segment data length of maximum accumulation time window.Domestic and international consciousness decision domain studies have shown that the process packet of decision
Include stimulation perception, it is noted that the processes such as evidence accumulation and Motor execution.It is perceived in stimulation, the stages progress decision information such as pays attention to
Accumulation be to classification results it is unhelpful, increase the decision-making time of brain machine interface system instead, reduce the real-time of system.
Therefore, evidence accumulation is positioned, specifies the starting point of effective information accumulation, can reduce based on sequential decision model
Dynamic classifier calculation amount, the maximum accumulation time window for shortening algorithm is long, to improve the real-time performance of brain machine interface system.
Summary of the invention
In order to overcome problems of the prior art, it is an object of the present invention to provide a kind of brain machine interface system decision models
Parameter optimization method.The present invention originated the evidence accumulation during decision using linear space integration single probe method
Cheng Jinhang positioning, to advanced optimize brain machine interface system decision model parameter.This method was by originating evidence accumulation
Cheng Jinhang positioning excludes the time loss that dynamic classifier carries out the accumulation of brain electrical feature in unrelated procedures, not only reduces sequential
The accumulated time of information, and the efficiency of decision-making is improved, to further increase the real-time of brain machine interface system.
In order to achieve the above-mentioned object of the invention, it solves the problems of in the prior art, the technical solution that the present invention takes
It is: a kind of parameter optimization method of brain machine interface system decision model, comprising the following steps:
Step 1, acquisition EEG signals training data are simultaneously pre-processed, and are moved using brain wave acquisition equipment acquisition random point
EEG signals under direction discernment task are as training data, random point fortune different according to the direction of motion and experiment difficulty
Dynamic direction includes a low difficulty left side, the low difficulty right side, middle difficulty is left, middle difficulty is right, a highly difficult left side and highly difficult right six kinds of differences identify
Task, and collected EEG signals are filtered, remove artefact pretreatment；According to random spot moving direction label to acquisition
EEG signals carry out segment processing, and to distinguish different experiment examinations time, split time difficulty same for same subject
The different experiments examination time of degree needs to be consistent, and the split time length that is taken needs to occur comprising stimulation to making a response
The process of process, i.e. subject progress decision；
Step 2, linear space integration single probe method data classification and evidence accumulation initiating process positioning, by linear
Space integration single probe method identifies two kinds of data categories in experiment, such as the identification to the direction of motion, Zuo Huo
The right side, the identification to experiment difficulty is difficult or easy, specifically includes following sub-step:
(a) structural classification initial data X, classification initial data X should include two class data, such as the left and right direction of motion two
Class data, i.e. X1And X2, wherein X1It is a nchannels×Tsample×N1Data matrix, X2It is a nchannels×Tsample
×N2Data matrix, wherein N1With N2Respectively represent the experiment examination number that two class data include, TsampleFor what is used in step 1
The sampling number of split time, nchannelsFor the lead number of use, and the initial data X=[X that classifies1,X2], it is a nchannels
×Tsample×(N1+N2) matrix, and obtained classification initial data X-form is converted into (Tsample×Ntrials)×
nchannelsMatrix form in next Classification and Identification, NtrialsWhat the data for progress Classification and Identification were included owns
Experiment examination number, Ntrials=N1+N2；
(b) sliding window is constructed, the window width of sliding window is δ sampled point, and single sliding distance is τ sampled point, for adopting
Number of samples is TsampleData can be divided into altogetherA sliding window, whereinIndicate the lower meaning being rounded
Think, the position of each sliding window is [1+ (i-1) τ, δ+(i-1) τ] respectively, i=1,2 ..., K, using data in window to point
Class device is trained, the time starting point accumulated by the variation tendency of classification accuracy come positive evidence；
(c) category sequence L is constructed, in original brain electricity classification data and sub-step (b) according to obtained in sub-step (a)
Sliding window, construction is (δ × a N comprising category the sequence L, L of element { 0,1 }trialsThe column vector of) × 1, wherein δ is to slide
The window of dynamic window is long, NtrialsNumber is tried to carry out all experiments that the data of Classification and Identification are included；
(d) logistic regression linear classifier is used, the classification initial data in each sliding window constructed in step (b) is estimated
Count an optimal spatial weighted vector wτ,δ, which can carry out maximization identification to two class data in the window, pass through formula
(1) classification results y is calculated,
Y=wTX+b (1)
In formula, w is the space weighted vector of the linear classifier, wTIndicate that the transposed matrix of w, b are bias term, X is point
Class initial data, it is assumed that the sample of classification initial data, which sorts out probability, meets formula (2),
In formula, p (c=+1 | X) indicates that X is judged as the probability of class c=+1, and p (c=-1 | X) indicate that X is judged as class
The probability of c=-1 obtains optimal space weight w by iteration weight weighted least-squares method, and specific iterative process passes through public affairs
Formula (3) realized,
In formula, X is classification initial data, XTFor the transposition of X, p is the classification probability vector of sample, and * represents inner product .* generation
Table vector product, d indicate the column vector comprising brain electricity sample category, i.e. L in sub-step (c), diag () are indicated a vector
It is converted into diagonal matrix, g represents gradient vector, and H is the gloomy battle array in sea that Fisher information matrix obtains, and Λ is punishing in iterative process
Penalty factor, during execution, the bias term b of script linear classification is integrated into the weight w of space, is retouched by formula (4)
It states,
Space weight w is initialized as 10-3*ones(nchannels+ 1,1), wherein ones (nchannels+ 1,1) one is indicated
(nchannels+ 1) all 1's matrix × 1, nchannelsThe lead number being acquired is represented, iteration process is until space weight
Until w restrains, so that the space weight w of optimal classification is obtained, since entire assorting process is to be directed to starting point as 1+ (i-
1) τ, i=1,2 ..., K, data in the sliding window of a length of δ of window, therefore optimal classification space weight w is considered as w at this timeτ,δ, whereinExpression sampling number is TsampleThe sliding window number that separates of data, what τ indicated to slide every time
Distance；
(e) by the space weight w for the optimal classification sought in sub-step (d)τ,δIt brings into formula (1), obtains classification results
Y, and constructed using the probability distribution of formula (2) calculating y, while using the probability distribution result of classification results y and sub-step (c)
The drafting of category sequence experience linearity curve, i.e. ROC curve, and calculate the area Az under ROC curve；
(f) the classification results Az that data in time windows obtain is counted, draws the Az song that window changes at any time
Line should be the trend that a kind of rising is presented, i.e. evidence to the classifying quality of two generic tasks due to the beginning accumulated with evidence
The more accumulation the more sufficient, and the accuracy to make accurate judgment is higher, and ought to reach maximum when decision executes；Therefore, to Az
The curve of window variation is analyzed at any time, observes its start time for continuing ascent stage, is started to conclusion evidence accumulation
Moment, and start the moment as the accumulation of the information characteristics of the online brain machine interface system next based on sequential decision model
Moment only accumulates effective classification information to exclude the unrelated classification information of accumulation, thus reduces the processing time, reaches
To the purpose for the real-time for improving brain machine interface system.
The medicine have the advantages that a kind of parameter optimization method of brain machine interface system decision model, comprising the following steps:
(1) it acquires EEG signals training data and is pre-processed, (2) linear space integration single probe method data classification and evidence
Accumulate initiating process positioning.Compared with the prior art, the present invention uses linear space integration single detection method, to training number
Two kinds of experimental conditions in carry out Classification and Identification, by the accuracy rate of analysis identification with the change of time during single experiment
The trend of change carries out temporal positioning to the evidence accumulation in decision model, to optimize the letter of sequential decision model
Breath accumulation parameter, compared with previous sequential decision model, the invention is by carrying out the evidence accumulation in decision model
Positioning, eliminates invalid classifications information accumulation process, and it is long to shorten maximum accumulation time window, improves based on the decision model
The real-time of online brain machine interface system.
Detailed description of the invention
Fig. 1 is the method for the present invention flow chart of steps.
Fig. 2 is random point side-to-side movement direction discernment experimental paradigm figure used by brain wave acquisition of the present invention.
Fig. 3 is the variation tendency that all subject motion's direction discernment accuracy rate mean values are slided with sliding window in the present invention
Figure.
Fig. 4 is the variation tendency that all subject's difficulty recognition accuracy mean values are slided with sliding window in the present invention
Figure.
Specific embodiment
The present invention will be further explained below with reference to the attached drawings.
As shown in Figure 1, a kind of parameter optimization method of brain machine interface system decision model, comprising the following steps:
Step 1,10 subject's EEG signals training datas are acquired and are pre-processed, brain wave acquisition equipment is utilized
The conductive polar cap of NeuroScan and 36 acquires the brain telecommunications under the random point side-to-side movement direction discernment task of three kinds of experiment difficulty
Number it is used as training data, brain electricity sample rate is 500Hz, and the bandpass filter of 0.1-70Hz is arranged.Wherein brain wave acquisition is tested
Normal form is tested after a sound " Beep " prompt and is started as shown in Fig. 2, need subject to watch screen attentively in collection process, prompting sound
After 1.5s, random point stimulation occur.Random point stimulation includes two parts, one is unanimously moved to the same direction
Random point；The second is to the random point of all directions random motion.When random point occurs, subject needs to be identified as early as possible to same
The direction of motion for the random point that one direction unanimously moves, and according to judgement, makes corresponding reaction, such as right-hand man's key is anti-
It answers, once tested make a response, then random point disappears, and after stimulation disappears, subject carries out rest in 2 seconds, into trying next time
It tests.Experiment sets three kinds of different experiment difficulty by changing the ratio of the random point unanimously moved.Therefore by the brain of acquisition
Electric data are divided into six classes: low difficulty is left, and low difficulty is right, and middle difficulty is left, and middle difficulty is right, a highly difficult left side, the highly difficult right side.It is directed to
Each difficulty is recycled by the above process, each difficulty acquires the data of 210 correct responses.After acquiring signal, to adopting
The EEG signals collected are pre-processed, and remove most of high frequencies and flip-flop including the use of the filter of 0.1-30Hz,
Then independent component analysis is used, the methods of threshold method removes the artefacts such as eye electricity.Then, according to random spot moving direction label pair
Pretreated EEG signals carry out segment processing, and to distinguish different experiment examinations time, split time is for same subject
The different experiments examination time of same difficulty needs to be consistent, and the split time length that is taken needs to occur comprising stimulation to making
The process of the process of reaction, i.e. subject progress decision；After being segmented to data, 2 are entered step.
Step 2, linear space integration single probe method data classification and evidence accumulation initiating process positioning, by linear
Space integration single probe method identifies two kinds of data categories in experiment, such as the identification to the direction of motion, Zuo Huo
The right side, the identification to experiment difficulty is difficult or easy, specifically includes following sub-step:
(a) structural classification initial data X, classification initial data X should include two class data, and two class data definitions are X1And X2,
Wherein, X1It is a nchannels×Tsample×N1Data matrix, X2It is a nchannels×Tsample×N2Data matrix,
Wherein N1With N2Respectively represent the experiment examination number that two class data include, TsampleSampling for the split time used in step 1
Points, nchannelsFor the lead number of use, and the initial data X=[X that classifies1,X2], it is a nchannels×Tsample×(N1+
N2) matrix, and obtained classification initial data X-form is converted into (Tsample×Ntrials)×nchannelsMatrix form use
In next Classification and Identification, NtrialsNumber, N are tried to carry out all experiments that the data of Classification and Identification are includedtrials=
N1+N2；
Two kinds of identification contents are come in the present invention, are direction of motion identification and experiment difficulty identification respectively, therefore classify
Initial data has two classes:
The direction of motion is identified: the eeg data being segmented is divided into two class X by the label based on random pointL、XR, respectively
Indicate left direction data and right direction data, i.e., X above1、X2.It is reconstructed to obtain new brain electricity classification data X=[X againL,
XR], it is a nchannels×Tsample×(NL+NR) size matrix.Wherein TsampleFor the split time that is used in step 1
Sampling number, nchannelsFor the lead number of use, NLAnd NRRespectively represent the experiment number of left direction and right direction.
Difficulty level is identified: three kinds of experiment difficulty are divided into two kinds of difficulty identification missions, i.e., low difficulty and middle difficulty
Identification, low difficulty and highly difficult identification.The low difficulty brain electricity number corresponding with middle difficulty that three kinds are tested in difficulty first
According to taking-up, combination obtains new eeg data X=[XLD,XMD], wherein XLD、XMDRespectively represent low difficulty data and middle difficulty number
According to X that is, above1、X2.This stylish eeg data X is a nchannels×Tsample×(NLD+NMD) size matrix.Wherein
TsampleSampling number for the split time used in step 1, nchannelsFor the lead number of use, NLD、NMDIt respectively represents low
The experiment number of difficulty and middle difficulty.
Secondly the low difficulty three kinds tested in difficulty is taken out with highly difficult corresponding eeg data, and combination obtains another
New eeg data X=[XLD,XHD], wherein XLD、XHDRespectively represent low difficulty data and highly difficult data, i.e., X above1、X2。
This stylish eeg data X is a nchannels×Tsample×(NLD+NHD) size matrix.Wherein TsampleTo be adopted in step 1
The sampling number of split time, nchannelsFor the lead number of use, NLD、NHDRespectively represent low difficulty and highly difficult reality
Test number.
(b) sliding window is constructed；The window width of sliding window is δ sampled point, and single sliding distance is τ sampled point, for adopting
Number of samples is TsampleData can be divided into altogetherA sliding window, whereinIndicate the lower meaning being rounded
Think.The position of each sliding window is [1+ (i-1) τ, δ+(i-1) τ] respectively, i=1,2 ..., K.Using data in window to point
Class device is trained, the time starting point accumulated by the variation tendency of classification accuracy come positive evidence.Sliding window of the present invention
Window width is set as δ=30 sampled point, single sliding distance τ=15 sampled point.
(c), category sequence L is constructed；Original brain electricity classification data and sub-step (b) according to obtained in sub-step (a)
In sliding window, construction is (δ × a N comprising category the sequence L, L of element { 0,1 }trialsThe column vector of) × 1.δ is at this time
The window of sliding window is long, NtrialsNumber is tried to carry out all experiments that the data of Classification and Identification are included.
(d) logistic regression estimates optimal spatial weighted vector；Each sliding using logistic regression, to being constructed in step (b)
Classification initial data in window estimates an optimal spatial weighted vector wτ,δ, which can carry out two class data in the window
Identification is maximized, wherein logistic regression is a kind of linear classifier, classification results y is calculated by formula (1),
Y=wTX+b (1)
In formula, w is the space weighted vector of the linear classifier, wTIndicate that the transposed matrix of w, b are bias term, X is point
Class initial data, it is assumed that the sample of classification initial data, which sorts out probability, meets formula (2),
In formula, p (c=+1 | X) indicates that X is judged as the probability of class c=+1, and p (c=-1 | X) indicate that X is judged as class
The probability of c=-1 obtains optimal space weight w by iteration weight weighted least-squares method, and specific iterative process passes through public affairs
Formula (3) realized,
In formula, X is classification initial data, XTFor the transposition of X, p is the classification probability of sample, and * represents inner product, and .* represents arrow
Amount product, d indicate one include each brain electricity sample category { 1,0 } column vector, i.e. L in sub-step (c), diag () can be with
By a vector median filters, diagonally matrix, column vector g represent gradient, and H is the gloomy battle array in sea that Fisher information matrix obtains, and Λ is repeatedly
Penalty factor during generation.During execution, the bias term b of script linear classification is integrated into the weight w of space, because
This does not occur in above-mentioned iterative process, is described by formula (4),
Space weight w is initialized as 10-3*ones(nchannels+ 1,1), wherein ones (nchannels+ 1,1) one is indicated
(nchannels+ 1) all 1's matrix × 1, nchannelsThe lead number being acquired is represented, iteration process is until space weight
Until w restrains, so that the space weight w of optimal classification is obtained, since entire assorting process is to be directed to starting point as 1+ (i-
1) τ, i=1,2 ..., K, data in the sliding window of a length of δ of window, therefore optimal classification space weight w is considered as w at this timeτ,δ, whereinExpression sampling number is TsampleThe sliding window number that separates of data, what τ indicated to slide every time
Distance；
(e) linearity curve is experienced according to classification results drafting, i.e., ROC curve and solves area under the curve；It will be in sub-step (d)
The space weight w for the optimal classification soughtτ,δSubstitute into y=wTIn X+b, classification results y is obtained, and calculate y's using formula (2)
Probability distribution, while ROC curve is drawn using the category sequence of the probability distribution result of classification results y and sub-step (c) construction,
And calculate the area Az under ROC curve；
(f) Az analysis of trend and evidence accumulation is positioned；The classification results that data in time windows are obtained
Az is counted, and the Az curve that window changes at any time is drawn.Wherein the curve of direction of motion identification is as shown in figure 3, and test hardly possible
Degree identification curve is as shown in Figure 4.It should be that one kind is presented to the classifying quality of two generic tasks due to the beginning accumulated with evidence
The trend of rising, i.e. evidence accumulation it is much more sufficient, the accuracy to make accurate judgment is higher, and ought to decision execute when,
Reach maximum；Therefore, the Az curve that window changes at any time is analyzed, from its time for continuing ascent stage of discovery observation
Point is accumulated start time to conclusion evidence, and is connect the moment as the online brain machine next based on sequential decision model
The information characteristics of port system accumulate start time.It can be seen that Az is all from the 7th sliding window in two articles of curves from Fig. 3 and Fig. 4
Position start that the variation tendency of lasting rising is presented, occur approximately in 210ms or so after stimulation occurs greatly corresponding to time shaft, because
This is using 210ms after stimulation as the starting point of maximum accumulation time window, so that unrelated classification information before excluding this moment point, only right
Effective classification information is accumulated, and the processing time is thus reduced, and achievees the purpose that the real-time for improving brain machine interface system.
Claims (1)
1. a kind of parameter optimization method of brain machine interface system decision model, it is characterised in that the following steps are included:
Step 1, acquisition EEG signals training data are simultaneously pre-processed, and acquire random spot moving direction using brain wave acquisition equipment
EEG signals under identification mission are as training data, the random point movement side different according to the direction of motion and experiment difficulty
To including that low difficulty is left, low difficulty is right, middle difficulty is left, middle difficulty is right, highly difficult left and highly difficult right six kinds different identification missions,
And collected EEG signals are filtered, remove artefact pretreatment；According to random spot moving direction label to the brain of acquisition
Electric signal carries out segment processing, and to distinguish different experiment examinations time, split time difficulty same for same subject
The split time length that different experiments examination time needs to be consistent, and taken needs to occur comprising stimulation to the mistake made a response
The process of journey, i.e. subject progress decision；
Step 2, linear space integration single probe method data classification and evidence accumulation initiating process positioning, pass through linear space
Integrated single probe method identifies two kinds of data categories in experiment: the identification to the direction of motion, left or right；To experiment
The identification of difficulty, it is difficult or easy, specifically include following sub-step:
(a) structural classification initial data X, classification initial data X should include two class data: left and right two class data of the direction of motion, i.e.,
X1And X2, wherein X1It is a nchannels×Tsample×N1Data matrix, X2It is a nchannels×Tsample×N2Number
According to matrix, wherein N1With N2Respectively represent the experiment examination number that two class data include, TsampleFor the split time used in step 1
Sampling number, nchannelsFor the lead number of use, and the initial data X=[X that classifies1,X2], it is a nchannels×Tsample
×(N1+N2) matrix, and obtained classification initial data X-form is converted into (Tsample×Ntrials)×nchannelsMatrix
Form is used in next Classification and Identification, NtrialsNumber is tried to carry out all experiments that the data of Classification and Identification are included,
Ntrials=N1+N2；
(b) sliding window is constructed, the window width of sliding window is δ sampled point, and single sliding distance is τ sampled point, for sampled point
Number is TsampleData can be divided into altogetherA sliding window, whereinIndicate the lower meaning being rounded, often
The position of secondary sliding window is [1+ (i-1) τ, δ+(i-1) τ], i=1,2 ..., K, using data in window to classifier respectively
It is trained, the time starting point accumulated by the variation tendency of classification accuracy come positive evidence；
(c) category sequence L, the cunning in original brain electricity classification data and sub-step (b) according to obtained in sub-step (a) are constructed
Dynamic window, category sequence L, L of the construction comprising element { 0,1 } are (δ × a NtrialsThe column vector of) × 1, wherein δ is sliding window
Window it is long, NtrialsNumber is tried to carry out all experiments that the data of Classification and Identification are included；
(d) logistic regression linear classifier is used, to the classification initial data estimation one in each sliding window constructed in step (b)
A optimal spatial weighted vector wτ,δ, which can carry out maximization identification to two class data in the window, pass through formula (1) and count
Calculation obtains classification results y,
Y=wTX+b (1)
In formula, w is the space weighted vector of the linear classifier, wTIndicate that the transposed matrix of w, b are bias term, X is that classification is original
Data, it is assumed that the sample of classification initial data, which sorts out probability, meets formula (2),
In formula, p (c=+1 | X) indicates that X is judged as the probability of class c=+1, and p (c=-1 | X) indicate that X is judged as class c=-
1 probability, optimal space weight w is obtained by iteration weight weighted least-squares method, and specific iterative process passes through formula (3)
It is realized,
In formula, X is classification initial data, XTFor the transposition of X, p is the classification probability vector of sample, and * represents inner product, and .* represents arrow
Amount product, d indicate the column vector comprising brain electricity sample category, i.e. L in sub-step (c), diag () are indicated a vector median filters
Diagonally matrix, g represent gradient vector, and H is the gloomy battle array in sea that Fisher information matrix obtains, Λ be punishment in iterative process because
Son, during execution, the bias term b of script linear classification is integrated into the weight w of space, is described by formula (4),
Space weight w is initialized as 10-3*ones(nchannels+ 1,1), wherein ones (nchannels+ 1,1) one is indicated
(nchannels+ 1) all 1's matrix × 1, nchannelsThe lead number being acquired is represented, iteration process is until space weight
Until w restrains, so that the space weight w of optimal classification is obtained, since entire assorting process is to be directed to starting point as 1+ (i-
1) τ, i=1,2 ..., K, data in the sliding window of a length of δ of window, therefore optimal classification space weight w is considered as w at this timeτ,δ, whereinExpression sampling number is TsampleThe sliding window number that separates of data, what τ indicated to slide every time
Distance；
(e) by the space weight w for the optimal classification sought in sub-step (d)τ,δIt brings into formula (1), obtains classification results y, and
The probability distribution of y is calculated using formula (2), while utilizing the class of the probability distribution result of classification results y and sub-step (c) construction
Linearity curve, i.e. ROC curve are experienced in mark sequence drafting, and calculate the area Az under ROC curve；
(f) the classification results Az that data in time windows obtain is counted, draws the Az curve that window changes at any time, by
It should be the trend that a kind of rising is presented to the classifying quality of two generic tasks in being accumulated with evidence, i.e., evidence is accumulated
Much, more sufficient, the accuracy to make accurate judgment is higher, and ought to reach maximum when decision executes；Therefore, at any time to Az
Between window change curve analyzed, observe its continue ascent stage start time, to conclusion evidence accumulate start time,
And start time is accumulated using the moment as the information characteristics of the online brain machine interface system next based on sequential decision model,
To exclude the unrelated classification information of accumulation, only effective classification information is accumulated, the processing time is thus reduced, reaches raising
The purpose of the real-time of brain machine interface system.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610329708.0A CN106022256B (en) | 2016-05-18 | 2016-05-18 | A kind of parameter optimization method of brain machine interface system decision model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610329708.0A CN106022256B (en) | 2016-05-18 | 2016-05-18 | A kind of parameter optimization method of brain machine interface system decision model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106022256A CN106022256A (en) | 2016-10-12 |
| CN106022256B true CN106022256B (en) | 2019-03-05 |
Family
ID=57097513
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610329708.0A Active CN106022256B (en) | 2016-05-18 | 2016-05-18 | A kind of parameter optimization method of brain machine interface system decision model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106022256B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TWI625682B (en) * | 2017-12-01 | 2018-06-01 | 財團法人工業技術研究院 | Methods, systems and non-transitory computer-readable medium for parameter optimization |
| CN109542221B (en) * | 2018-11-09 | 2020-07-24 | 北京航空航天大学 | Marking system for accurately acquiring brain signals of near infrared spectrum upper limb movement state |
| WO2021237429A1 (en) * | 2020-05-25 | 2021-12-02 | Center For Excellence In Brain Science And Intelligence Technology, Chinese Academy Of Sciences | A systematic device and scheme to assess the level of consciousness disorder by using language related brain activity |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101436100A (en) * | 2008-12-04 | 2009-05-20 | 上海大学 | Brain-electrical signal detection system and detection method for brain and machine interface |
| KR20140040309A (en) * | 2012-09-24 | 2014-04-03 | 재단법인대구경북과학기술원 | Apparatus and method for controlling of rehabilitation robot |
| CN104462716A (en) * | 2014-12-23 | 2015-03-25 | 北京理工大学 | Method for designing brain-computer interface parameters and kinetic parameters of brain controlled vehicle based on human-vehicle-road model |
| CN104951082A (en) * | 2015-07-09 | 2015-09-30 | 浙江大学 | Brain-computer interface method for intensifying EEG (electroencephalogram) signals through stochastic resonance |
-
2016
- 2016-05-18 CN CN201610329708.0A patent/CN106022256B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101436100A (en) * | 2008-12-04 | 2009-05-20 | 上海大学 | Brain-electrical signal detection system and detection method for brain and machine interface |
| KR20140040309A (en) * | 2012-09-24 | 2014-04-03 | 재단법인대구경북과학기술원 | Apparatus and method for controlling of rehabilitation robot |
| CN104462716A (en) * | 2014-12-23 | 2015-03-25 | 北京理工大学 | Method for designing brain-computer interface parameters and kinetic parameters of brain controlled vehicle based on human-vehicle-road model |
| CN104951082A (en) * | 2015-07-09 | 2015-09-30 | 浙江大学 | Brain-computer interface method for intensifying EEG (electroencephalogram) signals through stochastic resonance |
Non-Patent Citations (3)
| Title |
|---|
| The extraction of LRP via functional data analysis techniques;Yanbin Zhao 等;《Journal of Neuroscience Methods》;20120430;第206卷(第1期);全文 |
| 脑机接口中一种改进的模式识别方法;杨帮华 等;《中国生物医学工程学报》;20060430;第25卷(第2期);全文 |
| 脑机接口中半监督学习算法研究;王江 等;《电子测量技术》;20140531;第37卷(第5期);全文 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106022256A (en) | 2016-10-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Aslan et al. | Automatic Detection of Schizophrenia by Applying Deep Learning over Spectrogram Images of EEG Signals. | |
| CN109645990B (en) | Computer mode identification method for electroencephalogram signals of epileptics | |
| CN111329474B (en) | Electroencephalogram identity recognition method and system based on deep learning and information updating method | |
| CN113693613B (en) | Electroencephalogram signal classification method, electroencephalogram signal classification device, computer equipment and storage medium | |
| CN110353673B (en) | Electroencephalogram channel selection method based on standard mutual information | |
| CN112381008B (en) | Electroencephalogram emotion recognition method based on parallel sequence channel mapping network | |
| CN106022256B (en) | A kind of parameter optimization method of brain machine interface system decision model | |
| Bouallegue et al. | A dynamic filtering DF-RNN deep-learning-based approach for EEG-based neurological disorders diagnosis | |
| CN110399846A (en) | A kind of gesture identification method based on multichannel electromyography signal correlation | |
| CN110543831A (en) | brain print identification method based on convolutional neural network | |
| CN110781751A (en) | Emotional electroencephalogram signal classification method based on cross-connection convolutional neural network | |
| CN109009098B (en) | Electroencephalogram signal feature identification method under motor imagery state | |
| CN106648087B (en) | Feature EEG processing method based on consciousness task | |
| Hernandez et al. | Nonlinear evolution via spatially-dependent linear dynamics for electrophysiology and calcium data | |
| CN115414051A (en) | Emotion classification and recognition method of electroencephalogram signal self-adaptive window | |
| Lee et al. | Inter-subject contrastive learning for subject adaptive eeg-based visual recognition | |
| Chevallier et al. | Riemannian classification for SSVEP-based BCI: Offline versus online implementations | |
| Zhao et al. | Incremental common spatial pattern algorithm for BCI | |
| Ahmed et al. | Effective hybrid method for the detection and rejection of electrooculogram (EOG) and power line noise artefacts from electroencephalogram (EEG) mixtures | |
| CN110458066B (en) | Age group classification method based on resting electroencephalogram data | |
| CN109144277B (en) | Method for constructing intelligent vehicle controlled by brain based on machine learning | |
| Akrout et al. | Artificial and convolutional neural network of EEG-based motor imagery classification: A comparative study | |
| CN105787459A (en) | ERP signal classification method based on optimal score sparse determination | |
| CN110516711B (en) | Training set quality evaluation method of MI-BCI system and optimization method of single training sample | |
| CN114469137A (en) | Cross-domain electroencephalogram emotion recognition method and system based on space-time feature fusion model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |