Summary of the invention
In order to be optimized to leading of P300 brain-computer interface, satisfy the demand of P300 brain-computer interface, the invention provides a kind of P300 brain-computer interface optimization method that leads, see hereinafter for details and describe:
A kind of P300 brain-computer interface optimization method that leads said method comprising the steps of:
(1) obtain the incident phase locking value between leading according to individual leading of predetermined number L, the value of L is the positive integer more than or equal to 2; Set up the distance coefficient matrix between respectively leading according to said incident phase locking value, the phase synchronism intensity that definition is respectively led;
(2) according to said distance coefficient matrix between respectively leading, obtain clustering tree through the hierarchical clustering method;
(3) set up the importance ranking table that leads according to said clustering tree and said phase synchronism intensity of respectively leading;
(4) the said importance ranking table that leads is carried out accuracy and calculate, obtain L accuracy value;
(5) said L accuracy value sorted, the combination of leading that the accuracy value is the highest is as the optimum combination of leading, and exports the combination of leading of said optimum.
Said leading according to predetermined number L in the step (1) obtained the incident phase locking value between leading; Set up the distance coefficient matrix between respectively leading according to said incident phase locking value, the phase synchronism intensity that definition is respectively led is specially:
If obtain the eeg data of n goal stimulus and the eeg data of m non-goal stimulus through overtesting, the incident phase locking value of then leading between x and the y is specially:
Wherein, the value of n and m is the positive integer more than or equal to 1;
<>
τRepresentative is carried out time average to the formula in the bracket;
Define said distance coefficient matrix between respectively leading, for two lead x and y arbitrarily, its distance coefficient matrix is specially:
According to said distance coefficient matrix between respectively leading, define said phase synchronism intensity of respectively leading, for any x that leads, its phase synchronism intensity is:
Said phase synchronism intensity of respectively leading according to said clustering tree and said in the step (3) is set up the importance ranking table that leads and is specially:
Suppose to have L eeg data that leads, obtain the clustering tree that a L leads, the activity that is provided with L-1 features sources is reflected in that L is individual lead on; Through the intercepting clustering tree; Individual leading of L is divided into the L-1 class, then exists one to comprise two classes of leading, be designated as the lead x and the y that leads; When the phase synchronism intensity of the x that leads during greater than the phase synchronism intensity of the y that leads, the x that then selects to lead gets rid of the y that leads; The importance of the y that leads that gets rid of comes the L position; Lead for remaining L-1; The activity that is provided with L-2 features sources be reflected in L-1 lead on, through the intercepting clustering tree, L-1 led is divided into the L-2 class; Then exist one to comprise two classes of leading; Be designated as the lead x and the y that leads, when the phase synchronism intensity of the x that leads during greater than the phase synchronism intensity of the y that leads, the x that then selects to lead gets rid of the y that leads; The importance of the y that leads that gets rid of then comes the L-1 position; By that analogy, up to only surplus one lead till, remaining at last leads, and comes L-(L-1) position, obtains the L importance ranking table that leads.
Said in the step (4) carries out accuracy to the said importance ranking table that leads to be calculated, and obtains L accuracy value and is specially:
The data of leading of L-(L-1) position are carried out accuracy calculate, obtain the first accuracy value A1; The data of leading of L-(L-1) position and L-(L-2) position are carried out accuracy calculate, obtain the second accuracy value A2; The data of leading of L-(L-1) position, L-(L-2) position and L-(L-3) position are carried out accuracy calculate, obtain the 3rd accuracy value A3; By that analogy, up to obtaining L accuracy value AL.
The beneficial effect of technical scheme provided by the invention is:
The invention provides a kind of P300 brain-computer interface optimization method that leads; Fusion event PLV of the present invention (Phase Locking Value; The phase locking value) and the hierarchical clustering method; The features sources number being inferred decision-making, select the optimum combination of leading on this basis, is a kind of brand-new P300 brain-computer interface optimisation technique of leading; This invention can effectively reduce the redundancy of BCI system and lead; Reduce the instability risk of BCI system; Simplify procedures, online degree and its commercialization of promotion of improving the BCI system are offered help, can obtain considerable economic and social benefit in field of human-computer interaction.
Embodiment
For making the object of the invention, technical scheme and advantage clearer, will combine accompanying drawing that embodiment of the present invention is done to describe in detail further below.
In order to be optimized to leading of P300 brain-computer interface, satisfy the demand of P300 brain-computer interface, the embodiment of the invention provides a kind of P300 brain-computer interface optimization method that leads, and sees hereinafter for details and describes:
101: lead according to predetermined number L and to obtain the incident phase locking value between leading; Set up the distance coefficient matrix between respectively leading according to the incident phase locking value between leading, the phase synchronism intensity that definition is respectively led;
Wherein, the predetermined number L that leads sets according to the needs in the practical application, and value is the positive integer more than or equal to 2, and when specifically realizing, the embodiment of the invention does not limit this.
If a certain signal s (t), its analytical function z (t) can obtain by through type (1-1)
Signal
is that the Hilbert of s (t) changes, and can write out the instantaneous phase of signal s (t):
Suppose that predetermined number is 2, promptly have two EEG signals s that lead
x(t) and s
y(t), its phase place is respectively
With
The phase locking value of x and y of then leading can be obtained by formula (1-4):
Wherein,<>
τRepresentative is carried out time average to the formula in the bracket, and the window width of its time window is τ (τ=25ms in the embodiment of the invention is that example describes).If the complete synchronism of the phase place of two lead signals,
Then be stable, PLV
X, y(t)=1; If this two lead signal phase place do not have synchronism, then
Be stochastic distribution, PL
X, y(t)=0.
If behind overtesting, obtain the eeg data of n goal stimulus target and the eeg data of m non-goal stimulus nontarget, then lead x and the incident phase locking value of leading between the y can be obtained by formula (1-6):
Wherein, the value of n and m is the positive integer more than or equal to 1.
The distance coefficient matrix D ist of definition between respectively leading, for two lead x and y arbitrarily, its distance coefficient matrix D ist:
According to the distance coefficient matrix D ist between respectively leading, the phase synchronism intensity Wg that respectively leads of definition, for any x that leads, its phase synchronism intensity is:
102: the distance coefficient matrix according between respectively leading, obtain clustering tree through the hierarchical clustering method;
Wherein, The hierarchical clustering method; Claiming the grade clustering procedure again, is present one of the clustering method the most fully that uses at most, studies, and its basic thought is through foundation and progressively upgrades the distance coefficient matrix between respectively leading; Find out and merge immediate two types, merged into till one type up to whole cluster objects.The embodiment of the invention utilizes the method (also having bee-line method, longest distance method, gravity model appoach and type method of average in addition) based on ward to carry out hierarchical clustering to leading, and obtains clustering tree.Fig. 4 is the structural representation of clustering tree, and wherein on behalf of difference, numeral 1,2,3,4,5 lead respectively, and ordinate is represented distance.Lines a leads 5 and is divided into 4 types, and wherein leading 12 is one type with leading, and other lead and respectively become one type; Lines b then leads 5 and is divided into 3 types, and wherein leading 12 is one type with leading, and leading 3 is one type, and leading 4 and 5 is one type; Lines c leads 5 and is divided into 3 types, and wherein leading 1,2 and 3 is one type, and leading 4 and 5 is one type.
103: set up the importance ranking table that leads according to the clustering tree and the phase synchronism intensity of respectively leading;
For example: certain two lead x and y, record be of the reflection of same features sources at scalp, will there be a lot of similar features in the signal of lead so x and y, thereby can produce the characteristic redundancy.In other words, if utilized the x that leads exactly, the y that leads so with very little, that is to say that the importance of the y that leads has descended to the influence of last classification accuracy rate.Therefore, confirm to reflect the number with features sources of leading of features sources, effective minimizing number that leads is had great significance.
That is, setting up the importance ranking table that leads according to clustering tree and the phase synchronism intensity of respectively leading is specially: suppose to have the individual eeg data that leads of L, obtain the clustering tree that a L leads; The activity that is provided with L-1 features sources be reflected in L lead on; Through the intercepting clustering tree, individual leading of L is divided into the L-1 class, then exist one to comprise two classes of leading; Be designated as the lead x and the y that leads; When the phase synchronism intensity of the x that leads during greater than the phase synchronism intensity of the y that leads, the x that then selects to lead gets rid of the y that leads (maybe when the phase synchronism intensity of the x that leads during smaller or equal to the phase synchronism intensity of the y that leads, the y that then selects to lead gets rid of the x that leads); The importance of the y that leads that gets rid of then comes L position (or the importance of the x that leads that gets rid of then comes the L position); Lead for remaining L-1, the activity that continues to be provided with L-2 features sources is reflected in that L-1 is individual lead on, through the intercepting clustering tree; Individual leading of L-1 is divided into the L-2 class; Then exist one to comprise two classes of leading, be designated as the lead x and the y that leads, when the phase synchronism intensity of the x that leads phase synchronism intensity greater than the y that leads; The x that then selects to lead gets rid of the y that leads (maybe when the phase synchronism intensity of the x that leads during smaller or equal to the phase synchronism intensity of the y that leads, the y that then selects to lead gets rid of the x that leads); The importance of the y that leads that gets rid of then comes L-1 position (or the importance of the x that leads that gets rid of then comes the L position); By that analogy, up to only surplus one lead till, remaining at last leads, and comes L-(L-1) position, obtains the importance ranking table that a L leads.
Suppose the eeg data that has 64 to lead, to have obtained one 64 clustering tree of leading through two steps in front.Because do not know brain inside have actually the activity in several characteristic source be reflected in these 64 lead on, so carry out the decision-making of inferring of features sources number.The first step, the activity that is provided with 63 features sources be reflected in 64 lead on, so, exist a redundancy to lead.In order to find this redundancy to lead, through the intercepting clustering tree, 64 led is divided into 63 types.So, have so class, wherein contain two and lead, be designated as x and y, other types then have only one to lead.Think that lead x and y is the reflection to same features sources, therefore need make screening to it, and the foundation of selecting is exactly the phase synchronism intensity Wg that leads.Suppose to lead the phase synchronism intensity Wg of x greater than the phase synchronism intensity Wg of the y that leads, then select x to get rid of y.The importance of the y that leads that gets rid of then comes the 64th.Second step, lead for remaining 63, the activity that continues to be provided with 62 features sources be reflected in 63 lead on, so, exist a redundancy to lead.In order to find this redundancy to lead, through the intercepting clustering tree, 63 led is divided into 62 types.So, have so class, wherein contain two and lead, be designated as x and y, other types then have only one to lead.Think that lead x and y is the reflection to same features sources, therefore need make screening to it.And the foundation of selecting is exactly the phase synchronism intensity Wg that leads.Suppose to lead the phase synchronism intensity Wg of x greater than the phase synchronism intensity Wg of the y that leads, then select x to get rid of y.The importance of the y that leads that gets rid of then comes the 63rd.By that analogy, up to only surplus one lead till, then last remaining that leads, its importance is maximum, and is last, obtains one 64 importance ranking table that leads.
104: the importance ranking table that leads is carried out accuracy calculate, obtain L accuracy value;
Wherein, this step is specially carries out accuracy calculating to the data of leading of L-(L-1) position, obtains the first accuracy value A1; The data of leading of L-(L-1) position and L-(L-2) position are carried out accuracy calculate, obtain the second accuracy value A2; The data of leading of L-(L-1) position, L-(L-2) position and L-(L-3) position are carried out accuracy calculate, obtain the 3rd accuracy value A3; By that analogy, up to obtaining L accuracy value AL.
Wherein, Accuracy is calculated as: carry out 5 folding cross validations calculating accuracy according to the eeg data of n goal stimulus target and the eeg data of m non-goal stimulus nontarget; Be about to all data and be divided into 5 parts of B1, B2, B3, B4 and B5 at random, get B1 earlier as test sample book, other 4 piece of data are as training sample; The sorter of setting up with training sample goes to calculate the accuracy of test sample book B1, obtains accuracy value R1.Get B2, B3, B4 and B5 then successively as test sample book, other are training sample, obtain accuracy value R2, R3, R4 and R5 respectively.The mean value of getting these 5 accuracy values at last is the accuracy of 5 folding cross validations.Wherein, During concrete the realization, also can carry out 6 folding cross validations according to the needs in the practical application and calculate accuracy, be about to all data and be divided into 6 parts of B1, B2, B3, B4, B5 and B6 at random; Get B1 earlier as test sample book; Other 5 piece of data are described in detail referring to preceding text as training sample, repeat no more at this.
105: L accuracy value sorted, and the combination of leading that the accuracy value is the highest is as the optimum combination of leading, and exports the optimum combination of leading.
Wherein, can carry out from high to low or from low to high ordering L accuracy value, or adopt other sort method, when specifically realizing, the embodiment of the invention does not limit this.
The feasibility of leading optimization method with an a kind of P300 brain-computer interface of verifying that simply the embodiment of the invention provides below, see hereinafter for details and describe:
A kind of P300 brain-computer interface that the application of the invention embodiment provides optimization method that leads; 64 of 5 experimenters are led be optimized; The result shows, on average can reduce 46 and lead under the constant situation of accuracy (obtaining through 3 stack eigenwerts) keeping; Verify the feasibility of the method that the embodiment of the invention provides, satisfied the needs in the practical application.
In sum; The embodiment of the invention provides a kind of P300 brain-computer interface optimization method that leads; Embodiment of the invention fusion event PLV and hierarchical clustering method; The features sources number being inferred decision-making, select the optimum combination of leading on this basis, is a kind of brand-new P300 brain-computer interface optimisation technique of leading; The embodiment of the invention can effectively reduce the redundancy of BCI system and lead; Reduce the instability risk of BCI system; Simplify procedures, online degree and its commercialization of promotion of improving the BCI system are offered help, can obtain considerable economic and social benefit in field of human-computer interaction.
It will be appreciated by those skilled in the art that accompanying drawing is the synoptic diagram of a preferred embodiment, the invention described above embodiment sequence number is not represented the quality of embodiment just to description.
The above is merely preferred embodiment of the present invention, and is in order to restriction the present invention, not all within spirit of the present invention and principle, any modification of being done, is equal to replacement, improvement etc., all should be included within protection scope of the present invention.