CN104635934A - Brain-machine interface method based on logic thinking and imaginal thinking - Google Patents

Abstract

The invention discloses a brain-computer interface method and device based on logic thinking and imaginal thinking and relates to the field of novel pattern of brain-computer interface technology and feature extraction and classification. An experiment requires a testee for mental calculation and a three-dimensional object to rotate, signals of six electrodes are collected, and a common spatial pattern is adopted to extract feature values. Aiming at the problem that the common spatial pattern is suitable for multiple electrodes and the feature values, a common spatial filter algorithm is improved, electroencephalogram signals of one electrode are subjected to separation of different frequency bands and time periods, and the electroencephalogram signals are subjected to feature extraction calculation. The feature values are subjected to classification of a support vector machine, and excellent classification effect can be realized by the data processing mode through offline data analysis. A brain-computer interface based on logic thinking and imaginal thinking is constructed, reliability of a system is improved, experiment imaging difficulty and visual fatigue caused by long-term experiments are reduced, comfortability of the brain-computer interface system is improved, and application population range is expanded.

Description

The brain-machine interface method of logic-based thinking and thinking in images

Technical field

The present invention relates to artificial intelligence field and EEG's Recognition field, particularly relate to the brain-machine interface method of a kind of logic-based thinking and thinking in images.

Background technology

Brain-computer interface (Brain-Computer Interface is called for short BCI) does not rely on normally by nervus peripheralis and sarcous output channel, is the communication system of a kind of direct connection brain and computing machine and external unit.Brain machine interface system have gather without wound, the temporal resolution advantage of advantage simple to operate and uniqueness.Brain machine interface system is made up of four modules usually: electroencephalogramsignal signal acquisition module, EEG feature extraction module, EEG signals tagsort module and peripheral control module.Characteristic extracting module and tagsort module are the most crucial parts of whole brain-computer interface, and EEG signals can be converted to by these two modules just can by the control signal of external unit identification.The feature of EEG signals has faint property, complicacy and instability, due to these characteristics of EEG signals, has higher requirement for normal form design and feature extraction and classifying.

The eeg signal acquisition mode realizing brain-computer interface at present mainly contains two kinds: implanted eeg signal acquisition mode and non-built-in mode eeg signal acquisition mode.Implanted eeg signal acquisition mode mainly comprises Magnetic resonance imaging (FMRI), magneticencephalogram (EMG) and near infrared spectrum (NIRS) etc.The acquisition mode of non-built-in mode compensate for inconvenience that implanted collection brings and has the various problems such as wound property, wherein electroencephalogram (Electroencephalograph, be called for short EEG) be the main mode realizing non-built-in mode brain-computer interface at present, also extensively thought most potential technology.Compared with implanted electrode, non-built-in mode electrode have temporal resolution high, be easy to advantages such as using, cost is lower, be easy to carry.The main non-built-in mode electrode acquisition mode adopted has Mental imagery, P300 and Steady State Visual Evoked Potential (Steady-State Visual Evoked Potentials is called for short SSVEP) now.

In recent years, utilize different thinking task EEG signals as a kind of new technique realizing BCI, received and paid close attention to widely.Thinking activities are one of marks of the wisdom of humanity.When people carry out Mental task, can there is change to a certain degree in brain specific region neurocyte bioelectrical activity, these bioelectrical activities can reflect in scalp surface, can be collected by electroencephalogramsignal signal collection equipment.Brain is activate corticocerebral zones of different and the physiological phenomenon that can induce different brain wave rhythms makes to utilize EEG signals to carry out classification to logical thinking and thinking in images becomes possibility carrying out different thinking activities.According to Neuropsychology theoretical analysis, imagine that mental arithmetic and Space Rotating are complicated processes in the heart, the behavior of calculating may comprise that numeral identifies, the understanding of numeric character implication, the selection of computing method, operation rule and specific calculations program performs, the temporary transient storage of intermediate result and extract and remember the tenure composition such as expression of result again.When dextro manuality subject calculates, primary activation left dorsolateral prefrontal cortex and bilateral top layer cortex; And when subject is when carrying out the thinking activities relevant with language, mental arithmetic, the α ripple of left hemisphere is less than right hemisphere; When carrying out the Tasks relevant with space imagination, situation is contrary.By realizing brain-computer interface to the classification of different Mental task spontaneous brain electricity signal, directly utilize Mental task to control external unit, closer to the working control pattern of people to external unit, send and utilized evoked brain potential signal, the difficulty that the unnecessary artefact of generation and Evoked ptential extract, locate.And relative to other the pattern such as SSVEP, the head pattern of logical thinking and thinking in images is more convenient to use, Long-Time Service preferably.

The method of seeking effective feature extraction and classification improves one of recognition accuracy gordian technique.Conventional feature extracting method has power spectrumanalysis, wavelet transformation, cospace pattern algorithm (common spatial pattern is called for short CSP) etc.CSP algorithm is considered to a kind of effective brain power mode analytical approach.The advantage of CSP does not need to select the specific frequency band of subject in advance, and also without the need for the space distribution information of EEG signals, but task based access control processes EEG signal; Shortcoming needs a large amount of proper vectors and a large amount of electrodes, and too much electrode can cause the decline of whole brain-computer interface portability.The object of CSP algorithm is searching spatial filter, make two kinds of signals after algorithm process, distinguished to greatest extent, algorithm is based on two Simultaneous Diagonalization of Covariance Matrices, and while making a wherein class signal variance maximum, the variance of another kind of signal is minimum.

Support vector machine classifier shows a lot of distinctive advantage in non-linear, the small sample of solution and high dimensional pattern identification, and can promote the use of in the other machines problems concerning study such as Function Fitting.Support vector machine has outstanding learning performance and can maximize Geometry edge district and the characteristic minimizing experience error simultaneously, is thus widely used in the classification problem of brain-computer interface as sorter.Support vector machine main thought is that the Non-linear Kernel function selected in advance by certain is by the feature space of the spatial mappings of linearly inseparable to the linear separability of a higher-dimension, at this space utilization structural risk minimization structure optimal separating hyper plane, making interval between the inhomogeneity sample that this plane is nearest, classifying face both sides maximum, can realize by solving a convex quadratic programming problem on the construction problem of optimal hyperlane.The important advantage of of support vector machine method is the number that the complexity of obtained sorter can adopt support vector, instead of the dimension of transformation space is portrayed.

Summary of the invention

Goal of the invention: in order to overcome deficiency of the prior art, the invention provides the brain-machine interface method of a kind of logic-based thinking and thinking in images, and solving original CSP algorithm needs a large amount of proper vector and electrode and cause the technical matters that calculates and operation is all inconvenient.

Technical scheme: the two cerebral hemispheres of brain also exists the obvious division of labor, and it is verified, as the synchronization in Mental imagery/desynchronization phenomenon (ERD/ERS), similar with Mental imagery, when carrying out different logical thinkings and thinking in images activity, certain EEG signals can be produced in corresponding brain area, the two cerebral hemispheres of brain can detect that EEG signals is asymmetric equally, this feature makes to utilize EEG signals to distinguish different Mental task becomes possibility, also becomes simultaneously and utilizes different Mental task to one of basis designing brain-computer interface.

Based on the above-mentioned fact, for realizing object of the present invention, the technical scheme of employing is as follows:

The brain-machine interface method of logic-based thinking and thinking in images, comprises the following steps that order performs:

Step one, electroencephalogramsignal signal acquisition module is set at subject's head, by the normal form of logical thinking task and thinking in images task, utilizes brain wave acquisition method of singly leading to collect the original EEG signals of subject under different stimulated state;

Step 2, original EEG signals carried out to the three-dimensional process of space, time and frequency;

Step 3, under cospace pattern to after step 2 process EEG signals extract proper vector;

Step 4, by support vector machine to extracting the proper vector cross validation classification obtained in step 3, obtain eeg signal classification result and export.

Further, in the present invention, when gathering original EEG signals, the crosslinking electrode comprised as upper/lower positions is set in electroencephalogramsignal signal acquisition module: on prefrontal lobe, top and temporal lobe, all arrange a pair crosslinking electrode, and two the crosslinking electrode positional symmetry belonged to a pair are arranged, specifically comprise F3 and F4 laying respectively at prefrontal lobe both sides, lay respectively at C3 and C4 of top both sides and lay respectively at P3 and P4 of temporal lobe both sides.According to Neuropsychology theoretical analysis, when dextro manuality subject calculates, primary activation left dorsolateral prefrontal cortex and bilateral top layer cortex; And when subject is when carrying out the thinking activities relevant with language, mental arithmetic, the α ripple of left hemisphere is less than right hemisphere; When carrying out the Tasks relevant with space imagination, situation is contrary; Therefore, crosslinking electrode is set according to above-mentioned position and can obtains the EEG signals effectively distinguishing logical thinking and thinking in images.

Further, in the present invention, the normal form design of single experiment is as follows: first subject keeps the tranquility at least 2 seconds, and EEG signals corresponding under gathering this tranquility as a reference; Then logical thinking task or thinking in images task is carried out, wherein logical thinking task is for occur that in screen formula, subject carry out mental arithmetic according to formula, thinking in images task for there is a shape in screen, subject imagines that this shape rotates at three dimensions.Brain-computer interface output category signal can be applied to concrete control module, the classification of different EEG signals is utilized to control control objectives, research shows that traditional Mental imagery imagines comparatively difficulty concerning subject, logical thinking and thinking in images are then comparatively simple, therefore adopt these two different thoughtcast to test subject.And the single acquisition time that often kind of thinking task is corresponding and single experiment need a lasting process, in order to the convenient test of subject and the data processing in later stage, the time that can set single acquisition EEG signals was a fixing duration, as 8 seconds.

Further, in the present invention, in step one, adopt brain wave acquisition method of singly leading to gather the original EEG signals of these 6 crosslinking electrodes of F3, F4, C3, C4, P3, P4, reference electrode is arranged on left side mastoid location, and the earth signal that leads is arranged on forehead place.

Further, in the present invention, the original EEG signals obtained in step one is carried out pre-service in the following manner: gather electro-ocular signal by paired two crosslinking electrodes, to original EEG signals through removing above-mentioned electro-ocular signal, Muscle artifacts, removal baseline wander, data sectional process, and then obtain pretreated EEG signals.

Off line data analysis is carried out to the EEG signals that pre-service completes, adopts the analytical approach of spatial domain-time-frequency domain:

First Spatial domain analysis is carried out, because the EEG signals gathered has the interference of spatial noise, cause for a thinking task brain may have multiple region enliven, as far as possible the effect of Laplacian space filtering eliminates these spatial noises, the distinctive feature of brain area that outstanding thinking task is corresponding.Therefore the EEG signals after pre-service being completed carries out Laplacian space filtering, here need to arrange 4 crosslinking electrodes respectively in the adjacent position up and down of these 6 crosslinking electrodes of F3, F4, C3, C4, P3, P4 in advance, the basic thought of Laplacian space filtering is that the mean value of the EEG signals of EEG signals and its four the adjacent up and down crosslinking electrodes collected by the crosslinking electrode of needs subtracts each other, eliminate spatially to the interference of EEG signals, the EEG signals after obtaining spatial filtering.

Next carries out frequency-domain analysis and time-domain analysis.During frequency-domain analysis, Short Time Fourier Transform will be carried out to domain space through the filtered EEG signals of Laplacian space, then EEG signals is carried out butterworth filter, extract most active five the main frequency ranges of following human thinking activities, 4Hz-7Hz, 7Hz-10Hz, 10Hz-13Hz respectively, 13Hz-20Hz and 20Hz-30Hz, forms 5 frequency domain EEG signals; In normal form design, specify that subject carries out Mental task length task time is at every turn 5 seconds, during time-domain analysis, need to neglect first 500 milliseconds and last 500 milliseconds in 5 second time period of carrying out Mental task task, because within these two times, subject often has no idea to concentrate on, remaining four seconds are divided into four sections, every each period of 1 second, form 4 time domain EEG signals.

By the three dimensional analysis of above-mentioned spatial domain-time-frequency domain, the EEG signals of 5 conventional frequency sections of single electrode and the EEG signals of 4 time periods, form the proper vector of 20 dimensions, meet the requirement of the eigenwert quantity using single electrode cospace pattern algorithm, then use single electrode cospace pattern algorithm to carry out characteristic vector pickup.

Described single electrode cospace pattern algorithm, concrete steps are as follows:

I-th normalized covariance matrix of sample is

$C_i=\frac{X_i{X}_i^T}{\mathrm{trace}\left(X_i{X}_i^T\right)}$

X in formula _irepresent the EEG signals that i-th thinking task gathers, dimension is frequency × time, represents the dimension of time and frequency, temporarily thinks C for the ease of understanding _imatrix is M × N rank matrixes.T represents transpose of a matrix, and matrix diagonals line element sum is asked in trace representative, namely asks matrix trace.

Order

$C_l=\underset{i{\∈R}_l}{\mathrm{\Σ}}{C}_i$

$C_m=\underset{i{\∈R}_m}{\mathrm{\Σ}}{C}_i$

C _l, C _mrepresent the covariance matrix of logical thinking task and thinking in images task respectively, R _land R _mrepresent two kinds of different thinking task-set respectively.Two class thinking task covariance sums are expressed as C=C _l+ C _m, then Eigenvalues Decomposition is carried out to C, obtains

In formula, Λ is the diagonal matrix of M × M rank eigenwert composition, U ₀be M × M characteristic of correspondence vector matrix, can obtain whitening matrix is

The effect of whitening matrix P makes two covariance matrix C _land C _mcan simultaneous diagonalization after albefaction, so covariance matrix can change into

S _m＝PC _mP ^T＝U∧ _mU ^T

S _l＝PC _lP ^T＝U∧ _lU ^T

Wherein, S _l, S _mhave common proper vector U, and characteristic of correspondence value sum is I, i.e. ∧ _m+ ∧ _l=I.

Step below, for selecting spatial filter, by calculating, defines projection matrix W=U ^tp is spatial filter.For the EEG signals that i-th thinking task collects, the signal after spatial filtering is Z _i=WX _i.

Usually, two class samples very different when projecting in new coordinate axis after the conversion of above-mentioned spatial filter.Usually get maximum or minimum m stack features vector as the feature of this sample, obtain proper vector by computing below

$f_i=\log \left(\frac{{\mathrm{VAR}}_i}{\underset{i=1}{\overset{2}{\mathrm{\Σ}}}{\mathrm{VAR}}_i}\right)$

In formula, i represents different thinking tasks, and the sorter f value drawn being input to support vector machine carries out training and classifies; The proper vector number being obtained a kind of thinking task by analysis is l × 2m.

Accordingly, in brain-computer interface of the present invention, in step 4, proper vector inputted in the support vector machine trained, classification also cross validation obtains the classification results of the EEG signals that each electrode collects.

The basic thought of support vector machine finds optimum lineoid to make the data of two states apart from maximum, better classification accuracy rate can be obtained, original support vector machine problem can be expressed as following form, to meet under constraint condition (1), make function (2) obtain minimum value.

y _i(w ^Tx _i-b)-1≥0 (1)

$\min \frac{1}{2}{\left|\right|w\left|\right|}^2---\left(2\right)$

In formula, x _irepresent the input of support vector machine, y _ifor x _igeneric, w is the weight coefficient vector of classifying face, and b is classification thresholds.

In High dimensional data model, affect the selection that the topmost problem of svm classifier algorithm is kernel function, in the brain-computer interface of logic-based thinking and thinking in images, the kernel function of selection is Radial basis kernel function, and expression-form is as follows,

k(x _i,x _j)＝exp(-g||x _i-x _j|| ²)

In order to allow the tag along sort of mistake, introduce error ξ and penalty factor c, so expression formula is rewritten as follows,

y _i(w ^Tx _i-b)-1+ξ _i≥0 (3)

$\min \frac{1}{2}{\left|\right|w\left|\right|}^2+c\underset{i}{\mathrm{\Σ}}{\mathrm{\ξ}}_{\text{i}}---\left(4\right)$

Beneficial effect:

Logical thinking and thinking in images combine in the difference of the active regions of brain and brain-computer interface by the present invention, construct the brain-computer interface of logic-based thinking and thinking in images, and the method that have employed spatial domain-time-frequency domain three-dimensional carries out data processing, possibility is provided to single electrode cospace pattern, decrease number of electrodes, improve rate of information transmission and the reliability of brain-computer interface;

And traditional thinking task adopts encephalic technology to check area-of-interest, implement comparatively complicated, and cospace pattern and support vector cassification are used in thinking task by the present invention, resolution is high and simple to operate;

Choose thinking task for carrying out mental arithmetic and imagination space rotational tasks, the task of comparing traditional Mental imagery is more easy, also the difficulty of subject's operation is effectively reduced, compared with SSVEP, there is no very strong vision burden after long-time use, improve the applicable crowd of control system simultaneously yet.

Accompanying drawing explanation

Fig. 1 is the structural representation of the brain-computer interface of logic-based thinking of the present invention and thinking in images;

Fig. 2 is different thinking task basic comprising schematic diagram of the present invention;

Fig. 3 is that employing brain wave acquisition equipment of the present invention leads and arranges schematic diagram;

Fig. 4 is the frequency domain signal diagrams of logical thinking task of the present invention;

Fig. 5 is the frequency domain signal diagrams of thinking in images task of the present invention;

Fig. 6 is the feature vector chart of the logical thinking task based on single electrode cospace schema extraction of the present invention;

Fig. 7 is the feature vector chart of the thinking in images task based on single electrode cospace schema extraction of the present invention;

Fig. 8 is support vector machine Selecting parameter figure of the present invention.

Embodiment

Below in conjunction with accompanying drawing, the present invention is further described.

As shown in Figure 1, the brain-computer interface device of a kind of logic-based thinking and thinking in images, be made up of display 1, signal acquiring system 2, signal processor 3 and external control device 4, display 1 is connected with signal processor 3 respectively with signal acquiring system 2, and wherein signal processor 3 controls display 1 and shows required normal form content; Information acquisition system 2 gathers, process EEG signals and be sent to signal processor 3; Signal processor 3 pairs of EEG signals carry out characteristic vector pickup and classification, and export result to external control device 4; Here, information acquisition system 2 is made up of brain electricity cap, brain electricity amplification module, brain electricity filtration module, eeg recording and pretreatment module, EEG signals processed offline module, EEG signals online acquisition module; The sample frequency of brain electricity amplification module is 250Hz, the bandpass filter of brain electricity filtration module to be filter range be 0.1Hz to 100Hz; Because needs remove eye electricity, Muscle artifacts, therefore also need two paired electrodes and gather electro-ocular signal in thinking task.

Case study on implementation one

The present embodiment provides the brain-machine interface method of a kind of logic-based thinking and thinking in images, as shown in Figure 1, comprising: visual stimulus module, electroencephalogramsignal signal acquisition module, characteristic extracting module, tagsort module and control system; Visual stimulus module is connected with electroencephalogramsignal signal acquisition module, and electroencephalogramsignal signal acquisition module is connected with characteristic extracting module, and characteristic extracting module is connected with tagsort module, and tagsort module is connected with control system.Wherein, described visual stimulus module is in order to bring out the EEG signals for logical thinking and thinking in images; Described electroencephalogramsignal signal acquisition module is in order to gather EEG signals; Described characteristic extracting module is in order to carry out the extraction that single electrode cospace pattern carries out eigenwert by EEG signals complete for pre-service; Described tagsort module, in order to the eigenwert extracted to be trained, is selected optimum parameter, is then classified, cross validation etc.; Described control system, in order to the EEG signals of having classified to be transferred to the external unit of needs control by bluetooth or infrared equipment, makes brain-computer interface be applied in practicality.

Described visual stimulus module comprises mental arithmetic stimulation and Space Rotating stimulating unit.Concrete normal form is as shown in the time scale of Fig. 2, task at first, before subject is sitting in display 1, the screen of range display 11 to 2 meter, then task explanation is carried out to subject and normal form is trained, when single thinking task at first, there is "+" in screen, represent and prepare to start, subject is ready, then there will be the blank screen time about 2 seconds, using the EEG signals that collects in this time of the 2 seconds signal as tranquility, can at follow-up Data processing as a reference; Then the 3rd second time screen on there is the prompting of different thinking task, in time there is formula " 100-7 " in screen, subject is within 5 seconds follow-up imagination times of rising for the 4th second, ceaselessly deduct 7 with 100, when there is letter " L " in screen, subject is within 5 seconds follow-up imagination times of rising for the 4th second, and imagination letter is in three-dimensional rotation; When terminating this thinking task after full 5 seconds of the imagination duration.During experiment, need carry out 5 groups of thinking tasks altogether, often in group, 2 kinds of thinking tasks carry out 10 times, and single task continues 8 seconds, and every group task interval 10-15 minute, needs about 2 hours altogether from preparing to complete.

Described electroencephalogramsignal signal acquisition module comprises brain electricity cap, brain electricity amplification module, brain electricity filtration module, eeg recording module, EEG signals processed offline module, EEG signals online acquisition module, single crosslinking electrode is Ag/AgCl material and is arranged on brain electricity cap, and described single crosslinking electrode is provided with amplifier; Brain electricity cap is in order to be worn over head; Single crosslinking electrode is in order to gather EEG EEG signals; Because EEG signals is very faint, so need brain electricity amplification module to amplify EEG signals, the enlargement factor of brain electricity amplification module is 20000 times.

What adopt in the present invention is the electrode connection mode singly led, according to the region that the brain when carrying out thinking task relatively enlivens, the EEG signals of P3 and P4 electrode place totally 6 main electrodes gather F3 and F4 being positioned at prefrontal lobe, C3 and C4 being positioned at top, being positioned at temporal lobe, in the process gathering brain electricity, reference electrode is positioned at the position of left side mastoid process, GND signal is at forehead place, and the distribution of concrete electrode as shown in Figure 3.In order to remove the artefact of eyes, electro-ocular signal is gathered by two paired crosslinking electrodes.In invention, brain electricity amplification module is 64 passage brain wave acquisition equipment, and sample frequency is 250Hz, and in gatherer process, bandpass filter scope is 0.1Hz to 100Hz.

Described EEG feature extraction module comprises again data preprocessing module, data sectional processing module, single electrode cospace module; Data preprocessing module is connected with brain electricity amplification module, and the EEG signals in order to extracting is carried out removal eye and moved and Muscle artifacts interference, removes baseline interference etc.; Data sectional processing module is connected with data preprocessing module, and the EEG signals in order to pre-service to be completed carries out the process of spatial domain-time-frequency domain three-dimensional data; Single electrode cospace module is connected with data sectional processing module, obtains logical thinking and thinking in images brain electrical feature vector in order to use single electrode cospace pattern algorithm to the EEG signals of carrying out three-dimensional data process.

For the course of work of actual conditions summary EEG feature extraction module, this process comprises the following steps:

(1) EEG signals that brain wave acquisition equipment collects is carried out the pre-treatment step such as artefact removal, baseline calibration;

(2) EEG signals pre-service completed, carries out segmentation according to each single task EEG signals of 8 seconds in time domain, regards the zero point that data start when thinking task being occurred as, and namely the segmentation of single task is from-2 seconds to 6 seconds;

Next (3) F3, F4, C3, C4, P3, P4 EEG signals that totally 6 crosslinking electrodes collect separately is subtracted each other with the mean value of the EEG signals collected of four adjacent around separately electrodes, obtain the EEG signals after Laplacian space filtering;

(4) EEG signals through spatial filtering obtained is carried out stage extraction, obtain time domain EEG signals;

(5) EEG signals through spatial filtering obtained is carried out butterworth filter, choosing 5 frequency ranges, is 4Hz-7Hz, 7Hz-10Hz, 10Hz-13Hz, 13Hz-20Hz and 20Hz-30Hz respectively, obtains frequency domain EEG signals, as shown in Figure 4 and Figure 5;

(6) by the described EEG signals through spatial domain-time-frequency domain process, carry out single electrode cospace pattern feature extraction, the proper vector under the two kinds of thinking tasks obtained as shown in Figure 6 and Figure 7.

Described characteristic extracting module is in order to the eigenwert of recognition logic thinking task and thinking in images task.First, the proper vector of the EEG signals in single thinking task after spatial domain-time-frequency domain three dimensions process is extracted according to described single electrode cospace module, these proper vectors are input to the training carrying out sorter in support vector machine (SVM), choose optimum parameter, comprise error, penalty factor etc., the sorter trained as shown in Figure 8, is then carried out the classification of two kinds of thinking tasks, and carries out cross validation by SVM Selecting parameter and accuracy.The sorter of design can well complete the classification of two kinds of thinking tasks, selects 5 subjects to carry out testing the brain-computer interface svm classifier result of the logic-based thinking of the present invention of acquisition and thinking in images as shown in table 1.Because the crosslinking electrode adopted in using is less in the present invention, thus improve rate of information transmission and the convenience of whole brain-computer interface.

Table 1

Crosslinking electrode	Subject 1	Subject 2	Subject 3	Subject 4	Subject 5
						F3	93.9	74.8	69	70	74
F4	84.2	81.3	61	57	69
						C3	89.6	66.2	69	60	63
C4	92.6	73	76	68	78
						P3	61.6	70.5	74	58	67
P4	61.1	78.6	64	61	69

In order to the advantage of outstanding the inventive method, following table 2 lists the experimental data of existing method.

Table 2

Contrast existing method and the inventive method, can find out and use the feature extracting method of single electrode cospace pattern and the tagsort method of support vector machine, optimal classification accuracy can obtain higher lifting, and the poorest classification accuracy rate also has raising by a relatively large margin except first method, prove that method of the present invention can obtain reasonable classification accuracy rate, accuracy is enough to allow subject control brain-computer interface.

Case study on implementation two

The implementation case provides the brain-computer interface control method of a kind of logic-based thinking and thinking in images, due to the experience that the subject participating in brain-computer interface does not have brain-computer interface to control mostly, so whole process comprises training system and control system.

Training system comprises:

(1) training of evoked brain potential signal, first testedly carry out the training of SSVEP task, the time of about 1 minute, observe for the most obvious evoked brain potential with or without direct reaction, prove this tested adaptation brain-computer interface test if responded, next step training test can be carried out;

(2) subject carries out the training of thinking task brain-computer interface, concrete training patterns is as follows, carry out one group of thinking task, allow and be testedly familiar with normal form content, simultaneously in electroencephalogramsignal signal collection equipment, time domain and the frequency-region signal of EEG signals can be observed, signal is processed, can find out tested when carrying out thinking task EEG signals whether normal;

Through the training of above two steps, tested before formally starting in order to make, there is basic understanding to whole task, also in thinking task process, needed the content imagined to accomplish to subject oneself and be familiar with; The object of training is to collect better EEG signals, can eliminate the interference of a lot of noise signal, also can reduce the difficulty to subsequent operation process, obtain better classification accuracy rate and the rate of information throughput.

Control system comprises:

(1) electrical brain stimulation module sends Induced by Stimulation human brain and produces the imagination for logical thinking and thinking in images task, and produces evoked brain potential signal in corresponding region;

(2) gathered EEG signal by electroencephalogramsignal signal acquisition module, detailed process is that subject brings the brain electricity cap meeting head size, is coated with is covered with brain electricity cream at corresponding electrode; And gather EEG signals by these electrodes; The signal collected is inputed to amplifier, and pre-service etc. is carried out to signal;

(3) carry out feature extraction and classifying to EEG signals, detailed process is the signal that will collect, and carries out the signal transacting of spatial domain-time-frequency domain, then the EEG signals processed is carried out single electrode cospace pattern algorithm and calculates, obtain eigenwert; The eigenwert obtained is inputted into support vector machine, carries out the selection and calculation of parameter, obtain optimized parameter, and the classification accuracy rate obtained;

(4) classification results is transferred to external unit, external unit is for telecar, three kinds of states are rotated owing to having calmness, mental arithmetic and imagination space, the three-dimensional motion of object can be controlled, tranquil control is advanced or is retreated, mental arithmetic external unit performs order, and imagination space revolving outer equipment performs order of turning right.

The present invention devises the brain-machine interface method of a kind of logic-based thinking and thinking in images, to realize a kind of applied research of novel brain-computer interface technology; Realize being optimized CSP algorithm, have employed single electrode CSP.This system has high accuracy, the high rate of information throughput, and acquisition electrode is few, property easy to use.Compared to the brain-computer interface of Mental imagery, the brain-computer interface of logic-based thinking and thinking in images reduces operation easier, simultaneously compared to the brain-computer interface of SSVEP and P300 evoked brain potential signal, reduce the vision burden under that task, improve the applicable crowd of brain-computer interface.Further paractical research is carried out to this brain-computer interface, feature extraction and sorting algorithm are optimized and research, obtain the brain machine interface system that Performance comparision is perfect, be applied in practicality, and reasonable social benefit can be brought.

The above is only the preferred embodiment of the present invention; be noted that for those skilled in the art; under the premise without departing from the principles of the invention, can also make some improvements and modifications, these improvements and modifications also should be considered as protection scope of the present invention.

Claims (8)

1. the brain-machine interface method of logic-based thinking and thinking in images, is characterized in that: comprise the following steps that order performs:

Step one, electroencephalogramsignal signal acquisition module is set at subject's head, by the normal form of logical thinking task and thinking in images task, utilizes brain wave acquisition method of singly leading to collect the original EEG signals of subject under different stimulated state;

Step 2, pre-service is carried out to original EEG signals, then carry out the three-dimensional process of space, time and frequency;

Step 3, utilize single electrode cospace pattern algorithm to after step 2 process EEG signals extract proper vector;

Step 4, by support vector machine to extracting the proper vector cross validation classification obtained in step 3, obtain eeg signal classification result and export.

2. the brain-machine interface method of logic-based thinking according to claim 1 and thinking in images, it is characterized in that: when gathering original EEG signals, the crosslinking electrode comprised as upper/lower positions is set in electroencephalogramsignal signal acquisition module: on prefrontal lobe, top and temporal lobe, all arrange a pair crosslinking electrode, and paired two crosslinking electrode positional symmetry are arranged, specifically comprise F3 and F4 laying respectively at prefrontal lobe both sides, lay respectively at C3 and C4 of top both sides and lay respectively at P3 and P4 of temporal lobe both sides.

3. the brain-machine interface method of logic-based thinking according to claim 1 and thinking in images, it is characterized in that: the normal form design of single experiment is as follows: first subject keeps the tranquility at least 2 seconds, and EEG signals corresponding under gathering this tranquility as a reference; Then logical thinking task or thinking in images task is carried out, wherein logical thinking task is for occur that in screen formula, subject carry out mental arithmetic according to formula, thinking in images task for there is a shape in screen, subject imagines that this shape rotates at three dimensions.

4. the brain-machine interface method of logic-based thinking according to claim 2 and thinking in images, it is characterized in that: in step one, brain wave acquisition method of singly leading is adopted to gather the original EEG signals of these 6 crosslinking electrodes of F3, F4, C3, C4, P3, P4, reference electrode is arranged on left side mastoid location, and the earth signal that leads is arranged on forehead place.

5. the brain-machine interface method of logic-based thinking according to claim 2 and thinking in images, it is characterized in that: the original EEG signals obtained in step one is carried out pre-service in the following manner: gather electro-ocular signal by paired two crosslinking electrodes, to original EEG signals through removing above-mentioned electro-ocular signal, Muscle artifacts, removal baseline wander, data sectional process, and then obtain pretreated EEG signals.

6. the brain-machine interface method of logic-based thinking according to claim 2 and thinking in images, it is characterized in that: the detailed process of step 2 is as follows: for single experiment, EEG signals after pre-service being completed carries out Laplacian space filtering, obtains and eliminates spatially to the EEG signals after the interference of EEG signals; Then will be equally divided into 4 sections through the filtered EEG signals of Laplacian space on the one hand, form time domain EEG signals; Also will carry out Short Time Fourier Transform to domain space through the filtered EEG signals of Laplacian space simultaneously, extract most active 5 frequency bands of thinking activities, form frequency domain EEG signals.

7. the brain-machine interface method of logic-based thinking according to claim 6 and thinking in images, it is characterized in that: in step 3, use single electrode cospace pattern algorithm to the frequency domain EEG signals of the single electrode drawn in step 2 and time domain EEG signals form 20 dimensions proper vector extract.

8. the brain-machine interface method of logic-based thinking according to claim 7 and thinking in images, it is characterized in that: in step 4, proper vector inputted in the support vector machine trained, classification also cross validation obtains the classification results of the EEG signals that each electrode collects.