CN104173124A - Upper limb rehabilitation system based on biological signals - Google Patents

Abstract

The invention discloses an upper limb rehabilitation system based on biological signals. The upper limb rehabilitation system based on the biological signals is characterized in that a brain electric cap is attached to the surface of a brain, brain electric signals can be sensed through the brain electric cap, features of the brain electric signals are extracted and classified sequentially through a CSP feature extraction algorithm and a classifier with an adaptive LDA classification algorithm after amplification, filtering and noise reduction are performed on the brain electric signals, and then the brain electric signals are translated into drive instructions used to control drive motor equipment and pneumatic tendon auxiliary equipment, a system central processing unit fuses and sorts the drive instructions and motion state data collected by a motion state collector, and then feeds the drive instructions and the motion state data, which are fused and sorted, back to the drive motor equipment, the pneumatic tendon auxiliary equipment and a mobile terminal, and therefore a mechanical arm is driven to perform rehabilitation exercise. Work personnel remotely monitor motion and rehabilitation states of patients through the mobile terminal, and the patients also can invoke rehabilitation games in a display platform through a voice input and output device so as to perform auxiliary rehabilitation training. Accordingly, the upper limb rehabilitation system based on the biological signals improves training effectiveness of the patients, and simultaneously guarantees safety and stability in the training.

Description

A kind of upper limb healing system based on bio signal

Technical field

The invention belongs to control with computational intelligence technical field, more specifically say, relate to a kind of upper limb healing system based on bio signal.

Background technology

In modern society, brain diseasess such as apoplexy, hemiplegia and the upper extremity exercise obstacle that causes has brought a lot of difficulties to the life of a lot of middle-aged and elderly people.They need the rehabilitation training of science can help them to recover extremity motor function, and under this background, the upper limb healing system based on bio signal has caused people's concern as a kind of efficient apparatus of rehabilitation training of upper limbs.

Total about 1,000 hundred million neurocytes in human brain, wherein cerebral cortex has 14,000,000,000 cells, and each neurocyte has 10000 neural connections simultaneously, and they have formed and have sent out assorted huge nerve cell network.Neurocyte is mainly comprised of dendron, aixs cylinder and cyton three parts.In neutral net, information exchange is crossed nerve conflict and is conducted through the outstanding orientation of carrying out, and can cause that inner ion concentration changes in this process, and postsynaptic membrane changes, thereby produce faint electric current, is brain electricity physiological signal.

The physiological signal that EEG signals is wanted as body weight for humans.According to research, show, the main feature of EEG signals is:

1), randomness and stationarity are quite strong

Human brain is to remain a quite complicated system at scientific level up to now, in brain, each region connects each other and is restricting with part, therefore make EEG signals produce the various variations of not grasped yet so far, the physiologic factor that also can be regarded as formation EEG signals is changing all the time.

2), EEG signals has non-linear

Because human brain is the nonlinear system of a structure and function complexity, therefore the EEG signals of its generation also has non-linear feature, it is all very inconvenient that this analyzes with processing for us, and traditional signal processing method based on linear system is substantially no longer applicable to process EEG signals.

3), the amplitude of EEG signals very a little less than

The amplitude of EEG signals is 5 μ V～100 μ V, is generally 50 μ V; Frequency range is 0.5Hz～35Hz, and internal resistance is from tens kilo-ohms to hundreds of kilo-ohm not grade and easily variation, and signal to noise ratio is low, reaches as high as 1:105.

4), the rhythmicity of brain wave

Brain wave is the complicated complex wave of a kind of frequency range between 0.5～35Hz.According to its frequency, can be divided into Delta (δ, 0.5～3Hz), Theta (θ, 4～8Hz), Alpha (α, 8～13Hz), Beta (β, 14～30Hz).

Therefore, it is particularly important that the extraction of EEG signals just becomes, at present, the extraction of EEG signals adopts brain electricity cap to be attached to brain surface's mode conventionally, and as shown in Figure 1, brain electricity cap is 10-20 international standard system electrode placement methods, wherein the first width is left view, and the second width is top view.In accordance with international practices, odd electrode is positioned over brain left hemisphere, and even number motor is positioned over right hemisphere.

When people's brain sends motion intent instructions, corresponding brain region there will be decorrelation synchronism, the signal waveform energy that is embodied as the collection of EEG signals interface has larger change, by the EEG signals to zones of different, monitors, and can analyze the motion intention that draws patient.

The electrode that judgement is observed for bimanual movements intention mainly distributes around C3 and C4, and ground electrode is left ear-lobe A1.Frequently-used data sample frequency is 250Hz, and sample devices is the electrode cap of 10-15 passage.

Yet traditional rehabilitation system often cannot be by the rehabilitation training of patient's complete independently, and in rehabilitation exercise process, kinestate cannot intuitively show and remote monitoring, and can not be in real time by feedback patient's motor feedback to rehabilitation system, realize the trim process of rehabilitation system; Secondly, existing most of rehabilitation exercise equipment, its operation is often uninteresting, easily make patient's heart irritated, and the power frequency that is easily subject to other electronic devices and components on rehabilitation system is disturbed, and makes the collection of EEG signals inaccurate, is difficult to carry out effective rehabilitation training.

Summary of the invention

The object of the invention is to overcome the deficiencies in the prior art, a kind of upper limb healing system based on bio signal is provided, by brain electricity cap, extract patient's EEG signals, analyze patient's motion intention, thereby driving device arm is assisted patient moving, improved the effectiveness of patient's training, the safety while simultaneously having guaranteed training.

For achieving the above object, a kind of upper limb healing system based on bio signal of the present invention, is characterized in that, comprising:

One EEG signals extraction equipment is a brain electricity cap, brain electricity cap is attached to brain surface, by the metal electrode induction EEG signals of brain electricity cap, the built-in chip of brain electricity cap amplifies EEG signals again successively, filtering and denoising, by CSP feature extraction algorithm, the EEG signals after denoising is converted into again to the digital signal of high identification, digital signal is classified by having the grader of self adaptation LDA sorting algorithm again, complete feature identification, digital signal after identification is converted into the driving instruction of controlling drive motors equipment and pneumatic muscle auxiliary facilities, finally by the wireless device in brain electricity cap, send,

One system central processor, comprise data processor and wireless transmitter, be arranged on the top of back bracket, for receiving, the kinestate data that gather in the driving instruction that storage brain electricity cap sends and kinestate harvester, wherein, data processor carries out data parsing to the driving instruction and the kinestate data that receive according to default communication protocol, identify valid data value, again part relevant to kinestate data in valid data value is encoded, finally the data after coding are encapsulated and feed back to drive motors equipment and pneumatic muscle auxiliary facilities as feedback signal, the kinestate data of storage are sent to mobile terminal by wireless transmitter simultaneously,

One back bracket, is attached to the back of human body, and its top is connected with carbon fiber arm ectoskeleton, and flushes with the shoulder of human body, for bearing the weight of rehabilitation equipment, and hangs the system central processor at back;

One carbon fiber arm ectoskeleton, comprise upper arm and forearm, upper arm and forearm are two carbon fiber boards that fit tightly and be screwed, junction, joint at upper arm and back bracket, upper arm and forearm connects to form mechanical arm with screw respectively, and on upper arm and forearm, all there is equally spaced screw hole, can make position, the wrong hole of carbon fiber board install, for regulating mechanical arm length;

Two drive motors equipment, every group of drive motors equipment comprises a drive motors and one drive circuit plate, to be installed on respectively the ectoskeletal top arm's tip of carbon fiber arm, and the junction, joint of upper arm and forearm, drive circuit board is for gathering the driving frequency of drive motors, and drive motors is for the motion of driving device arm;

Two pneumatic muscle auxiliary drive apparatus, are installed on respectively upper arm and back bracket, upper arm and junction, forearm joint, for cutting down mechanical shaking and driving force compensation;

One motor encoder, be installed on drive motors tail end, and connect drive circuit board and kinestate harvester by data line, by the monitoring to drive motors kinestate, obtain the kinestate data of mechanical arm, and these kinestate data are sent to kinestate harvester;

One electromyographic signal collection electrode, comprises and electrode signal acquisition and the data line of an adhesive type electrode signal acquisition is sticked on to patient's arm, for gathering patient's electromyographic signal, and transfers to kinestate harvester;

One heart rate gathers elastic wristband, comprises built-in pulse collection electrode and data line, is worn on patient's wrist, for gathering patient's heart rate, and transfers to kinestate harvester;

One kinestate harvester is surface-mounted integrated circuit, be arranged on the outside of mechanical arm, on surface-mounted integrated circuit, be reserved with and connect the data line interface that motor encoder, electromyographic signal collection electrode and heart rate gather wrist strap, after connecting by data line, from motor encoder electromyographic signal collection electrode and heart rate, gather elastic wristband reception data, the data that receive are encapsulated, and be sent to system central processor as current kinestate data;

One voice command receiving equipment is a phonetic incepting microphone, is installed on forearm tail end, and also detachable use separately, for input speech signal;

One voice message equipment, comprises a speech processing module, is arranged on display screen back; The voice signal of input is converted into serial ports command signal by speech processing module, and for controlling the display interface menu of display platform, display platform sends feedback signal by voice message equipment again after receiving serial ports command signal;

One display platform, is installed on rehabilitation system the place ahead, and its display interface menu is controlled by phonetic order, is built-in with bluetooth module and rehabilitation training game; Bluetooth module can directly read EEG signals from potentials extraction equipment, and the auxiliary patient of rehabilitation training game carries out rehabilitation training;

Brain electricity cap is attached to brain surface, by the metal electrode induction EEG signals of brain electricity cap, the built-in chip of brain electricity cap to EEG signals amplify successively, filtering and denoising, by CSP feature extraction algorithm, the EEG signals after denoising is converted into again to the digital signal of high identification, digital signal is classified by having the grader of self adaptation LDA sorting algorithm again, complete feature identification, the digital signal after identification is converted into the driving instruction of controlling drive motors equipment and pneumatic muscle auxiliary facilities;

Drive instruction to be transferred to system central processor by the mode of wireless transmission, the kinestate data that data processor in system central processor gathers in conjunction with kinestate harvester again, according to default communication protocol, carry out data parsing, identify valid data value, again part relevant to kinestate data in valid data value is encoded, finally the data after coding are encapsulated and feed back to drive motors equipment and pneumatic muscle auxiliary facilities as feedback signal, thereby driving device arm is done rehabilitation exercise according to feedback signal;

Simultaneity factor central processing unit is sent to mobile terminal by kinestate data by inner wireless transmitter, medical personnel can pass through kinestate and the rehabilitation state of mobile terminal remote monitored patient, the rehabilitation training that patient also can call in display platform by voice command receiving equipment is played, and carries out auxiliary rehabilitation exercise.

Wherein, the method that in described brain electricity cap, built-in chip carries out denoising to EEG signals is: by default wavelet algorithm, EEG signals is carried out to denoising, idiographic flow is:

At function space L ²(R), in, establish wavelet function C _Ψmeet:

$C_{\mathrm{\&Psi;}}={\&Integral;}_{R^{+}}\frac{\left|\widehat{\mathrm{\&Psi;}}\left(x\right)\right|}{\left|w\right|}\mathrm{dw}<\&infin;$

Wherein, R is all real number fields, and w is constant and w ≠ 0, and Ψ (x) is called wavelet mother function, fourier transformation for Ψ (x); For convenience of narration, make wavelet mother function Ψ (x) be:

$\mathrm{\&Psi;}(a,b)\left(x\right)=\frac{1}{\sqrt{\left|a\right|}}\mathrm{\&Psi;}\left(\frac{x-b}{a}\right)$

Wherein, a and b are contraction-expansion factor and shift factor, control respectively stretching and the translation transformation of the female ripple of small echo;

The EEG signals f that is L by length (x) carries out continuous wavelet transform:

$W_f(a,b)=\frac{1}{\sqrt{a}}{\&Integral;}_{R}f\left(x\right)\mathrm{\&Psi;}\left(\frac{x-b}{a}\right)\mathrm{dx}$

The wavelet transformation of this EEG signals f (x) carries out inverse transformation:

$f{\left(x\right)}^{\&prime;}=C_{\mathrm{\&Psi;}}{\&Integral;\&Integral;}_{R\&times;R^{+}}{W}_f(a,b)\mathrm{\&Psi;}\left(\frac{x-b}{a}\right)\mathrm{dadb}$

Wherein, R * R ⁺represent that this double integral is the integration in real number field;

According to the convolution property of small echo and down-sampling characteristic, f (x) ' can only be decomposed into log at most ₂l layer,, when ground floor decomposes, makes convolution operation with high frequency coefficient and the low frequency coefficient of small echo signal respectively by f (x) ', more respectively convolution signal is later done to down-sampling, obtains low frequency scale coefficient and high frequency detail coefficients; From the second layer decomposes, last layer is decomposed to the low frequency scale coefficient obtaining and make convolution and down-sampling with high frequency coefficient and the low frequency coefficient of small echo signal respectively, obtain low frequency scale coefficient and high frequency detail coefficients after this layer of decomposition, to the last one deck log ₂l, finally from log ₂l layer starts, and low frequency scale coefficient and high frequency detail coefficients are carried out doing contrary convolution with low frequency coefficient and the high frequency coefficient of small echo signal respectively after unit time delay, obtains the low frequency scale coefficient of last layer, the like until ground floor obtains the EEG signals after denoising.

As improvement of the present invention, sampling CSP feature extraction algorithm extracts EEG signals, and its concrete grammar is:

K.1, calculate the covariance matrix of EEG signals

EEG signals after denoising is divided into right-hand man's two class EEG signals, i.e. X, Y, what every class EEG signals was all used C * N is that real number matrix represents, and wherein, C is signalling channel number, and N is sampled point, and the covariance matrix of X and Y can be expressed as:

${\mathrm{\&Sigma;}}_X=\frac{{\mathrm{XX}}^T}{\mathrm{trace}\left({\mathrm{XX}}^T\right)},{\mathrm{\&Sigma;}}_Y=\frac{{\mathrm{YY}}^T}{\mathrm{trace}\left({\mathrm{YY}}^T\right)}$

Wherein, ∑ _xwith ∑ _ybe respectively the covariance matrix for X and Y, [] ^tfor the transposition of [], trace (XX ^t) and trace (YY ^t) be respectively the diagonal element sum of X and Y;

K.2, structure spatial filter

K.2.1 determine the existence condition of spatial filter

Construct a spatial filter W, while making CSP feature extraction algorithm carry out feature extraction, meet:

max∑var(W ^T·X)，s.t.∑var[W ^T·(X+Y)]＝1

Wherein, var () is variance computing function; Variance computing function to above formula launches, and the existence condition that obtains spatial filter W is:

max∑W ^T·Σ _X·W，s.t.∑W ^T·(Σ _X+Σ _Y)·W＝1

K.2.2, according to the existence condition of spatial filter, construct spatial filter

Solution matrix P, and P meets P (∑ _x+ ∑ _y) P ^t=I, wherein, I is unit matrix;

According to the definition of covariance matrix: covariance matrix is for being symmetrical matrix, symmetrical matrix and be still symmetrical matrix, by symmetrical matrix (∑ _x+ ∑ _y) carry out singular value decomposition and obtain (∑ _x+ ∑ _y)=U φ U ^t, wherein, U is (∑ _x+ ∑ _y) eigenvectors matrix, φ is (∑ _x+ ∑ _y) eigenvalue diagonal matrix, order obtain and meet P (Σ _x+ Σ _y) P ^tthe matrix P of=I;

Order solve orthogonal matrix R and diagonal matrix D, R and D are met ${\widehat{\mathrm{\&Sigma;}}}_X={\mathrm{RDR}}^T;$

Right carry out singular value decomposition, obtain positive definite symmetric matrices will the orthogonal matrix R that forms of characteristic vector, will eigenvalue as diagonal element, remaining element is made as to 0, construct diagonal matrix D;

According to ${\widehat{\mathrm{\&Sigma;}}}_X+{\widehat{\mathrm{\&Sigma;}}}_Y=I,{\widehat{\mathrm{\&Sigma;}}}_Y=R\&CenterDot;(I-D)\&CenterDot;R^T,$ Construct spatial filter W=R ^tp;

K.3, designing filter group W _sub

Select m column vector of eigenvalue maximum in diagonal matrix D and m column vector of eigenvalue minimum, their corresponding 2m column vectors in spatial filter W are formed to new bank of filters W _sub;

K.4, the EEG signals after denoising is carried out to feature extraction, obtain the digital signal F (x) of high identification

By the EEG signals of any time, by spatial filter W Filtering Processing, C for the EEG signals after processing * L real number matrix S represents, real number matrix S is split as K submatrix again, i.e. S=[S (1), and S (2) ..., S (K)] ^t, each submatrix represents with the real number matrix S () of C * N, and this K submatrix is passed through to bank of filters W successively _subprocessing obtains the matrix of K 2m * N again according to formula wherein, representing matrix S _cSPin the value of the capable j row of the i of l submatrix, l=1,2 ..., K, i=1,2 ..., 2m, j=1,2 ..., N, obtains each eigenvalue f _l, finally by eigenvalue f _lcombination obtains characteristic vector x (l)=[f _l1, f _l2..., f _l2m] ^t, real number matrix S obtains high identification digital signal after feature extraction so $F\left(x\right)=[x\left(1\right),x\left(2\right),\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;x\left(K\right)]=\left[\begin{array}{c}{f}_{11},f_{12},\&CenterDot;\&CenterDot;\&CenterDot;,f_{1K}\\ f_{21},f_{22},\&CenterDot;\&CenterDot;\&CenterDot;,f_{2K}\\ \&CenterDot;\\ \&CenterDot;\\ \&CenterDot;\\ f_{l1},f_{l2},\&CenterDot;\&CenterDot;\&CenterDot;,f_{\mathrm{lK}}\end{array}\right].$

Further improvement of the present invention is, sampling has the grader of self adaptation LDA sorting algorithm classifies to EEG signals, and concrete grammar is:

T.1), adopt slip window sampling to carry out Estimation of Mean to EEG signals

Length be the EEG signals of L by feature extraction after, all F (x) are designated as to set { x (j) }, its Estimation of Mean expression formula is:

$\mathrm{\&mu;}=\frac{1}{L}\underset{j=1}{\overset{L}{\mathrm{\&Sigma;}}}x\left(j\right)$

One time slide window is set, will gathers { x (j) } and constantly carry out Estimation of Mean at t, its expression formula is:

$\mathrm{\&mu;}\left(t\right)=\frac{1}{\underset{j=0}{\overset{h-1}{\mathrm{\&Sigma;}}}{w}_j}\underset{j=0}{\overset{h-1}{\mathrm{\&Sigma;}}}{w}_j\&CenterDot;x(t-j)$

Wherein, h represents sliding window width, w _jrepresent window weight coefficient, x (t-j) represents (t-j) EEG signals constantly;

Calculate t-1 Estimation of Mean constantly, by t-1 Estimation of Mean constantly, obtain t Estimation of Mean μ (t) constantly:

μ(t)＝(1-UC)·μ(t-1)+UC·x(t)

Wherein, wherein x (t) is t input EEG signals constantly, and UC represents to upgrade coefficient;

T.2), estimate the covariance of EEG signals

Length be the EEG signals of L by feature extraction after, all F (x) are designated as to set { x (j) }, the account form of its covariance matrix Σ is:

$\mathrm{\&Sigma;}=\frac{1}{L}\underset{j=1}{\overset{L}{\mathrm{\&Sigma;}}}[x\left(j\right)-\mathrm{\&mu;}]{[x\left(j\right)-\mathrm{\&mu;}]}^T$

Set { x (j) } in the estimated value of t covariance matrix Σ (t) is constantly so:

Σ(t)＝(1-UC)·Σ(t-1)+UC·[1,x(t)] ^T·[1,x(t)]

To t covariance Σ (t) the constantly conversion of inverting:

$\mathrm{\&Sigma;}{\left(t\right)}^{-1}=\frac{1}{1-\mathrm{UC}}\&CenterDot;[\mathrm{\&Sigma;}{(t-1)}^{-1}-\frac{1}{\frac{1-\mathrm{UC}}{\mathrm{UC}}+x{\left(t\right)}^T\&CenterDot;v\left(t\right)}v\left(t\right)\&CenterDot;v{\left(t\right)}^T]$

Wherein, ν (t)=Σ (t-1) ^-1x (t);

T.3), Estimation of Mean and covariance are carried out to online adaptive estimation, complete reappraising of linear classifier parameter

c _t＝F{D[x(t)]}＝F[b(t-1)+w(t-1) ^T·x(t)]

${\mathrm{\&mu;}}_{c_t}\left(t\right)=(1-\mathrm{UC})\&CenterDot;{\mathrm{\&mu;}}_{c_t}(t-1)+\mathrm{UC}\&CenterDot;x\left(t\right)$

w(t)＝∑(t) ^-1·[μ ₂(t)-μ ₁(t)]

$b\left(t\right)=-w{\left(t\right)}^T\&CenterDot;\frac{1}{2}\&CenterDot;[{\mathrm{\&mu;}}_1\left(t\right)+{\mathrm{\&mu;}}_2\left(t\right)]$

Wherein, c _trepresent classification labelling c _t∈ { 1,2}; F{D[x (t)] } represent the classification discriminant function of EEG signals x (t), as D[x (t)] and during > 0, c _tvalue is 2, as D[x (t)] during < 0, c _tvalue is that 1, w (t) is the t slope of linear classifier constantly, and b (t) is the t intercept of linear classifier constantly.

Goal of the invention of the present invention is achieved in that

The present invention is based on the upper limb healing system of bio signal, brain electricity cap is attached to brain surface, by the metal electrode induction EEG signals of brain electricity cap, EEG signals is amplified, filtering and denoising, by CSP feature extraction algorithm and the grader with self adaptation LDA sorting algorithm, EEG signals is carried out to feature extraction and classification successively again, complete feature identification, thereby be converted into the driving instruction of controlling drive motors equipment and pneumatic muscle auxiliary facilities, system central processor is again in conjunction with the current kinestate data that drive instruction and kinestate harvester to gather, encode, encapsulation feeds back to drive motors equipment and pneumatic muscle auxiliary facilities rear number, thereby driving device arm is done rehabilitation exercise according to feedback signal, simultaneity factor central processing unit is sent to mobile terminal by kinestate data, staff is by kinestate and the rehabilitation state of mobile terminal remote monitor patients, the rehabilitation training that patient also can call in display platform by phonetic entry, outut device is played, and carries out auxiliary rehabilitation exercise.Improved like this effectiveness of patient's training, the safety and stability while simultaneously having guaranteed training.

Meanwhile, the upper limb healing system that the present invention is based on bio signal also has following beneficial effect:

(1), the present invention is attached to the mode of patient's head by brain electricity cap, sports consciousness that can Real-Time Monitoring patient, has improved the subjective initiative of Rehabilitation training;

(2), the present invention makes improvements brain electricity cap, increased CSP feature extraction algorithm and have the grader of self adaptation LDA sorting algorithm in brain electricity cap, more accurate while making like this extraction of EEG signals;

(3), in the present invention, another improvement is to have increased electromyographic signal collection electrode, heart rate collection elastic wristband and kinestate harvester, electromyographic signal and heart rate that like this can Real-Time Monitoring patient, by kinestate harvester, the data of monitoring are carried out Real-time Collection and fed back to system central processor again, it is whole that system central processor gathers automatic fine tuning in conjunction with the EEG signals of feedback data and extraction to rehabilitation training joint, guaranteed like this effectiveness, stability and the safety of training;

(4), also increased remote control function in the present invention, by the mode of Long-distance Control, medical personnel can pass through kinestate and the rehabilitation state of mobile terminal remote monitored patient, have further increased the safety of training;

(5), further improvement of the present invention is that the rehabilitation training that patient can call in display platform by voice command input equipment is played, carry out auxiliary rehabilitation exercise, thereby improved the effectiveness of rehabilitation training.

Accompanying drawing explanation

Fig. 1 is the distribution of electrodes schematic diagram of brain electricity cap;

Fig. 2 is a kind of specific embodiment structure chart of upper limb healing system that the present invention is based on bio signal;

Fig. 3 is the schematic diagram that the present invention is based on the upper limb healing system of bio signal;

Fig. 4 is the spectrogram of ERD/ERS phenomenon;

Fig. 5 is the schematic diagram that EEG signals is carried out wavelet decomposition;

Fig. 6 is that dbN small echo carries out the spectrogram of denoising to filtered EEG signals;

Fig. 7 is that Haar small echo carries out the spectrogram of denoising to filtered EEG signals.

The specific embodiment

Below in conjunction with accompanying drawing, the specific embodiment of the present invention is described, so that those skilled in the art understands the present invention better.Requiring particular attention is that, in the following description, when perhaps the detailed description of known function and design can desalinate main contents of the present invention, these are described in here and will be left in the basket.

Embodiment

Fig. 2 is a kind of specific embodiment structure chart of upper limb healing system that the present invention is based on bio signal.

In the present embodiment, as shown in Figure 2, a kind of upper limb healing system based on bio signal of the present invention, comprising:

One EEG signals extraction equipment 1 is a brain electricity cap, brain electricity cap is attached to brain surface, by the metal electrode induction EEG signals of brain electricity cap, the built-in chip of brain electricity cap amplifies EEG signals again successively, filtering and denoising, by CSP feature extraction algorithm, the EEG signals after denoising is converted into again to the digital signal of high identification, digital signal is classified by having the grader of self adaptation LDA sorting algorithm again, complete feature identification, digital signal after identification is converted into the driving instruction of controlling drive motors equipment and pneumatic muscle auxiliary facilities, finally by the wireless device in brain electricity cap, send,

One system central processor 2, comprise data processor and wireless transmitter, be arranged on the top of back bracket 3, for receiving, the kinestate data that gather in the driving instruction that storage brain electricity cap sends and kinestate harvester 6, wherein, data processor carries out data parsing to the driving instruction and the kinestate data that receive according to default communication protocol, identify valid data value, again part relevant to kinestate data in valid data value is encoded, finally the data after coding are encapsulated and feed back to drive motors equipment 4 and pneumatic muscle auxiliary facilities 7 as feedback signal, the kinestate data of storage are sent to mobile terminal by wireless transmitter simultaneously,

One back bracket 3, is attached to the back of human body, and its top is connected with carbon fiber arm ectoskeleton 5, and flushes with the shoulder of human body, for bearing the weight of rehabilitation equipment and hanging the system central processor 2 at back;

One carbon fiber arm ectoskeleton 5, comprise upper arm and forearm, upper arm and forearm are two carbon fiber boards that fit tightly and be screwed, junction, joint at upper arm and back bracket 3, upper arm and forearm connects to form mechanical arm with screw respectively, and on upper arm and forearm, all there is equally spaced screw hole, can make position, the wrong hole of carbon fiber board install, for regulating mechanical arm length;

Two drive motors equipment 4, every group of drive motors equipment comprises a drive motors and one drive circuit plate, the top arm's tip of carbon fiber arm ectoskeleton 5 will be installed on respectively, and the junction, joint of upper arm and forearm, drive circuit board is for gathering the driving frequency of drive motors, and drive motors is for the motion of driving device arm;

Two pneumatic muscle auxiliary drive apparatus 7, are installed on respectively the junction, joint of upper arm and back bracket 3, upper arm and forearm, for cutting down mechanical shaking and driving force compensation;

One motor encoder 8, be installed on drive motors tail end, and connect drive circuit board and kinestate harvester 6 by data line, by the monitoring to drive motors kinestate, obtain the kinestate data of mechanical arm, and these kinestate data are sent to kinestate harvester 6;

One electromyographic signal collection electrode 9, comprises and electrode signal acquisition and the data line of an adhesive type electrode signal acquisition is sticked on to patient's arm, for gathering patient's electromyographic signal, and transfers to kinestate harvester 6;

One heart rate gathers elastic wristband 10, comprises built-in pulse collection electrode and data line, is worn on patient's wrist, for gathering patient's heart rate, and transfers to kinestate harvester 6;

One kinestate harvester 6 is surface-mounted integrated circuit, be arranged on the outside of mechanical arm, on surface-mounted integrated circuit, be reserved with and connect the data line interface that motor encoder 8, electromyographic signal collection electrode 9 and heart rate gather wrist strap 10, after connecting by data line, from motor encoder 8, electromyographic signal collection electrode 9 and heart rate, gather elastic wristband 10 and receive data, the data of reception are encapsulated, and be sent to system central processor 2 as current kinestate data;

One voice command receiving equipment 11 is a phonetic incepting microphone, is installed on forearm tail end, and also detachable use separately, for input speech signal;

One voice message equipment 12, comprises a speech processing module, is arranged on display screen back; The voice signal of input is converted into serial ports command signal by speech processing module, and for controlling the display interface menu of display platform 13, display platform 13 sends feedback signal by voice message equipment 12 again after receiving serial ports command signal;

One display platform 13, is installed on rehabilitation system the place ahead, and its display interface menu is controlled by phonetic order, is built-in with bluetooth module and rehabilitation training game; Bluetooth module can directly read EEG signals from potentials extraction equipment, and the auxiliary patient of rehabilitation training game carries out rehabilitation training;

Fig. 3 is the schematic diagram that the present invention is based on the upper limb healing system of bio signal.

In the present embodiment, rehabilitation system is attached to patient's health by binder, drives the work of rehabilitation system by mechanically operated mode, thereby drive patient to carry out rehabilitation training, its idiographic flow is as shown in Figure 3:

S1, collection EEG signals

The brain electricity cap with EEG signals sensor is attached to brain surface, by the metal electrode induction EEG signals of brain electricity cap, in the present embodiment, adopt the CURRY Scan NuAmps Express brain electricity cap of Neuroscan company to carry out EEG signals extraction to patient;

When the brain of human body carries out the different thinking activities relevant to motion, the frequecy characteristic of EEG signals EEG (Electroencephalogram) has the different forms of expression.When human body keeps clear-headed, attention is dispersed, and brain is while imagining limb motion, and the elementary sensorimotor cortex of brain can show obvious EEG activity in alpha rhythm (8～12Hz) and beta response (18～26Hz) frequency range.And when people carries out a certain specific limb motion imagination, the energy of the EEG signal that can make specific region in brain within the scope of alpha rhythm and beta response reduces, the phenomenon that desynchronizes ERD (event ralateddesynchronization).Contrary, after brain stops corresponding motion and imagines and loosen, the energy of alpha rhythm and beta response can recover again, i.e. synchronism ERS (event relatedsynchronization);

As shown in Figure 4, it is the most obvious that ERD/ERS phenomenon shows in alpha rhythm, as shown in Fig. 4 (b), in beta response slightly a little less than, as Fig. 4 (a), on two width figure, all can find out in some period, signal energy is lower, even region zero, and some period energy is very high, and this has represented the embodiment of ERD/ERS phenomenon in this frequency range.But signal is totally comparatively mixed and disorderly, cannot be directly used in the judgement of motion intention, signal could need to be used through pretreatment;

In step S2～S6, EEG signals is specifically processed below:

S2, EEG signals are amplified

Hardware power amplifier by built-in chip in brain electricity cap, amplifies EEG signals;

S3, EEG signals filtering

EEG signals after amplifying is input to successively to 50Hz wave trap and 3Hz～40Hz band filter of built-in chip in brain electricity cap, the 50Hz alternating current of filtering EEG signals, low frequency clutter and high frequency supurious wave;

S4, EEG signals denoising

Filtered EEG signals is carried out denoising by the default wavelet algorithm of built-in chip in brain electricity cap, and idiographic flow is:

At function space L ²(R), in, establish wavelet function C _Ψmeet:

$C_{\mathrm{\&Psi;}}={\&Integral;}_{R^{+}}\frac{\left|\widehat{\mathrm{\&Psi;}}\left(x\right)\right|}{\left|w\right|}\mathrm{dw}<\&infin;$

Wherein, R is all real number fields, and w is constant and w ≠ 0, and Ψ (x) is called wavelet mother function, fourier transformation for Ψ (x); For convenience of narration, make wavelet mother function Ψ (x) be:

$\mathrm{\&Psi;}(a,b)\left(x\right)=\frac{1}{\sqrt{\left|a\right|}}\mathrm{\&Psi;}\left(\frac{x-b}{a}\right)$

Wherein, a and b are contraction-expansion factor and shift factor, control respectively stretching and the translation transformation of the female ripple of small echo; The continuous wavelet transform form of EEG signals f (x) is expressed as:

$W_f(a,b)=\frac{1}{\sqrt{a}}{\&Integral;}_{R}f\left(x\right)\mathrm{\&Psi;}\left(\frac{x-b}{a}\right)\mathrm{dx}$

The wavelet transformation of this EEG signals f (x) carries out inverse transformation:

$f{\left(x\right)}^{\&prime;}=C_{\mathrm{\&Psi;}}{\&Integral;\&Integral;}_{R\&times;R^{+}}{W}_f(a,b)\mathrm{\&Psi;}\left(\frac{x-b}{a}\right)\mathrm{dadb}$

Wherein, R * R ⁺represent that this double integral is the integration in real number field;

Wavelet function has polytype, and conventional have a Haar small echo, dbN small echo, and symN small echos etc., when concrete application, select according to standards such as actual support length, symmetry and regularities.

According to the convolution property of small echo and down-sampling characteristic, f (x) ' can only be decomposed into log at most ₂l layer, in the present embodiment, as shown in Figure 5, f (x) ' is divided into 3 layers, when ground floor decomposes, f (x) ' is made to convolution operation with high frequency coefficient and the low frequency coefficient of small echo signal respectively, more respectively convolution signal is later done to down-sampling, obtain low frequency scale coefficient cA1 and the high frequency detail coefficients cD1 of ground floor;

Again ground floor is decomposed to the scale coefficient cA1 obtaining and make convolution operation with high frequency coefficient and the low frequency coefficient of small echo signal respectively, more respectively convolution signal is later done to down-sampling, obtain low frequency scale coefficient cA2 and the high frequency detail coefficients cD2 of the second layer; In like manner obtain low frequency scale coefficient cA3 and the high frequency detail coefficients cD3 of the 3rd layer;

During signal process down-sampling, its length of every once sampling is original half, therefore, may make the dimension of signal too low after some layers of decomposition, so in order to obtain the constant denoised signal of length, conventionally can carry out signal reconstruction by unit time delay and the method for contrary convolution;

As shown in Figure 5, since the 3rd layer, low frequency scale coefficient cA3 and high frequency detail coefficients cD3 are carried out doing contrary convolution with low frequency coefficient and the high frequency coefficient of small echo signal respectively after unit time delay, obtain the low frequency scale coefficient cA2 of last layer, the like until ground floor, obtain the denoised signal isometric with f (x) ', now this signal packet, containing most EEG signals features, has been removed the impact of high-frequency interferencing signal;

In the present embodiment, selected respectively dbN small echo and Haar small echo to carry out denoising to filtered EEG signals;

S4.1), use the alpha rhythm signal place denoising of dbN small echo to filtered EEG signals, denoising result as shown in Figure 6, wherein, Fig. 6 (a) is the EEG signals after denoising, the result that Fig. 6 (b) processes for setting threshold, Fig. 6 (c) is for being used the result after default threshold is processed, the result that Fig. 6 (d) processes for soft-threshold.From upper figure, different threshold settings can make the denoising situation of signal different, and the too high meeting of threshold value filters useful information, as Fig. 6 (b); Too low denoising effect is not obvious, and as Fig. 6 (d), so use default threshold in the present invention when dbN small echo is processed, its result is as Fig. 6 (c);

S4.2), use Haar small echo the alpha rhythm signal of the EEG signals after denoising to be carried out the result of denoising under different threshold values, denoising result as shown in Figure 7, wherein, Fig. 7 (a) is the EEG signals after denoising, Fig. 7 (b) for setting threshold be the result of 1.5 o'clock, Fig. 7 (c) for setting threshold be the result of 3.5 o'clock, Fig. 7 (d) is default value denoising result.

Can find out, the denoising effect of different small echos differs, for haar small echo, although the effect of acquiescence denoising is comparatively obvious, obviously not as dbN small echo is used effective while giving tacit consent to denoising.And use the pressure denoising of lower threshold value can obtain good denoising result, through pretreated signal, there is obvious identification to improve.

S5, the EEG signals by the default CSP feature extraction algorithm of built-in chip in brain electricity cap after to denoising are carried out feature extraction

In the present embodiment, the recovering aid mechanical arm that mechanical arm is 2DOF, therefore, corresponding Characteristic Extraction algorithm can be realized and extract that imagination arm moves upward, the different characteristic quantity of arm while moving downward.

Cospace pattern CSP (Common Spatial Patterns) is a kind of technology of carrying out multi-lead space filtering for two categorical datas, and the target of CSP feature extraction algorithm is structure spatial filter, and its concrete grammar is:

EEG signals after denoising is divided into right-hand man's two class EEG signals, i.e. X, Y, what every class EEG signals was all used C * N is that real number matrix represents, and wherein, C is signalling channel number, and N is sampled point, and the covariance matrix of X and Y can be expressed as:

${\mathrm{\&Sigma;}}_X=\frac{{\mathrm{XX}}^T}{\mathrm{trace}\left({\mathrm{XX}}^T\right)},{\mathrm{\&Sigma;}}_Y=\frac{{\mathrm{YY}}^T}{\mathrm{trace}\left({\mathrm{YY}}^T\right)}$

Wherein, ∑ _xwith ∑ _ybe respectively the covariance matrix for X and Y, [] ^tfor the transposition of [], trace (XX ^t) and trace (YY ^t) be respectively the diagonal element sum of X and Y;

Construct a spatial filter W, when CSP carries out feature extraction, meet:

max∑var(W ^T·X)，s.t.∑var[W ^T·(X+Y)]＝1

Wherein, var () is variance computing function; Variance computing function to above formula launches, and the existence condition that obtains spatial filter W is:

max∑W ^T·Σ _X·W，s.t.∑W ^T·(Σ _X+Σ _Y)·W＝1

Wherein, max (A) s.t (B) is illustrated in and under the condition that meets B, solves maximum A;

The concrete grammar that constructs spatial filter according to the existence condition of spatial filter is:

Solution matrix P, and P meets P (∑ _x+ ∑ _y) P ^t=I, wherein, I is unit matrix;

According to the definition of covariance matrix: covariance matrix is for being symmetrical matrix, symmetrical matrix and be still symmetrical matrix, by symmetrical matrix (∑ _x+ ∑ _y) carry out singular value decomposition and obtain (∑ _x+ ∑ _y)=U φ U ^t, wherein, U is ∑ _x+ ∑ _yeigenvectors matrix, φ is ∑ _x+ ∑ _yeigenvalue diagonal matrix, order obtain and meet P (Σ _x+ Σ _y) P ^tthe matrix P of=I;

Order solve orthogonal matrix R and diagonal matrix D, R and D are met ${\widehat{\mathrm{\&Sigma;}}}_X={\mathrm{RDR}}^T;$

Right carry out singular value decomposition, obtain positive definite symmetric matrices will the orthogonal matrix R that forms of characteristic vector, will eigenvalue as diagonal element, remaining element is made as to 0, construct diagonal matrix D; Current having tried to achieve meets P (∑ _x+ ∑ _y) P ^tthe matrix P of=I is with satisfied matrix R, structural matrix W=R ^tp, ∑ W makes to satisfy condition ^t(Σ _x+ Σ _y) W=1;

Select m column vector of eigenvalue maximum in diagonal matrix D and m column vector of eigenvalue minimum, their corresponding 2m column vectors in spatial filter W are formed to new bank of filters W _sub;

By the EEG signals of any time, by spatial filter W Filtering Processing, C for the EEG signals after processing * L real number matrix S represents, real number matrix S is split as K submatrix again, i.e. S=[S (1), and S (2) ..., S (K)] ^t, each submatrix represents with the real number matrix S () of C * N, and this K submatrix is passed through to bank of filters W successively _subprocess, obtain the matrix of K 2m * N again according to formula wherein, representing matrix S _cSPin the value of the capable j row of the i of l submatrix, l=1,2 ..., K, i=1,2 ..., 2m, j=1,2 ..., N, obtains each eigenvalue f _l, finally by eigenvalue f _lcombination obtains characteristic vector x (l)=[f _l1, f _l2..., f _l2m] ^t, real number matrix S obtains high identification digital signal after feature extraction so $F\left(x\right)=[x\left(1\right),x\left(2\right),\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;x\left(K\right)]=\left[\begin{array}{c}{f}_{11},f_{12},\&CenterDot;\&CenterDot;\&CenterDot;,f_{1K}\\ f_{21},f_{22},\&CenterDot;\&CenterDot;\&CenterDot;,f_{2K}\\ \&CenterDot;\\ \&CenterDot;\\ \&CenterDot;\\ f_{l1},f_{l2},\&CenterDot;\&CenterDot;\&CenterDot;,f_{\mathrm{lK}}\end{array}\right].$

In the present embodiment, rehabilitation system is to carry out for upper limb, therefore gets m=1, i.e. x (l)=[f _l1, f _l2] ^t, real number matrix S obtains high identification digital signal after feature extraction so $F\left(x\right)=[x\left(1\right),x\left(2\right),\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;x\left(K\right)]=\left[\begin{array}{c}{f}_{11},f_{12},\&CenterDot;\&CenterDot;\&CenterDot;,f_{1K}\\ f_{21},f_{22},\&CenterDot;\&CenterDot;\&CenterDot;,f_{2K}\end{array}\right].$

S6, by the grader in brain electricity cap, the digital signal F (l) of high identification is classified

In the present embodiment, employing has self adaptation LDA (Linear Discriminant Analysis, linear discriminant analysis) grader of sorting algorithm is classified to the digital signal F (x) of high identification, and sorted digital signal is as the driving instruction of controlling drive motors equipment and pneumatic muscle auxiliary facilities;

The data of the new input of self adaptation LDA algorithm utilization in grader reappraise classifier parameters, to adapt to current data feature, improve classification accuracy rate.To the self adaptation LDA sorting algorithm of grader, be specifically described as follows below:

S6.1), adopt slip window sampling to carry out Estimation of Mean to EEG signals

Length be the EEG signals of L by feature extraction after, all F (x) are designated as to set { x (j) }, its Estimation of Mean expression formula is:

$\mathrm{\&mu;}=\frac{1}{L}\underset{j=1}{\overset{L}{\mathrm{\&Sigma;}}}x\left(j\right)$

One time slide window is set, will gathers { x (j) } and constantly carry out Estimation of Mean at t, its expression formula is:

$\mathrm{\&mu;}\left(t\right)=\frac{1}{\underset{j=0}{\overset{h-1}{\mathrm{\&Sigma;}}}{w}_j}\underset{j=0}{\overset{h-1}{\mathrm{\&Sigma;}}}{w}_j\&CenterDot;x(t-j)$

Wherein, h represents sliding window width, w _jrepresent window weight coefficient, x (t-j) represents (t-j) EEG signals constantly;

Calculate t-1 Estimation of Mean constantly, by t-1 Estimation of Mean constantly, obtain t Estimation of Mean μ (t) constantly:

μ(t)＝(1-UC)·μ(t-1)+UC·x(t)

Wherein, wherein x (t) is t input EEG signals constantly, and UC represents to upgrade coefficient;

S6.2), estimate the covariance of EEG signals

Length be the EEG signals of L by feature extraction after, all F (x) are designated as to set { x (j) }, the account form of its covariance matrix Σ is:

$\mathrm{\&Sigma;}=\frac{1}{L}\underset{j=1}{\overset{L}{\mathrm{\&Sigma;}}}[x\left(j\right)-\mathrm{\&mu;}]{[x\left(j\right)-\mathrm{\&mu;}]}^T$

Set { x (j) } in the estimated value of t covariance matrix Σ (t) is constantly so:

Σ(t)＝(1-UC)·Σ(t-1)+UC·[1,x(t)] ^T·[1,x(t)]

To t covariance Σ (t) the constantly conversion of inverting:

$\mathrm{\&Sigma;}{\left(t\right)}^{-1}=\frac{1}{1-\mathrm{UC}}\&CenterDot;[\mathrm{\&Sigma;}{(t-1)}^{-1}-\frac{1}{\frac{1-\mathrm{UC}}{\mathrm{UC}}+x{\left(t\right)}^T\&CenterDot;v\left(t\right)}v\left(t\right)\&CenterDot;v{\left(t\right)}^T]$

Wherein, ν (t)=Σ (t-1) ^-1x (t);

S6.3), Estimation of Mean and covariance are carried out to online adaptive estimation, complete reappraising of classifier parameters

c _t＝F{D[x(t)]}＝F[b(t-1)+w(t-1) ^T·x(t)]

${\mathrm{\&mu;}}_{c_t}\left(t\right)=(1-\mathrm{UC})\&CenterDot;{\mathrm{\&mu;}}_{c_t}(t-1)+\mathrm{UC}\&CenterDot;x\left(t\right)$

w(t)＝∑(t) ^-1·[μ ₂(t)-μ ₁(t)]

$b\left(t\right)=-w{\left(t\right)}^T\&CenterDot;\frac{1}{2}\&CenterDot;[{\mathrm{\&mu;}}_1\left(t\right)+{\mathrm{\&mu;}}_2\left(t\right)]$

Wherein, c _trepresent classification labelling c _t∈ { 1,2}; F{D[x (t)] } represent the classification discriminant function of EEG signals x (t), as D[x (t)] and during > 0, c _tvalue is 2, as D[x (t)] during < 0, c _tvalue is that 1, w (t) is the t slope of linear classifier constantly, and b (t) is the t intercept of linear classifier constantly, initial value μ (0), Σ (0) ^-1can choose suitable EEG signals trains and obtains;

The kinestate data of S7, motor encoder collection machinery arm, as the movement velocity of mechanical arm, acceleration, driving moments etc., are sent to kinestate harvester by the kinestate data of collection;

S8, electromyographic signal collection attachment of electrodes are on arm, gather patient's electromyographic signal, as control shoulder joint, elbow joint and the carpal muscle group electromyographic signal of (comprising triangular muscle, biceps brachii m., triceps brachii, shoulder sleeve flesh and forearm flexor group etc.), by data wire, be sent to kinestate harvester; In the present embodiment, the myoelectric apparatus MyoMove of Shanghai Nuo Cheng company is used in the collection of electromyographic signal;

S9, heart rate gather elastic wristband, are worn in patient's wrist, gather Heart Rate, and transfer to kinestate harvester and process;

S10, kinestate harvester are surface-mounted integrated circuit, on surface-mounted integrated circuit, be reserved with and connect the data line interface that motor encoder, electromyographic signal collection electrode and heart rate gather wrist strap, after connecting by data line, encapsulated motor encoder, electromyographic signal collection electrode and heart rate gather the data that elastic wristband transmits, and the data after encapsulation are sent to system central processor as current kinestate data;

S11, system central processor are planned as a whole allotment to receiving data

System central processor receives, stores the data of brain electricity cap and the transmission of kinestate harvester, wherein, the driving instruction that brain electricity cap sends control drive motors equipment and pneumatic muscle auxiliary facilities by wireless transmission method is to system central processor, and kinestate harvester sends kinestate data to system central processor by data line;

Data processor in system central processor carries out data parsing to the data of brain electricity cap and the transmission of kinestate harvester according to default communication protocol, dispel header, telegram end, identify valid data value, again part relevant to kinestate data in valid data value is encoded, according to default communication protocol, add that header, telegram end and type identification complete encapsulation again, data after encapsulation feed back to drive motors equipment and pneumatic muscle auxiliary facilities as feedback signal, and driving device arm is done rehabilitation training; The kinestate data of storage are sent to mobile terminal by inner wireless transmitter simultaneously;

During rehabilitation training, data processor in system central processor is finely tuned drive motors equipment and pneumatic muscle auxiliary facilities according to the data of kinestate harvester transmission again, such as: when kinestate harvester collects that mechanical arm movement velocity is too fast, patient's rapid heart rate etc., regulate the feedback signal that sends to drive motors equipment and pneumatic muscle auxiliary facilities, reduce the operating frequency of drive motors;

The remote monitoring of S12, mobile terminal

In the present embodiment, mobile terminal is connected by wireless mode with system central processor, medical personnel can pass through kinestate and the rehabilitation state of mobile terminal remote monitor patients like this, medical personnel can utilize according to current kinestate mobile terminal to send instruction to system central processor simultaneously, realize medical personnel rehabilitation system is carried out to long-range fine setting;

S13, by the auxiliary rehabilitation exercise of speech ciphering equipment

In the present embodiment, by mike, voice signal is inputted, by being arranged on the speech processing module at display screen back, voice signal is converted into serial ports command signal, thereby the display interface of controlling display platform, display platform sends feedback signal by voice message equipment after receiving serial ports command signal; Bluetooth module in display platform reads the EEG signals in brain electricity cap in real time, and be presented at display interface, patient can be according to the state showing, the rehabilitation of calling in display platform by the mode of phonetic entry in conjunction with self-condition is played, and carries out auxiliary rehabilitation exercise.

Although above the illustrative specific embodiment of the present invention is described; so that those skilled in the art understand the present invention; but should be clear; the invention is not restricted to the scope of the specific embodiment; to those skilled in the art; as long as various variations appended claim limit and definite the spirit and scope of the present invention in, these variations are apparent, all utilize innovation and creation that the present invention conceives all at the row of protection.

Claims (4)

1. the upper limb healing system based on bio signal, is characterized in that, comprising:

One EEG signals extraction equipment is a brain electricity cap, brain electricity cap is attached to brain surface, by the metal electrode induction EEG signals of brain electricity cap, the built-in chip of brain electricity cap amplifies EEG signals again successively, filtering and denoising, by CSP feature extraction algorithm, the EEG signals after denoising is converted into again to the digital signal of high identification, digital signal is classified by having the grader of self adaptation LDA sorting algorithm again, complete feature identification, digital signal after identification is converted into the driving instruction of controlling drive motors and pneumatic muscle auxiliary facilities, finally by the wireless device in brain electricity cap, send,

One system central processor, comprise data processor and wireless transmitter, be arranged on the top of back bracket, for receiving, the kinestate data that gather in the driving instruction that storage brain electricity cap sends and kinestate harvester, wherein, data processor carries out data parsing to the driving instruction and the kinestate data that receive according to default communication protocol, identify valid data value, again part relevant to kinestate data in valid data value is encoded, finally the data after coding are encapsulated and feed back to drive motors equipment and pneumatic muscle auxiliary facilities as feedback signal, the kinestate data of storage are sent to mobile terminal by wireless transmitter simultaneously,

One back bracket, is attached to the back of human body, and its top is connected with carbon fiber arm ectoskeleton, and flushes with the shoulder of human body, for bearing the weight of rehabilitation equipment, and hangs the system central processor at back;

One carbon fiber arm ectoskeleton, comprise upper arm and forearm, upper arm and forearm are two carbon fiber boards that fit tightly and be screwed, junction, joint at upper arm and back bracket, upper arm and forearm connects to form mechanical arm with screw respectively, and on upper arm and forearm, all there is equally spaced screw hole, can make position, the wrong hole of carbon fiber board install, for regulating mechanical arm length;

Two drive motors equipment, every group of drive motors equipment comprises a drive motors and one drive circuit plate, to be installed on respectively the ectoskeletal top arm's tip of carbon fiber arm, and the junction, joint of upper arm and forearm, drive circuit board is for gathering the driving frequency of drive motors, and drive motors is for the motion of driving device arm;

Two pneumatic muscle auxiliary drive apparatus, are installed on respectively upper arm and back bracket, upper arm and junction, forearm joint, for cutting down mechanical shaking and driving force compensation;

One motor encoder, be installed on drive motors tail end, and connect drive circuit board and kinestate harvester by data line, by the monitoring to drive motors kinestate, obtain the kinestate data of mechanical arm, and these kinestate data are sent to kinestate harvester;

One electromyographic signal collection electrode, comprises and electrode signal acquisition and the data line of an adhesive type electrode signal acquisition is sticked on to patient's arm, for gathering patient's electromyographic signal, and transfers to kinestate harvester;

One heart rate gathers elastic wristband, comprises built-in pulse collection electrode and data line, is worn on patient's wrist, for gathering patient's heart rate, and transfers to kinestate harvester;

One kinestate harvester is surface-mounted integrated circuit, be arranged on the outside of mechanical arm, on surface-mounted integrated circuit, be reserved with and connect the data line interface that motor encoder, electromyographic signal collection electrode and heart rate gather wrist strap, after connecting by data line, from motor encoder electromyographic signal collection electrode and heart rate, gather elastic wristband reception data, the data that receive are encapsulated, and be sent to system central processor as current kinestate data;

One voice command receiving equipment is a phonetic incepting microphone, is installed on forearm tail end, and also detachable use separately, for input speech signal;

One voice message equipment, comprises a speech processing module, is arranged on display screen back; The voice signal of input is converted into serial ports command signal by speech processing module, and for controlling the display interface menu of display platform, display platform sends feedback signal by voice message equipment again after receiving serial ports command signal;

One display platform, is installed on rehabilitation system the place ahead, and its display interface menu is controlled by phonetic order, is built-in with bluetooth module and rehabilitation training game; Bluetooth module can directly read EEG signals from potentials extraction equipment, and the auxiliary patient of rehabilitation training game carries out rehabilitation training;

Brain electricity cap is attached to brain surface, by the metal electrode induction EEG signals of brain electricity cap, the built-in chip of brain electricity cap to EEG signals amplify successively, filtering and denoising, by CSP feature extraction algorithm, the EEG signals after denoising is converted into again to the digital signal of high identification, digital signal is classified by having the grader of self adaptation LDA sorting algorithm again, complete feature identification, the digital signal after identification is converted into the driving instruction of controlling drive motors equipment and pneumatic muscle auxiliary facilities;

Drive instruction to be transferred to system central processor by the mode of wireless transmission, the kinestate data that data processor in system central processor gathers in conjunction with kinestate harvester again, according to default communication protocol, carry out data parsing, identify valid data value, again part relevant to kinestate data in valid data value is encoded, finally the data after coding are encapsulated and feed back to drive motors equipment and pneumatic muscle auxiliary facilities as feedback signal, thereby driving device arm is done rehabilitation exercise according to feedback signal;

Simultaneity factor central processing unit is sent to mobile terminal by kinestate data by inner wireless transmitter, medical personnel can pass through kinestate and the rehabilitation state of mobile terminal remote monitored patient, the rehabilitation training that patient also can call in display platform by voice command receiving equipment is played, and carries out auxiliary rehabilitation exercise.

2. the upper limb healing system based on bio signal according to claim 1, is characterized in that, the method that in described brain electricity cap, built-in chip carries out denoising to EEG signals is: by default wavelet algorithm, EEG signals is carried out to denoising, idiographic flow is:

At function space L ²(R), in, establish wavelet function C _Ψmeet:

$C_{\mathrm{\&Psi;}}={\&Integral;}_{R^{+}}\frac{\left|\widehat{\mathrm{\&Psi;}}\left(x\right)\right|}{\left|w\right|}\mathrm{dw}<\&infin;$

Wherein, R is all real number fields, and w is constant and w ≠ 0, and Ψ (x) is called wavelet mother function, fourier transformation for Ψ (x); For convenience of narration, make wavelet mother function Ψ (x) be:

$\mathrm{\&Psi;}(a,b)\left(x\right)=\frac{1}{\sqrt{\left|a\right|}}\mathrm{\&Psi;}\left(\frac{x-b}{a}\right)$

Wherein, a and b are contraction-expansion factor and shift factor, control respectively stretching and the translation transformation of the female ripple of small echo;

The EEG signals f that is L by length (x) carries out continuous wavelet transform:

$W_f(a,b)=\frac{1}{\sqrt{a}}{\&Integral;}_{R}f\left(x\right)\mathrm{\&Psi;}\left(\frac{x-b}{a}\right)\mathrm{dx}$

The wavelet transformation of this EEG signals f (x) carries out inverse transformation:

$f{\left(x\right)}^{\&prime;}=C_{\mathrm{\&Psi;}}{\&Integral;\&Integral;}_{R\&times;R^{+}}{W}_f(a,b)\mathrm{\&Psi;}\left(\frac{x-b}{a}\right)\mathrm{dadb}$

Wherein, R * R ⁺represent that this double integral is the integration in real number field;

According to the convolution property of small echo and down-sampling characteristic, f (x) ' can only be decomposed into log at most ₂l layer,, when ground floor decomposes, makes convolution operation with high frequency coefficient and the low frequency coefficient of small echo signal respectively by f (x), more respectively convolution signal is later done to down-sampling, obtains low frequency scale coefficient and high frequency detail coefficients; From the second layer decomposes, last layer is decomposed to the low frequency scale coefficient obtaining and make convolution and down-sampling with high frequency coefficient and the low frequency coefficient of small echo signal respectively, obtain low frequency scale coefficient and high frequency detail coefficients after this layer of decomposition, to the last one deck log ₂l, finally from log ₂l layer starts, and low frequency scale coefficient and high frequency detail coefficients are carried out doing contrary convolution with low frequency coefficient and the high frequency coefficient of small echo signal respectively after unit time delay, obtains the low frequency scale coefficient of last layer, the like until ground floor obtains the EEG signals after denoising.

3. the upper limb healing system based on bio signal according to claim 1, is characterized in that, described CSP feature extraction algorithm is:

3.1, calculate the covariance matrix of EEG signals

EEG signals after denoising is divided into right-hand man's two class EEG signals, i.e. X, Y, what every class EEG signals was all used C * N is that real number matrix represents, and wherein, C is signalling channel number, and N is sampled point, and the covariance matrix of X and Y can be expressed as:

${\mathrm{\&Sigma;}}_X=\frac{{\mathrm{XX}}^T}{\mathrm{trace}\left({\mathrm{XX}}^T\right)},{\mathrm{\&Sigma;}}_Y=\frac{{\mathrm{YY}}^T}{\mathrm{trace}\left({\mathrm{YY}}^T\right)}$

Wherein, ∑ _xwith ∑ _ybe respectively the covariance matrix for X and Y, [] ^tfor the transposition of [], trace (XX ^t) and trace (YY ^t) be respectively the diagonal element sum of X and Y;

3.2, structure spatial filter

3.2.1 determine the existence condition of spatial filter

Construct a spatial filter W, while making CSP feature extraction algorithm carry out feature extraction, meet:

max∑var(W ^T·X)，s.t.∑var[W ^T·(X+Y)]

Wherein, var () is variance computing function; Variance computing function to above formula launches, and the existence condition that obtains spatial filter W is:

max∑W ^T·Σ _X·W，s.t.∑W ^T·(Σ _X+Σ _Y)·W＝1

3.2.2, according to the existence condition of spatial filter, construct spatial filter

Solution matrix P, and P meets P (∑ _x+ ∑ _y) P ^t=I, wherein, I is unit matrix;

According to the definition of covariance matrix: covariance matrix is for being symmetrical matrix, symmetrical matrix and be still symmetrical matrix, by symmetrical matrix (∑ _x+ ∑ _y) carry out singular value decomposition and obtain (∑ _x+ ∑ _y)=U φ U ^t, wherein, U is (∑ _x+ ∑ _y) eigenvectors matrix, φ is (∑ _x+ ∑ _y) eigenvalue diagonal matrix, order obtain and meet P (Σ _x+ Σ _y) P ^tthe matrix P of=I;

Order solve orthogonal matrix R and diagonal matrix D, R and D are met ${\widehat{\mathrm{\&Sigma;}}}_X={\mathrm{RDR}}^T;$

Right carry out the decomposition of singular value value, obtain positive definite symmetric matrices will the orthogonal matrix R that forms of characteristic vector, will eigenvalue as diagonal element, remaining element is made as to 0, construct diagonal matrix D;

According to ${\widehat{\mathrm{\&Sigma;}}}_X+{\widehat{\mathrm{\&Sigma;}}}_Y=I,{\widehat{\mathrm{\&Sigma;}}}_Y=R\&CenterDot;(I-D)\&CenterDot;R^T,$ Construct spatial filter W=R ^tp;

3.3, designing filter group W _sub

Select m column vector of eigenvalue maximum in diagonal matrix D and m column vector of eigenvalue minimum, their corresponding 2m column vectors in spatial filter W are formed to new wave filter W _sub;

3.4, the EEG signals after denoising is carried out to feature extraction, obtain the digital signal F (x) of high identification

By the EEG signals of any time, by spatial filter W Filtering Processing, C for the EEG signals after processing * L real number matrix S represents, real number matrix S is split as K submatrix again, i.e. S=[S (1), and S (2) ..., S (K)] ^t, each submatrix represents with the real number matrix S () of C * N, and this K submatrix is passed through to bank of filters W successively _subprocess, obtain the matrix of K 2m * N again according to formula wherein, representing matrix S _cSPin the value of the capable j row of the i of l submatrix, l=1,2 ..., K, i=1,2 ..., 2m, j=1,2 ..., N, obtains each eigenvalue f _l, finally by eigenvalue f _lcombination obtains characteristic vector x (l)=[f _l1, f _l2..., f _l2m] ^t, real number matrix S obtains high identification digital signal after feature extraction so $F\left(x\right)=[x\left(1\right),x\left(2\right),\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;x\left(K\right)]=\left[\begin{array}{c}{f}_{11},f_{12},\&CenterDot;\&CenterDot;\&CenterDot;,f_{1K}\\ f_{21},f_{22},\&CenterDot;\&CenterDot;\&CenterDot;,f_{2K}\\ \&CenterDot;\\ \&CenterDot;\\ \&CenterDot;\\ f_{l1},f_{l2},\&CenterDot;\&CenterDot;\&CenterDot;,f_{\mathrm{lK}}\end{array}\right].$

4. the upper limb healing system based on bio signal according to claim 1, is characterized in that, the self adaptation LDA sorting algorithm of described grader is:

4.1), adopt slip window sampling to carry out Estimation of Mean to EEG signals

Length be the EEG signals of L by feature extraction after, all F (x) are designated as to set { x (j) }, its Estimation of Mean expression formula is:

$\mathrm{\&mu;}=\frac{1}{L}\underset{j=1}{\overset{L}{\mathrm{\&Sigma;}}}x\left(j\right)$

One time slide window is set, will gathers { x (j) } and constantly carry out Estimation of Mean at t, its expression formula is:

$\mathrm{\&mu;}\left(t\right)=\frac{1}{\underset{j=0}{\overset{h-1}{\mathrm{\&Sigma;}}}{w}_j}\underset{j=0}{\overset{h-1}{\mathrm{\&Sigma;}}}{w}_j\&CenterDot;x(t-j)$

Wherein, h represents sliding window width, w _jrepresent window weight coefficient, x (t-j) represents (t-j) EEG signals constantly;

Calculate t-1 Estimation of Mean constantly, by t-1 Estimation of Mean constantly, obtain t Estimation of Mean μ (t) constantly:

μ(t)＝(1-UC)·μ(t-1)+UC·x(t)

Wherein, wherein x (t) is t input EEG signals constantly, and UC represents to upgrade coefficient;

4.2), estimate the covariance of EEG signals

Length be the EEG signals of L by feature extraction after, all F (x) are designated as to set { x (j) }, the account form of its covariance matrix Σ is:

$\mathrm{\&Sigma;}=\frac{1}{L}\underset{j=1}{\overset{L}{\mathrm{\&Sigma;}}}[x\left(j\right)-\mathrm{\&mu;}]{[x\left(j\right)-\mathrm{\&mu;}]}^T$

Set { x (j) } in the estimated value of t covariance matrix Σ (t) is constantly so:

Σ(t)＝(1-UC)·Σ(t-1)+UC·[1,x(t)] ^T·[1,x(t)]

To t covariance Σ (t) the constantly conversion of inverting:

$\mathrm{\&Sigma;}{\left(t\right)}^{-1}=\frac{1}{1-\mathrm{UC}}\&CenterDot;[\mathrm{\&Sigma;}{(t-1)}^{-1}-\frac{1}{\frac{1-\mathrm{UC}}{\mathrm{UC}}+x{\left(t\right)}^T\&CenterDot;v\left(t\right)}v\left(t\right)\&CenterDot;v{\left(t\right)}^T]$

Wherein, ν (t)=Σ (t-1) ^-1x (t);

4.3), Estimation of Mean and covariance are carried out to online adaptive estimation, complete reappraising of linear classifier parameter

c _t＝F{D[x(t)]}＝F[b(t-1)+w(t-1) ^T·x(t)]

${\mathrm{\&mu;}}_{c_t}\left(t\right)=(1-\mathrm{UC})\&CenterDot;{\mathrm{\&mu;}}_{c_t}(t-1)+\mathrm{UC}\&CenterDot;x\left(t\right)$

w(t)＝∑(t) ^-1·[μ ₂(t)-μ ₁(t)]

$b\left(t\right)=-w{\left(t\right)}^T\&CenterDot;\frac{1}{2}\&CenterDot;[{\mathrm{\&mu;}}_1\left(t\right)+{\mathrm{\&mu;}}_2\left(t\right)]$

Wherein, c _trepresent classification labelling c _t∈ { 1,2}; F{D[x (t)] } represent the classification discriminant function of EEG signals x (t), as D[x (t)] and during > 0, c _tvalue is 2, as D[x (t)] during < 0, c _tvalue is that 1, w (t) is the t slope of linear classifier constantly, and b (t) is the t intercept of linear classifier constantly.