CN103190905B - Multi-channel surface electromyography signal collection system based on wireless fidelity (Wi-Fi) and processing method thereof - Google Patents

Abstract

The invention discloses a multi-channel surface electromyography signal collection system based on the wireless fidelity (Wi-Fi) and a processing method thereof. The collection system comprises an electromyography signal conditioning circuit, a center control unit and an 802.11 controller. A surface electromyography electrode transmits original signals of collected surface electromyography signals to the electromyography signal conditioning circuit to conduct signal conditioning, and the conditioned signals are transmitted to a processing platform through the center control unit 202. The system utilizes a self-returning augmented reality (AR) model to conduct characteristic extraction on the surface electromyography signals, utilizes a support vector machine method to recognize moving direction and predict action force of human body lower limb ankles, is capable of effectively overcoming the shortcoming that a traditional electromyography signal processing method is low in recognition accuracy, and a recognition result is only disperse two-value action and the like.

Description

Multi-channel surface myoelectric signal acquiring system and processing method based on Wi-Fi

Technical field

The present invention relates to the acquisition and processing to the surface electromyogram signal at skin surface place, refer to particularly a kind of multi-channel surface myoelectric signal acquiring system and processing method based on Wi-Fi.

Background technology

Surface myoelectric (surface electromyography, abbreviation SEMG) signal is comprehensive on skin surface place time and space of muscle electrical activity.Different neuromuscular systems can produce the bioelectrical signals with different qualities when carrying out arbitrarily and non-randomness is movable, by extracting and study different bioelectrical signals characteristics, can effectively identify human motion action, diagnose disorder of muscle and instruct rehabilitation medical etc.In recent years, SEMG is not only widely used in the fields such as medical diagnosis on disease, rehabilitation medicine, motion physical culture, and as a kind of man-machine interaction input mode of novelty and receive much concern.

Yet traditional Surface Electromyography Signal Acquisition System adopts the transmission network of wired or low-speed wireless mostly, realize the electromyographic signal transfer of data of the control appliances such as capture card and computer.As use low rate radio frequency chip CC2500 as the data link of capture card and control terminal, or document adopts wired SEMG electrode collection surface electromyographic signal.The wired electromyographic signal collection jig of tradition has the advantages such as high accuracy, low delay, but has the shortcomings such as serious power frequency is disturbed, portable difference; Though low-speed wireless capture card can overcome the shortcomings of wired electromyographic signal collection, portability, ease for use and the safety of acquisition system have been strengthened, but the wireless link of low rate has been sacrificed the real-time of acquisition system, the indexs such as high accuracy under some special applications demand, high sampling rate have even been reduced.In addition, traditional wireless electromyographic signal collection system need to customize communication hardware and the communication protocol of the control appliance ends such as computer, and the versatility of traditional electromyographic signal collection equipment is greatly limited.

Meanwhile, electromyographic signal processing method also becomes the emphasis of research gradually, and relevant electromyographic signal has also had larger progress in the research of the aspects such as muscle function diagnosis, athletic performance identification and muscle fatigue evaluation.Utilize the row mode classification of going forward side by side of some feature of human body surface myoelectric signal that myoelectricity capture card extracts, can obtain kinesiology's relevant information, and then drive artificial limb or rehabilitation medicine equipment assisting patient to complete corresponding muscle movement.Wherein the method for electromyographic signal feature extraction and pattern classification is the research emphasis in processing of bioelectric signals field all the time.Recent study person studies from the feature extracting method of the aspect effects on surface electromyographic signals such as time domain, frequency domain, time-frequency domain respectively.Traditional Time Domain Analysis is electromyographic signal to be regarded as to the function of time, conventional Time Domain Analysis comprises absolute value integral mean (MeanAbsolute Value, abbreviation MAV), root-mean-square value (root meam square, abbreviation RMS), zero passage count, variance and second moment etc. or by extracting AR model coefficient as the characteristic vector of classification.In frequency-domain analysis method, conventionally adopt Fourier transformation to complete the conversion between discrete series and frequency spectrum, conventional index comprises power spectral density, frequency of average power (mean power frequency, abbreviation MPF), median frequency (median frequency, abbreviation MF) etc.In recent years, the research of using Time-Frequency Analysis method to extract electromyographic signal feature has also had larger progress.Typical method is as short time discrete Fourier transform (Short-Time Fourier Transform, abbreviation STFT), wavelet transformation (Wavelet Transform, abbreviation WT), wavelet package transforms (Wavelet Packet Transform, abbreviation WPT), Wigner-Ville (the Wigner-Ville Distribution that distributes, abbreviation WVD), Complex Cepstrum Coefficient, linear predictor coefficient (Linear Predictive Coefficient, abbreviation LPC) etc.The sorting technique of processing of bioelectric signals domain pattern identification mainly comprises BP neutral net, Fuzzy Pattern Recognition and support vector machine etc.

Along with the extensive use of surface electromyogram signal in fields such as the biomedical engineerings such as pattern recognition, rehabilitation medicine and sports sciences, the generalization ability to analysis, identification and the processing of the acquisition precision of electromyographic signal collection card, real-time, portability etc. and electromyographic signal process software is proposed to higher requirement.

Summary of the invention

In view of this, the object of the present invention is to provide a kind of multi-channel surface myoelectric signal acquiring system and surface electromyogram signal processing method based on Wi-Fi, to overcome above-mentioned deficiency of the prior art.

For achieving the above object, the invention provides a kind of multi-channel surface myoelectric signal acquiring system based on Wi-Fi, comprising:

The surface myoelectric utmost point, the primary signal that contains surface electromyogram signal for obtaining skin surface place;

Preamplifier, is connected with the described surface myoelectric utmost point, for amplifying described primary signal, and the primary signal after being amplified;

Wave trap, is connected with described preamplifier, for removing the electromagnetic interference signal of primary signal after described amplification;

Band filter, is connected with described wave trap, for eliminating primary signal after the described amplification signal except surface electromyogram signal, obtains the first processing signals;

Final amplifier, is connected with described band filter, for to described the first processing signals gain and level lifting, obtains the second processing signals;

Analog-to-digital conversion module, is connected with described final amplifier, for to described the second processing signals analog digital conversion, samples and obtains original discrete electromyographic signal sequence;

Wireless communication module, is connected with described analog-to-digital conversion module, for described original discrete electromyographic signal sequence is transmitted by Wi-Fi wireless network;

Processing platform, is connected with described wireless communication module by Wi-Fi wireless network, receives described original discrete electromyographic signal sequence.

In addition, the present invention also provides a kind of processing method to above-mentioned institute collection surface electromyographic signal, comprises the following steps:

The AR coefficient of the original discrete electromyographic signal sequence that extraction human motion produces at skin surface place is as human body lower limbs direction of motion feature.The sample that the root-mean-square value of original discrete electromyographic signal and lower limb actual force size are formed is as the training sample set of lower limb active force forecast model.

Further, described lower extremity movement direction character is by using Yule-Walker equation solution AR model to calculate 4 rank AR coefficients of 4 passage electromyographic signal sequences of current collection, and forms the characteristic vector μ of action to be identified;

Described human motion amount of force prediction is by calculating the amount of force of lower limb electromyographic signal RMS and actual measurement, the training sample ν of common anabolic action power forecast model.

Described eigenvalue μ and training sample ν are inputted respectively to the size that C-SVM and ε-SVR obtain human motion direction and motion active force.

The multi-channel surface myoelectric signal acquiring system cost that the present invention is based on Wi-Fi is low, real-time good, and accuracy rate is high.And utilize autoregression (AutoRegressive, abbreviation AR) model effects on surface electromyographic signal is carried out feature extraction, use support vector machine (Support Vector Machine, abbreviation SVM) method is to the identification of the human body lower limbs ankle direction of motion and the prediction of active force, can effectively overcome that traditional electromyographic signal processing method accuracy of identification is low, recognition result is only the shortcomings such as discrete two-value action.

Accompanying drawing explanation

Fig. 1 is the use schematic diagram that the present invention is based on the multi-channel surface myoelectric signal acquiring system of Wi-Fi;

Fig. 2 is the structured flowchart that the present invention is based on the multi-channel surface myoelectric signal acquiring system of Wi-Fi;

Fig. 3 is the structured flowchart of electromyographic signal modulate circuit in Fig. 1;

Fig. 4 is the process chart of surface electromyogram signal.

The specific embodiment

Below in conjunction with the drawings and specific embodiments, the present invention is described in further detail.

As shown in Figure 1, myoelectricity conducting wire 100 is connected with human body, the primary signal that the surface myoelectric utmost point 101 of myoelectricity conducting wire 100 front ends contains surface electromyogram signal for gathering skin surface place, the multi-channel surface myoelectric signal acquiring system 200 that the present invention is based on Wi-Fi is connected with myoelectricity conducting wire 100, after the original electromyographic signal of myoelectricity conducting wire 100 acquisition skin surfaces, the multi-channel surface myoelectric signal acquiring system 200 of input based on Wi-Fi carries out signal processing, and transfers to relevant treatment platform by Wi-Fi.Processing platform is the blood processor with Wi-Fi radio communication function, and in the present embodiment, processing platform is that computer 300 or handheld-type intelligent terminal 400(are as panel computer etc.).

The present invention is based on Wi-Fi multi-channel surface myoelectric signal acquiring system 200 structure as shown in Figure 2, this acquisition system comprises: electromyographic signal modulate circuit 201, middle control unit 202,802.11 baseband controllers 203 are connected with lithium battery and electric power controller 204.。The surface myoelectric utmost point 101 transfers to electromyographic signal modulate circuit 201 by the primary signal of the surface electromyogram signal of acquisition and carries out signal condition, and the signal after conditioning transfers to processing platform by Wi-Fi after the ADC module converts of middle control unit 202 is original discrete electromyographic signal sequence.Electromyographic signal modulate circuit 201 is all connected with lithium battery and electric power controller 204 with middle control unit 202.

As shown in Figure 3, the present embodiment electromyographic signal modulate circuit 201 used comprises preamplifier 301, wave trap 302, band filter 303, final amplifier 304 successively.Preamplifier 301 is connected with the surface myoelectric utmost point 101, for amplifying primary signal, and the primary signal after being amplified; Wave trap 302 is for removing the electromagnetic interference signal that amplifies rear primary signal; Band filter 303, for eliminating primary signal after the amplification signal except surface electromyogram signal, obtains the first processing signals; Final amplifier 304, for to the first processing signals gain and level lifting, obtains the second processing signals.The front stage gain of 26dB that the present embodiment preamplifier 301 used forms for AD8295 instrumentation amplifier; The 50Hz active filter that the rear class amplifier that wave trap 302 used is AD8295 instrumentation amplifier forms; Band filter 303 used is 20Hz ~ 500Hz band filter; Final amplifier 304 used is LMV324AD operational amplifier.

In the present embodiment, control unit 202 and comprise analog-to-digital conversion module and wireless communication module, wherein Wi-Fi communication interface adopts WM631 module (802.11 baseband controller 203) to combine with STM32F103RCT6 type processor, and both are by SDIO interface communication.Wherein STM32F103RCT6 type processor complete 802.11, ICP/IP protocol stack and 802.11 controller drivers, WM631 module completes the functions such as 802.11 MAC layer Access Control and radio communication.In addition, the inner analog-to-digital conversion module of multi-channel surface myoelectric signal acquiring system 200 based on Wi-Fi adopts the ADC module of STM32F103RCT6 type processor inside to realize, and processor in real time original discrete electromyographic signal sequence finally transfers to network by Wi-Fi communication interface.Middle control unit 202 also comprises network indication 205 and user interface 206.

The processing of carrying out following steps of the surface electromyogram signal that the present invention gathers the multi-channel surface myoelectric signal acquiring system 200 based on Wi-Fi:

Step S401: because AR model is a kind of linear prediction model, be a known N data, can release N point data or below above by model, so the AR model of take estimates to contain electromyographic signal useful feature as basic spectrum, by calculating the corresponding AR coefficient of AR model as the characteristic vector of sample action, can effectively distinguish different sample actions.The transfer function of AR model and AR coefficient corresponding relation are as follows:

$H_{\mathrm{AR}}\left(Z\right)=\frac{1}{1+\underset{k=1}{\overset{p}{\mathrm{\Σ}}}{a}_k{Z}^{-k}}$ ，

In formula, p is the exponent number of selected AR model, and ak is p rank AR model coefficients, and p and ak determine the feature that spectrum is estimated jointly.

The present invention calculates respectively 4 passage 4 rank AR model parameter a11, a12, a13, a14, a21 ... a44, and the AR coefficient sets of 4 passages is combined into one-dimensional vector ε={ a11, a12, a13, a14, a21 ... a44}, forms the eigenvalue μ of action to be identified, the characteristic vector using μ as support vector machine pattern classification algorithm (C-SVM) pattern recognition.AR coefficient solves and uses Yule-Walker equation (correlation method) to solve, and by following formula, calculates:

$\mathrm{\ϵ}=\underset{i,j=0}{\overset{p}{\mathrm{\Σ}}}{a}_{\mathrm{pj}}R(i-j)a_{\mathrm{pj}}=a^T\mathrm{Ra}$ ，

In formula, R matrix is solved and is obtained by sampling auto-correlation function, samples autocorrelative computational methods as follows:

$R\left(k\right)=\frac{1}{N}\underset{k=0}{\overset{N-1-k}{\mathrm{\Σ}}}x\left(n\right)x(n+k),0\≤k\≤N-1$ ，

X in formula (n) represents the value that sampling sequence is ordered at n, and N represents sampling sequence length.

In order to reduce Yule-Walker equation solution dimension, solving of equation adopts Levinson-Durbin algorithm, by parameter iteration derivation AR (k+1) the rank model parameter of AR (k) rank model.For AR (p) model, recursion calculates until k+1=p, can derive the AR coefficient of p rank AR model.

Step S402: calculate respectively the discrete electromyographic signal sequence of 4 passage time domain root-mean-square R1, R2, R3 and R4 and be combined into one-dimensional vector ε={ R1, R2, R3, R4}, and by its normalization.The present embodiment is when gathering original discrete electromyographic signal sequence, adopt three-dimensional force sensor to measure in real time lower limb amount of force as the sample of human motion actual force size, by the training sample ν of the common anabolic action power of the sample forecast model of the root-mean-square value of original discrete electromyographic signal and lower limb amount of force, the training sample set using ν as support vector machine nonlinear regression (ε-SVR) pattern recognition.Wherein root-mean-square calculates by following formula:

$\mathrm{RMS}=\sqrt{\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{z}^2\left(i\right)}$

In formula, the discrete electromyographic signal sequence of z (i) expression single channel z (i) | and i=1,2 ..., N}, N is the discrete electromyographic signal sequence length of single channel.

Step S403: eigenvalue μ and ν are inputted respectively to the size that C-SVM and ε-SVR obtain human motion direction and motion active force.

The present invention chooses C-SVM and ε-SVR algorithm of support vector machine, identify respectively human motion direction of action and predicting function power size, its principle is: its essence of supporting vector machine model by lower dimensional space vector x nonlinear mapping to higher dimensional space φ (x), and ask for optimum hyperplane in higher-dimension φ space,

Y=w. φ (x)+b, makes minimum,

$\begin{array}{c}x.t.w^T\mathrm{\φ}\left(x_i\right)+b-z_i\≤\mathrm{\ϵ}+{\mathrm{\ξ}}_i^{*}\\ \mathrm{zi}-w^T\mathrm{\φ}\left(x_i\right)-b\≤\mathrm{\ϵ}+{\mathrm{\ξ}}_i^{*},\mathrm{\ξ},{\mathrm{\ξ}}^{*}\≥0,i=\mathrm{1,2},...,l\end{array}$

Wherein w is hyperplane method vector, ξ, and ξ * is relaxation factor, and C is relaxation factor punishment parameter, and b is hyperplane biasing, and ε is for returning loss.Above formula is finally converted to lagrange duality problem and can obtains:

$\begin{array}{c}\underset{a,a^{*}}{\max }\{-\frac{1}{2}\underset{i=1}{\overset{l}{\mathrm{\Σ}}}\underset{j=1}{\overset{l}{\mathrm{\Σ}}}(a_i-a_i^{*})(a_j-a_j^{*}).K(z_i,z_j)-\mathrm{\ϵ}\underset{i=1}{\overset{l}{\mathrm{\Σ}}}(a_i+a_i^{*})+\underset{i=1}{\overset{l}{\mathrm{\Σ}}}{y}_i(a_i-a_i^{*})\}\\ s.t.\underset{i=1}{\overset{l}{\mathrm{\Σ}}}(a_i-a_i^{*})=0(0\≤a_i\≤C;0\≤a_i^{*}\≤C)\end{array}$

Finally obtain regression function

$f_I\left(z_0\right)=\underset{i=1}{\overset{l}{\mathrm{\Σ}}}(a_i-a_i^{*})K(z_i,z_0)+b$ ，

Z wherein _ifor the training sample v that comprises amount of force, z ₀for the root-mean-square vector of real-time electromyographic signal, f (z ₀) be z ₀corresponding prediction of output amount of force, a _i, a _i* be Lagrange multiplier coefficient vector, b is amount of bias, and K is kernel function.Realize the kernel function of the present invention's process software used and select RBF core, that is:

K(x _i,x _j)=exp(-γ||x _i-x _j|| ²)，

Wherein γ is kernel functional parameter, γ >0.C=1，p=0.1。The present invention realizes C-SVM and ε-SVR algorithm of support vector machine by LibSVM storehouse.

The present invention uses OpenGL storehouse to realize the emulation of human body dummy model, and utilizes 3DS MAX software, adopts body modeling method to set up respectively the multi-level model such as human body, thigh, shank and foot.

The present invention imports anthropometric dummy in the platform lower leaf time of VC++6.0, by summit-leg-of-mutton method, carries out the drafting of anthropometric dummy.Model external shape is mainly to adopt triangle to be similar to, gore with respect to observer towards utilizing the judgement of vertex scheme vector.Solving vertex scheme vector adopts and take the meansigma methods of the normal vector of all that this point is summit and obtain.

In addition, the shape library based on OpenGL utilizes double buffering technology to realize high-speed plotting, and when front buffer carries out models show, back buffer district carries out the drafting of next model, has avoided the scintillation occurring in procedure for displaying.In the present embodiment, dummy model is controlled the anthropometric dummy lower limb ankle direction of motion and movement velocity in real time according to the type of action of lower limb electromyographic signal pattern recognition and amount of force.When human body dummy model based on OpenGL shows by conversion OpenGL view techniques to each hierarchical model realize translation, Rotation and Zoom converts, realize lower limb ankle normal attitude, protract, rear quarters, outward turning and 5 groups of states of inward turning.Process software changes the exercise data of each hierarchical model by reading in real time human motion direction and amount of force, i.e. the amount of translation, Rotation and Zoom is again played up scene in window, thus the athletic performance of final reproducing virtual human body.

The beneficial effect of patent of the present invention is: by system hardware of the present invention, gathering 4 channel sample speed is 16KHZ, sampling resolution 12bit, and minimum signal amplitude is 10uV.Use 802.11 communication protocols simultaneously, realize the seamless link of the digital terminals such as capture card and PC, mobile phone, enriched electromyographic signal diversification processing platform.In addition, the present invention can realize the action recognition rate up to 91.2%, active force precision of prediction and the multivariant ankle motion virtual models show that error is 9.8% to the processing method of collection surface electromyographic signal.

The present invention calculates time domain and the frequency domain parameters such as the mean-square value (RMS) of peroneus longus, tibia longue, extensor digitorum longus and peroneus brevis of human body lower limbs shank and frequency of average power (MPF) in real time by electromyographic signal process software.Wherein RMS characterizes muscle health state to a certain extent, and MPF presents downward trend under muscle fatigue state.Software is by the discrete original electromyographic signal sequence Z of intercepting 1000 byte lengths _i, i=0,1 ..., 999, and utilize formula to calculate respectively RMS, MPF parameter.RMS and MPF result of calculation by output display at software interface with for assessment of muscle fatigue and health status.The computing formula of RMS parameter:

$\mathrm{RMS}=\sqrt{\frac{1}{1000}\underset{i=0}{\overset{999}{\mathrm{\Σ}}}{z_i}^2}$ ,

Z wherein _ifor the original discrete electromyographic signal sequence of sampling and obtaining.

During MPF calculation of parameter, first software carry out FFT conversion to original electromyographic signal, and establishing electromyographic signal sequence is x (n), and alternative approach is:

$X\left(k\right)=\underset{n=0}{\overset{N-1}{\mathrm{\Σ}}}x\left(n\right)e^{\frac{2\mathrm{\πjnk}}{N}}$ ，

The X obtaining (k) sequence is corresponding with actual frequency, obtains the spectrum distribution of signal by corresponding conversion.According to symmetry and the MPF computational methods of FFT conversion, have:

$\mathrm{MPF}=\frac{\underset{k=0}{\overset{N/2}{\mathrm{\Σ}}}\left(\right|X\left(k\right)|*\frac{k}{N}*f_s)}{\underset{k=0}{\overset{N/2}{\mathrm{\Σ}}}\left|X\left(k\right)\right|}$ ，

By above formula, can try to achieve the corresponding MPF parameter of electromyographic signal sequence.

Process software requires muscle to do isometric contraction campaign, and actual RMS is compared with canonical parameter (choosing 1.0V), show that ratio value is as muscle health state estimation.Software also compares the actual MPF value in front and back simultaneously, and passing threshold relative method is obtained muscle fatigue index.

Embodiment

A male volunteers take below as example in detail technical solution of the present invention.

Gather the electromyographic signal of 4 muscle of 24 years old male volunteers lower limb shank peroneus longus (stretching ankle), tibia longue (foot is bent upwards), extensor digitorum longus (toe and ankle flexion) and peroneus brevis (ankle is crooked and stretch), and by the surface electromyogram signal processing method in invention, complete lower limb ankle normal attitude, protract, rear quarters, outward turning and the identification of 5 groups of athletic performances of inward turning and the prediction of active force.The result of action recognition drives OpenGL fantasy sport model to complete corresponding virtual acting.Concrete experimentation is as follows:

1) connect myoelectricity capture card and process software

The D15 interface that connects electromyographic signal conducting wire 100 and the multi-channel surface myoelectric signal acquiring system 200 based on Wi-Fi; Open multi-channel surface myoelectric signal acquiring system 200 power supplies based on Wi-Fi, use PC wireless network card scanning current network, and be connected to the wireless network that network name is called " marvel ", after successful connection, open electromyographic signal process software, and in internetwork connection mode, specify correct communication protocol, IP address and port numbers.It is 192.168.10.10 that this example acquiescence is used udp protocol, capture card IP address, and port numbers is 8080.Click after " connection button ", myoelectricity capture card has been connected with the network of process software.

2) electrode slice is connected with conducting wire

Use alcohol wipe volunteer lower limb shank surface, with cleaning skin, guarantee effectively adhesion of electrode slice.In peroneus longus, tibia longue, extensor digitorum longus and the peroneus brevis of volunteer's lower limb shank, paste respectively two plate electrodes, and on ankle joint, paste a plate electrode as right lower limb drive electrode.Drive leading of passage to emit successively the patch electrode on peroneus longus, tibia longue, extensor digitorum longus, peroneus brevis and the ankle joint with shank to be connected the CH1 of electromyographic signal conducting wire, CH2, CH3, CH4 and right lower limb.

3) electromyographic signal process software sample collection

This example sample collection process need completes ankle normal attitude, protracts, produce 0N, 10N, 20N, 30N and 40N 5 groups of active force test samples of active force over the ground under rear quarters, outward turning and 5 groups of athletic performances of inward turning and ankle normal attitude, the size of active force is recorded by power level sensor, and amounting to sample size is 10 groups.Click electromyographic signal process software and " start to test " button, software sends " sign on " to electromyographic signal collection card, and starts first group of test sample collection.Gatherer process is completed by electromyographic signal collection card, and real-time Transmission is to PC end electromyographic signal process software.Electromyographic signal collection card single channel sampling rate is 4KHz, and sampled point word length is 2 bytes, and every group of test sample sampling time is 5s, and therefore the data total amount of every group of test sample 4 passages is 4K*2*5*4 byte.Volunteer remains the action of experiment sample defined in sample collection process, between every group of experiment sample action, suitably loosen rest, to reduce the tired impact that training sample is produced, and complete according to this 10 groups of samples experiments of the series of experiments sample regulations such as the protracting of ankle, rear quarters.In gatherer process, electromyographic signal process software is deposited in experiment sample packet under software work at present catalogue, and in sample collection process, software is by information such as simultaneous display acquired signal waveform and frequency spectrums.

4) training sample feature extraction and generation SVM model

Click electromyographic signal process software " pattern recognition " button, myoelectricity process software reads the sample data of sample collection stage storage successively, then respectively 5 groups of sample action data and 5 groups of active force sample datas are carried out to AR coefficient and RMS feature extraction, obtain training sample characteristic vector ε and μ.If ε is the characteristic vector that AR coefficient extracts, μ is the characteristic vector that RMS extracts.

Training software calculates respectively the coefficient of its 4 rank AR model to 4 channel datas of 5 groups of athletic performance samples, finally obtain 5 stack features vectors and be respectively ε 1={ a ₁₁, a ₁₂, a ₁₃, a ₁₄..., ε 5={ a ₄₁, a ₄₂, a ₄₃, a ₄₄, a wherein _ijthe AR coefficient that represents the test sample that Chi passage gathers, ε can be expressed as that ε={ ε 1, and ε 2, and ε 3, and ε 4, ε 5}; Training software calculates respectively its RMS to 4 channel datas of 5 groups of active force samples, and finally obtain 5 groups of RMS values and be respectively μ 1 ..., μ 5, μ can be expressed as μ=μ 1 ... μ 5}.

The characteristic vector of training sample is formatted as LibSVM storehouse and trains desired form, finally uses the training of the complete paired samples of svm_train function and generates SVM model.After pattern recognition has been trained, software will eject " pattern drill completes, and has generated SVM model " prompted dialog frame, represent that training process completes.

5) line model identification

Line model cognitive phase, software continuous acquisition experiment sample data, and sampling myoelectricity signals sequence is done to truncation, blocking sequence length is 1000 bytes.Process software calculating is blocked AR coefficient and the RMS value of discrete series and is sent into respectively C-SVM pattern classifier and ε-SVR active force forecast model.C-SVM and ε-SVR algorithm are all used svm_predict function to complete the prediction to the identification of electromyographic signal athletic performance and active force.In this example volunteer by motion ankle, do protract, appropriate change movement velocity and amount of force when the action such as rear quarters, outward turning and inward turning, process software will dynamically show the type of action of current line model identification and the size of predicting function power, and export result to dummy model, complete virtual emulation process.Line model identification be take 250ms(sequence length as 1000Byte, and sample frequency is 4KByte) for the cycle, electromyographic signal is processed.

6) OpenGL virtual human body movement model

OpenGL motion model will predict the outcome according to the type of action of pattern recognition and active force, adjusts in real time the direction of motion and the speed of virtual human body lower limb ankle.Process software is adjusted the visual angle of 3DS model according to current type of sports, and step model is carried out to translation or rotation, and the speed of translation or rotation will be adjusted according to the size of predicting function power is linear.Process software reconfigures up-to-date 3DS hierarchical model, again plays up, to reach the athletic performance of true reappearance human body lower limbs ankle in window.

7) electromyographic signal conventional parameter calculates

Process software will calculate in real time time domain and the frequency domain parameters such as mean-square value (RMS) and frequency of average power (MPF) in electromyographic signal collection process, and more according to the selected muscle types of software interface, selectivity shows result of calculation, and compares and provide the relative fatigue of corresponding muscle and health index with standard or information before.

In above experimental procedure, process software is for reference by the spectrum information that shows in real time electromyographic signal time domain waveform and correspondence in analysis result.In order to verify effectiveness of the present invention, in this example, require volunteer lower limb ankle to select in 5 groups of actions arbitrarily group random motion and change amount of force arbitrarily among 0N ~ 40N, in experimentation, the actual motion state of ankle is compared with dummy model, variation in the primary part observation dummy model direction of motion and speed, the actual motion situation of its result and ankle is basically identical.In addition, for the effect that further checking is invented, require volunteer to select at random athletic performance and active force combination in example, every group of action repeats to add up more than 50 times, draws experiment statistics result.Through experiment statistics, process software of the present invention at identification lower limb ankle normal attitude, protract, during the action such as rear quarters, outward turning and inward turning discrimination up to 91.2%, within the scope of 0N ~ 40N, predicting function power time error is less than 9.8%, human body imitate cartoon effect and ankle actual motion situation are basically identical, and result can be applicable to the application controls such as ankle healing robot.

Finally, process software experimental result of the present invention has also been verified the effectiveness that the multi-channel surface myoelectric signal acquiring system 200 based on Wi-Fi gathers, and meets single channel 4KHz sampling rate, and 12bit sampling precision can meet electromyographic signal collection requirement.The size of lower extremity movement direction and active force can effectively be identified or predict to its processing method.In addition, the multi-channel surface myoelectric signal acquiring system based on Wi-Fi art designs can convenient be processed electromyographic signal in the terminals such as PC, mobile phone, panel computer, has expanded scope and the value of human body surface myoelectric signal application.

Claims (3)

1. a processing method for the multi-channel surface myoelectric signal acquiring system institute collection surface electromyographic signal based on Wi-Fi, the described multi-channel surface myoelectric signal acquiring system based on Wi-Fi comprises:

The surface myoelectric utmost point, the primary signal that contains surface electromyogram signal for obtaining skin surface;

Preamplifier, is connected with the described surface myoelectric utmost point, for amplifying described primary signal, and the primary signal after being amplified;

Wave trap, is connected with described preamplifier, for removing the electromagnetic interference signal of primary signal after described amplification;

Band filter, is connected with described wave trap, for eliminating primary signal after the described amplification signal except surface electromyogram signal, obtains the first processing signals;

Final amplifier, is connected with described band filter, for to described the first processing signals gain and level lifting, obtains the second processing signals;

Analog-to-digital conversion module, is connected with described final amplifier, for to described the second processing signals analog digital conversion, samples and obtains original discrete electromyographic signal sequence;

Wireless communication module, is connected with described analog-to-digital conversion module, for described original discrete electromyographic signal sequence is transferred to processing platform by Wi-Fi wireless network;

The processing method that it is characterized in that the described multi-channel surface myoelectric signal acquiring system institute collection surface electromyographic signal based on Wi-Fi comprises:

The AR coefficient of the original discrete electromyographic signal sequence that described processing platform extraction human motion produces at skin surface place, as human motion direction character, forms the characteristic vector μ of action to be identified; By the training sample ν of the common anabolic action power of the sample forecast model of the root-mean-square value of original discrete electromyographic signal and human motion actual force size;

Described characteristic vector μ and training sample ν are inputted respectively to the size that C-SVM and ε-SVR obtain human motion direction and motion active force;

According to described human motion direction and the amount of force obtaining, adjust in real time lower extremity movement direction and the speed of the human body dummy model based on OpenGL, with the kinestate of true reappearance human body lower limbs.

2. the processing method of the multi-channel surface myoelectric signal acquiring system institute collection surface electromyographic signal based on Wi-Fi according to claim 1, is characterized in that:

Described human motion direction character is by using Yule-Walker equation solution AR model to calculate the AR coefficient of original discrete electromyographic signal sequence.

3. the processing method of the multi-channel surface myoelectric signal acquiring system institute collection surface electromyographic signal based on Wi-Fi according to claim 1, is characterized in that:

The RMS of gauging surface electromyographic signal, MPF parameter, respectively as assessment muscle fatigue, health status index, wherein,

Described RMS calculates according to following formula:

In formula, the discrete electromyographic signal sequence of z (i) expression single channel z (i) | and i=1,2 ..., N}, N is the discrete electromyographic signal sequence length of single channel;

Described MPF calculates according to following formula:

The total length that in formula, N is Fourier transformation, fs is electromyographic signal sample frequency;

During MPF calculation of parameter, first described processing platform carries out FFT conversion to original electromyographic signal, to electromyographic signal sequence, is x (n), and alternative approach is:

The X obtaining (k) sequence is corresponding with actual frequency, obtains the spectrum distribution of signal by corresponding conversion.