CN105956547A - Decomposition method based on array surface electromyogram signal smoothing - Google Patents

Abstract

The invention provides a decomposition method based on array surface electromyogram signal smoothing. The method comprises the steps of: firstly, utilizing a self-adaptive time length smoothing method to carry out pre-processing on array surface electromyogram signals, and increasing signal characteristics; secondly, utilizing a convolution kernel compensation algorithm to extract motion unit granting sequences, and correcting the sequences according to a granting rule; and finally, arranging and optimizing all obtained granting sequences, and deleting repeated sequences. According to the invention, the signal characteristics are increased by the self-adaptive time length smoothing method before decomposition, so that a relatively ideal effect is obtained especially for the surface electromyogram signals with large interference. In addition, the method is easy to realize.

Description

Based on the decomposition method that array surface electromyogram signal is smooth

Technical field

The present invention relates to a kind of decomposition method smooth based on array surface electromyogram signal.

Background technology

Surface electromyogram signal (surface EMG, sEMG) is to utilize surface electrode to detect electromyographic signal from body surface, with needle electrode Electromyographic signal (Needle EMG, NEMG) is compared, and it has feature non-invasive, easy to patients, and therefore application prospect is wide Wealthy.Experiment shows, utilizes array sEMG can improve the verification and measurement ratio of moving cell (MU), particularly improves small magnitude motion The detection of unit activity current potential (MUAP) and recognition effect.Clinically, can more fully be understood by array sEMG Nervimuscular functional status, differentiates neurogenic and muscle-derived disease, it is judged that the position of nerve injury, degree and recovery, And rehabilitation medicine and sports medical science are also had significant by the detection analysis of array sEMG signal.

Array sEMG is decomposed the moving cell granting sequence substantially comprised sEMG and is classified, and at present, sEMG classifies Method mainly has: K means clustering algorithm, template matching method, artificial neural network (ANN) algorithm, real time linear aliasing blind signal Separation algorithm, independent element divide folding (ICA), convolution kernel backoff algorithm etc..K-means clustering algorithm needs to specify the classification number of cluster, And the priori of the unit granting that lacks exercise in electromyographic signal, it is difficult to classification is specified accurately.Template matching method due to Template obtains difficulty, applies limited.ANN can solve containing more superposed waveform situations and preferably eliminate when low signal-to-noise ratio absolutely To error, but, after ANN method is once trained, network just immobilizes, when the shape of template changes, and nerve net Network also needs re-training, so robustness is bad.ICA is a kind of fanaticism decomposition technique, it is assumed that constitute each of electromyographic signal It is separate that moving cell provides sequence, then signal decomposition is become some separate compositions.Convolution kernel backoff algorithm method Being a kind of blind signal decomposition method, it is more satisfactory that the method has been verified effect.The signal to noise ratio of array sEMG is relatively low, MUAP ripple The superposition degree that the variability of shape is strong and mutual is relatively big, and this is to cause it to decompose the main cause of difficulty.As a whole, array The research of formula Decomposition Surface EMG also in the exploratory stage, is one of the difficult point of myoelectricity research field.

Summary of the invention

In view of the above problems, it is an object of the invention to provide a kind of decomposition method smooth based on array surface electromyogram signal.

For achieving the above object, for array surface electromyogram signal, propose first by primary signal pretreatment, improve signal quality, Convolution kernel backoff algorithm is used to decompose again.In this process, not solution matrix, obtain muscle by convolution kernel compensation method and transport The number of moving cell and granting sequence.Owing to the method has carried out pretreatment to primary signal, enhance signal characteristic, relative to Other method, the method has the advantage that sEMG Decomposition Accuracy is high.

The invention discloses a kind of decomposition method smooth based on array surface electromyogram signal, it is characterized in that comprising the following steps:

Step one: filter array surface electromyogram signal, weakens interference；

Step 2: use self adaptation duration smoothing method to filtered surface electromyogram signal pretreatment, each channel signal is put down Sliding enhancing signal feature, obtains signal S, and method is as follows:

1) surface electromyogram signal being divided into time span is T_fThe signal of length, obtains k segment signal, according to moving cell granting Characteristic, desirable 20ms≤T_f≤60ms；

2) to kth segment signal S_k(k is 1,2,3 ...), find minima S_{k_min}With maximum S_{k_max}, calculate the difference between value Value V_k=S_{k_max}-S_{k_min}；

3) so K section obtains K difference V₁, V₂, Λ, V_k, obtain sliding window time span T of i-th section of self adaptation duration_i:

$T_i=L_b+\frac{max(V_1,V_2,\mathrm{...},V_k)}{V_i}{L}_e$

Wherein L_b, L_eIt is the constant of design, max (V₁,V₂,...,V_k) represent the maximum in K difference.

4) each section is used the method that sliding window is average, calculates every section surface electromyographic signal, and record the initial of every segment signal Point and final point value, it is assumed that starting point and the maximal end point of the i-th segment signal are designated as Y_isAnd Y_ie；

5) the maximal end point Y of the i-th-1 segment signal is compared_(i-1)eStarting point Y of the i-th segment signal_isIf, | Y_(i-1)e-Y_is| (N is to set to≤N Constant), then 2 are joined directly together；Otherwise average with the sliding window of length 5ms on 2 o'clock, it is achieved 2 smooth connections, Obtain smooth rear surface electromyographic signal S eventually.

Step 3: use convolution kernel backoff algorithm that surface electromyogram signal S is extracted and provide the moment；

Step 4: be modified providing the moment, supplements and the granting moment of deletion error, obtains one and provides moment sequence；

Step 5: repeat step 3-----step 4, arrange cycle-index, extracts multiple granting moment sequence；

Step 6: to all granting sequence classified finishings, delete the granting sequence vector repeated, optimum results.

The technical measures optimized also include:

Above-mentioned convolution kernel backoff algorithm is that the correlation extraction utilizing sEMG signal provides the moment, and cross-correlation matrix is expressed as:

C=E (S (n) S^T(n))

Wherein n is sampling instant, and S (n) is the array signal of the n-th sampling instant, S^TN () is the array letter of the n-th sampling instant Number transposition, E () is number sequence expectation.

N' moving cell granting sequence table is shown as sometime:

ξ (n')=S^T(n')C^-1S(n')

Wherein C^-1The inverse matrix of array signal cross-correlation matrix.

Compared with prior art, a kind of decomposition method smooth based on array surface electromyogram signal of the present invention, due to interference relatively Big sEMG signal waveform distortion is serious, and in order to ensure signal characteristic and promotion signal quality, the present invention uses self adaptation duration Smooth sEMG signal, and split time adjustable length, the time is the shortest, and effect is the best, but the time of calculating is the longest, time concrete Between length determine according to practical situation.The convolution kernel backoff algorithm that the present invention uses need not calculate moving cell and provides sequence and battle array Hybrid matrix between row sEMG signal, greatly reduces the calculating time, improves efficiency, easy to use.

Accompanying drawing explanation

Fig. 1 is flow chart of the present invention.

Detailed description of the invention

Being described in further detail the present invention below in conjunction with accompanying drawing 1, those skilled in the art can be by disclosed by this specification Hold and realize easily.

The invention discloses a kind of decomposition method smooth based on array surface electromyogram signal, comprise the following steps:

Step one: filter array surface electromyogram signal, weakens interference；

Step 2: use self adaptation duration smoothing method to filtered surface electromyogram signal pretreatment, each channel signal is put down Sliding enhancing signal feature, obtains signal S, and method is as follows:

1) surface electromyogram signal being divided into time span is T_fThe signal of length, obtains k segment signal, according to moving cell granting Characteristic, usual moving cell granting frequency is in the range of 10Hz-50Hz, so desirable T_f=30ms, final stage signal length Allow less than 30ms；

2) to kth segment signal S_k(k is 1,2,3 ...), find minima S_{k_min}With maximum S_{k_max}, calculate the difference between value Value V_k=S_{k_max}-S_{k_min}；

3) so K section obtains K difference V₁, V₂, Λ, V_k, obtain sliding window time span T of i-th section of self adaptation duration_i:

$T_i=L_b+\frac{max(V_1,V_2,\mathrm{...},V_k)}{V_i}{L}_e$

Wherein L_b, L_eIt is the constant of design, max (V₁,V₂,...,V_k) represent the maximum in K difference.L_b, L_eTake representative value L_b=5ms, L_e=1ms；

4) each section is used the method that sliding window is average, calculates every section surface electromyographic signal, and record the initial of every segment signal Point and final point value, it is assumed that starting point and the maximal end point of the i-th segment signal are designated as Y_isAnd Y_ie；

5) the maximal end point Y of the i-th-1 segment signal is compared_(i-1)eStarting point Y of the i-th segment signal_isIf, | Y_(i-1)e-Y_is| (N takes≤N N=40 μ V), then 2 are joined directly together；Otherwise average with the sliding window of length 5ms on 2 o'clock, it is achieved 2 smooth connections, Obtain smooth rear surface electromyographic signal S eventually.

Step 3: utilize convolution kernel backoff algorithm to calculate and provide sequence.Convolution kernel backoff algorithm is to utilize array sEMG signal each The dependency of channel signal, calculates and provides sequence.Detailed process is: first computing array sEMG signal cross-correlation matrix and Cross-correlation matrix inverse matrix, cross-correlation matrix is expressed as:

C=E (S (n) S^T(n))

Wherein n is sampling instant, and S (n) is the array signal of the n-th sampling instant, S^TN () is the array letter of the n-th sampling instant Number transposition, E () is number sequence expectation.Calculate inverse matrix C of cross-correlation matrix^-1, i.e.

C^-1=[E (S (n) S^T(n))]^-1

Then sampling instant n takes the intermediate value of sEMG signal energy, and energy calculates according to the following formula:

Δ=S^T(n)C^-1S(n)

Take the moment n corresponding to energy intermediate value Δ₀.Finally utilize equation below be calculated moving cell provide sequence:

ξ(n₀)=S^T(n₀)C^-1S(n₀)

Step 4: to above-mentioned granting moment ξ (n₀) be modified, supplement and the granting moment of deletion error.Due to moving cell Granting frequency is 10-50Hz, so in the range of providing time at intervals time 20ms-100ms, there is interval to providing moment sequence Provide the moment less than 20ms to delete, should supplementing according to frequency of granting moment is lacked for more than 100ms interval and provides the moment.

Step 5: repeating step 3-----step 4, arrange cycle-index, usual cycle-index can be set to 500 times, extracts multiple Provide moment sequence；

Step 6: to all granting sequence classified finishings, delete the granting sequence vector repeated, optimum results.SEMG has extracted Delete the granting sequence of repetition after one-tenth, reject and provide frequency irrational granting sequence, optimum results.

Claims (1)

1., based on the decomposition method that array surface electromyogram signal is smooth, it is characterized in that comprising the following steps:

Step one: filter array surface electromyogram signal, weakens interference；

Step 2: use self adaptation duration smoothing method to filtered surface electromyogram signal pretreatment, each channel signal is put down Sliding enhancing signal feature, obtains signal S, and method is as follows:

1) surface electromyogram signal being divided into time span is T_fThe signal of length, obtains k segment signal, according to moving cell granting Characteristic, desirable 20ms≤T_f≤60ms；

2) to kth segment signal S_k(k is 1,2,3 ...), find minima S_{k_min}With maximum S_{k_max}, calculate the difference between value Value V_k=S_{k_max}-S_{k_min}；

3) so K section obtains K difference V₁, V₂, Λ, V_k, obtain sliding window time span T of i-th section of self adaptation duration_i:

$T_i=L_b+\frac{max(V_1,V_2,\mathrm{...},V_k)}{V_i}{L}_e$

Wherein L_b, L_eIt is the constant of design, max (V₁,V₂,...,V_k) represent the maximum in K difference；

4) each section is used the method that sliding window is average, calculates every section surface electromyographic signal, and record the initial of every segment signal Point and final point value, it is assumed that starting point and the maximal end point of the i-th segment signal are designated as Y_isAnd Y_ie；

5) the maximal end point Y of the i-th-1 segment signal is compared_(i-1)eStarting point Y of the i-th segment signal_isIf, | Y_(i-1)e-Y_is| (N is to set to≤N Constant), then 2 are joined directly together；Otherwise average with the sliding window of length 5ms on 2 o'clock, it is achieved 2 smooth connections, Obtain smooth rear surface electromyographic signal S eventually；

Step 3: use convolution kernel backoff algorithm that surface electromyogram signal S is extracted and provide the moment；

Step 4: be modified providing the moment, supplements and the granting moment of deletion error, obtains one and provides moment sequence；

Step 5: then repeat step 3-----step 4, arrange cycle-index, extracts multiple granting moment sequence；

Step 6: to all granting sequence classified finishings, delete the granting sequence vector repeated, optimum results.