CN106293057A - Gesture identification method based on BP neutral net - Google Patents
Gesture identification method based on BP neutral net Download PDFInfo
- Publication number
- CN106293057A CN106293057A CN201610574817.9A CN201610574817A CN106293057A CN 106293057 A CN106293057 A CN 106293057A CN 201610574817 A CN201610574817 A CN 201610574817A CN 106293057 A CN106293057 A CN 106293057A
- Authority
- CN
- China
- Prior art keywords
- electromyographic signal
- normalized
- sample
- gesture
- neural network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F3/00—Input arrangements for transferring data to be processed into a form capable of being handled by the computer; Output arrangements for transferring data from processing unit to output unit, e.g. interface arrangements
- G06F3/01—Input arrangements or combined input and output arrangements for interaction between user and computer
- G06F3/011—Arrangements for interaction with the human body, e.g. for user immersion in virtual reality
- G06F3/015—Input arrangements based on nervous system activity detection, e.g. brain waves [EEG] detection, electromyograms [EMG] detection, electrodermal response detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2203/00—Indexing scheme relating to G06F3/00 - G06F3/048
- G06F2203/01—Indexing scheme relating to G06F3/01
- G06F2203/011—Emotion or mood input determined on the basis of sensed human body parameters such as pulse, heart rate or beat, temperature of skin, facial expressions, iris, voice pitch, brain activity patterns
Abstract
The present invention relates to a kind of gesture identification method based on BP neutral net.The method includes: gather the electromyographic signal that the multiple gesture motion of multiple sample produces;Electromyographic signal is normalized;Electromyographic signal after normalized is extracted multiple eigenvalues composition eigenvalue matrix;Eigenvalue matrix utilize BP neural network algorithm carry out model training to form BP neural network model;BP neural network model is stored in electromyographic signal collection equipment to carry out gesture identification.The present invention is in signal analysis and pattern recognition, and the method for normalizing of improvement effectively eliminates the difference between different crowd electromyographic signal, uses BP neural network classifier to carry out pattern recognition, greatly reduces false recognition rate while improving gesture identification rate.
Description
Technical field
The present invention relates to the pretreatment of electromyographic signal, feature extraction and algorithm for pattern recognition in gesture identification solution
Design, particularly relates to a kind of based on BP neutral net gesture identification method.
Background technology
Along with developing rapidly of microelectric technique, sensor technology and computer technology, handheld mobile device, Wearable
Equipment and microcomputer are gradually the most universal in people's daily life.But owing to using scene and the limit of mini-plant
System, traditional human-computer interaction device, the such as equipment such as keyboard, mouse can not meet the demand of people.Transportable little
Type gesture identification equipment is proposed as novel human-computer interaction device.
The muscle signal of telecommunication, since 1945 are used in control field, experienced by the development of last 100 years, the most studied
It is applied to medical diagnosis and bio-mechanical field.Along with biomedical technology, the development of artificial intelligence technology, use electromyographic signal
The method carrying out gesture identification is suggested and constantly explores.The generation of human biological electricity is the most non-linear, non-stationary process,
Signal amplitude is faint, and such as electromyographic signal amplitude is in μ V~mV level, and interindividual variation is bigger.How to eliminate individual variation,
Make gesture identification scheme have versatility, be just particularly important.
Summary of the invention
Therefore, for solving technological deficiency and the deficiency that prior art exists, the present invention provides a kind of based on the normalizing improved
The BP neutral net gesture identification method of change method.
In pattern recognition and signal analysis, neural network classifier has description input and output linearity and non-linear reflects
The advantage penetrating relation, has powerful learning capacity simultaneously.The present invention carries out normalizing to the muscle signal of telecommunication sample of Different Individual
Change, then extract many temporal signatures and be applied to neural network classifier and be analyzed, be identified gesture with this.Specific as follows
Step:
Gather the electromyographic signal that the multiple gesture motion of multiple sample produces;
Described electromyographic signal is normalized;
Described electromyographic signal after normalized is extracted multiple eigenvalues composition eigenvalue matrix;
Described eigenvalue matrix utilize BP neural network algorithm carry out model training to form BP neural network model;
Described BP neural network model is stored in described electromyographic signal collection equipment to carry out gesture identification.
In one embodiment of the invention, described electromyographic signal is normalized, including:
The electromyographic signal of the many gesture motion of multichannel of same sample is that (i, j), (i j) is discrete-time series, i to X to X
Represent the sequence number of the acquisition channel of described electromyographic signal collection equipment, and i=1~N, j express time serial number, use outer waving
Maximum | X | of the absolute value of gesture action N number of passage electromyographic signalmaxAs normalized reference standard, to described electromyographic signal
Carrying out described normalized, described normalized formula is:Wherein, (i, after j) being normalization for x
The discrete-time series of the electromyographic signal of the many gesture motion of multichannel.
In one embodiment of the invention, described electromyographic signal is normalized, including:
For electromyographic signal X of the many gesture motion of multichannel of same sample, (i, j), (i j) is discrete time sequence to X
Row, i represents the sequence number of the acquisition channel of described electromyographic signal collection equipment, and i=1~N, j express time serial number, outside utilization
The maximum of the absolute value of the same channel difference values of gesture of waving | X (n)max-X(n)min| it is normalized;At described normalization
Reason formula is:N represents the sequence number of the acquisition channel of described electromyographic signal collection equipment,
(i j) is the discrete-time series after normalization to x.
In one embodiment of the invention, described electromyographic signal is normalized, including:
For electromyographic signal X of the many gesture motion of multichannel of same sample, (i, j), (i j) is discrete time sequence to X
Row, i represents the sequence number of the acquisition channel of described electromyographic signal collection equipment, and i=1~N, j express time serial number, utilizes institute
The root mean square maximum stating outer gesture N channel of waving is normalized;Described normalized formula is:
Wherein, n represents the sequence number of the acquisition channel of described electromyographic signal collection equipment, X (n,
J) it is the discrete-time series of the n-th passage, j express time sequence;(i j) is the discrete-time series after normalization to x.
In one embodiment of the invention, described electromyographic signal is normalized, including:
For electromyographic signal X of the many gesture motion of multichannel of same sample, (i j) deducts same sample relaxation state
(i, j), wherein, (i, j), (i j) is discrete time sequence to X ' to X to electromyographic signal X ' after the electromyographic signal formation process of respective channel
Row, i represents the sequence number of the acquisition channel of described electromyographic signal collection equipment, and i=1~N, j express time serial number;Outside utilization
Gesture of waving is at the maximum of the absolute value of same acquisition channel difference | X ' (n)max-X′(n)min| it is normalized;Described
Normalized formula is:N represents that the collection of described electromyographic signal collection equipment is led to
The sequence number in road, (i j) is the discrete-time series after normalization to x.
In one embodiment of the invention, the described electromyographic signal after normalized is extracted multiple eigenvalue clusters
Become eigenvalue matrix, including:
For different sample difference gestures, from the plurality of spy of N number of acquisition channel of described electromyographic signal collection equipment
Levy the feature combination as corresponding gesture of at least one feature of middle selection, to form an eigenvalue matrix;
The plurality of eigenvalue includes absolute mean between the absolute mean (MAX) of described electromyographic signal, N number of acquisition channel
Ratio (R_MAV), root-mean-square (RMS), root-mean-square ratio (R_RMS), zero crossing (ZC), waveform length (WL) and symbol slope
Rate of change (SSC).
In one embodiment of the invention, BP neural network algorithm is utilized to carry out model training described eigenvalue matrix
To form BP neural network model, including:
Step one, random number initialize weight matrix;
Step 2, the plurality of eigenmatrix is normalized;Normalization is referenced as multiple same spy of sample N channel
The maximum difference levied;
Step 3, determine the nodes k of single hidden layer;
Step 4, sequentially input P learning sample, and assume that being currently entered sample is pth.
Step 5, calculate the output of each layer successively;Wherein, the input algorithm of hidden layer is:Output
Layer introduces nonlinear function:Wherein, p is the sample number being currently entered, wjiIt is i-th god
Through the weights of unit to jth neuron, netpjIt is the input of neuron j, opjIt is the output of neuron j;
Step 6, output layer neuron are output as:
Step 7, the error performance target function of pth sample:In formula, tplIt is neuron l
Target output;
The sample number that step 8, recording learning are crossed, if p < P, forwards step 5 to, if p=P, forwards step 9 to;
Step 9, weights according to the modified weight each layer of formula correction;The connection weight w of output layer and hidden layerljLearning algorithm:The connection weight w of hidden layer and input layerjiLearning algorithm:
In formula, n is iterations, and η is learning rate, η ∈ [0,1], uses Variable Step Algorithm to learn;
Step 10, the weights of described each layer being added factor of momentum α, weights now are:
wlj(n+1)=wlj(n)+△wlj+α(wlj(n+1)-wlj(n));
wji(n+1)=wji(n)+△wji+α(wji(n+1)-wji(n));Wherein, the value α ∈ [0,1] of factor of momentum;
Step 11, recalculate the output of each layer according to new weights, if each sample standard deviation meets output and target
The difference of output is less than number of threshold values, or has reached the study number of times preset, then stop.Otherwise forward step 5 to.
In one embodiment of the invention, described BP neural network model is stored in electromyographic signal collection equipment with
Carry out gesture identification, including:
Electromyographic signal is acquired by the N number of acquisition channel utilizing described electromyographic signal collection equipment;
The electromyographic signal collected is carried out in time domain feature extraction corresponding to obtain described BP neural network model
Eigenvalue matrix;
The described eigenvalue matrix described BP neural network model of input is identified.
In one embodiment of the invention, myoelectricity is believed by the N number of acquisition channel utilizing described electromyographic signal collection equipment
Number being acquired set parameter includes:
Arranging fixing data and processing length of window is L, and moving step length is L, and sample frequency F meets condition: L/F < 0.3s.
Above-described embodiment, to electromyographic signal by data acquisition, normalized, feature extraction, feature selection, and utilizes
Different gesture characteristic of correspondence value matrixs are trained and determine the eigenvalue matrix of optimum, i.e. at PC by BP neural network algorithm
BP neural network model is trained by end, is then stored in embedded system, which saves embedded system to model training
Time time loss and memory consumption.When expanding the training sample of model, Embedded service ability need not be considered simultaneously.Therefore
PC end can be sampled sample as much as possible, contains crowd's myoelectricity feature widely, makes the BP neural network model tool trained
There is more preferable versatility.
Accompanying drawing explanation
In order to the technical scheme of the present invention or prior art is more clearly described, embodiment or prior art will be retouched below
In stating, the required accompanying drawing used is briefly described.It should be evident that the accompanying drawing in describing below is some of the present invention
Embodiment, for those of ordinary skill in the art, on the premise of not paying creative work, it is also possible to attached according to these
Figure obtains other accompanying drawings.Below in conjunction with accompanying drawing, the detailed description of the invention of the present invention is described in detail.
The schematic flow sheet of a kind of based on BP neutral net the gesture identification method that Fig. 1 provides for the embodiment of the present invention;
The model schematic of a kind of BP neutral net that Fig. 2 provides for the embodiment of the present invention;
The algorithm flow chart of a kind of BP neutral net that Fig. 3 provides for the embodiment of the present invention;
The flow process signal of the another kind gesture identification method based on BP neutral net that Fig. 4 provides for the embodiment of the present invention
Figure.
Detailed description of the invention
For making the object, technical solutions and advantages of the present invention clearer, below in conjunction with the accompanying drawing of the present invention, to this
Bright technical scheme carries out clear, complete description.Obviously, described embodiment is a part of embodiment of the present invention, and not
It it is whole embodiments.Based on embodiments of the invention, those of ordinary skill in the art are not making creative work premise
Lower obtained every other embodiment, broadly falls into protection scope of the present invention.
Refer to the stream of a kind of based on BP neutral net the gesture identification method that Fig. 1, Fig. 1 provide for the embodiment of the present invention
Journey schematic diagram, the method can apply to include the fields such as electromyographic signal process, gesture identification, myoelectricity artifucial limb, bio-mechanical, especially
It relates to the normalization of electromyographic signal in gesture recognition system, feature extraction and algorithm for pattern recognition design, i.e. signal
Normalization, temporal signatures extraction and the application of sorting algorithm and optimization.Specifically, the method may include steps of:
Step a, gather the multiple gesture motion of multiple sample produce electromyographic signal;
Step b, electromyographic signal is normalized;
Step c, to after normalized electromyographic signal extract multiple eigenvalues composition eigenvalue matrix;
Step d, eigenvalue matrix utilize BP neural network algorithm carry out model training to form BP neural network model;
Step e, BP neural network model is stored in electromyographic signal collection equipment to carry out gesture identification.
Wherein, for step a, in order to ensure the accuracy to gesture identification, utilize big data thinking, choose no less than 50
Individual sample.It is to say, can 50 testers be tested, model is made to have versatility.
For step a, described electromyographic signal collection equipment is worn on the forearm of tester, and N number of acquisition channel is corresponding
It is attached at the various location of described appointed part.Wherein, the strange UIT of Bryant in electromyographic signal collection equipment e.g. Xi'an
The DTing equipment that skill Co., Ltd produces.
For step b, may include that
Step b1, the electromyographic signal of the many gesture motion of multichannel of same sample are that (i, j), (i, when being j) discrete for X for X
Between sequence, i represents the sequence number of the acquisition channel of described electromyographic signal collection equipment, and i=1~N, j express time serial number.Adopt
Maximum | X | with the absolute value of outer pendulum gesture motion N channel electromyographic signalmaxAs normalized reference standard.To described flesh
The signal of telecommunication carries out described normalized formula:Wherein, (i j) is the many handss of multichannel after normalization to x
The discrete-time series of the electromyographic signal of gesture action.
Alternatively, for step b, may include that
Step b2, for electromyographic signal X of the many gesture motion of multichannel of same sample, (i, j), (i is j) discrete to X
Time series, i represents the sequence number of the acquisition channel of described electromyographic signal collection equipment, and i=1~N, j express time serial number,
Utilize the maximum of the absolute value of the described outer same channel difference values of gesture of waving | X (n)max-X(n)min| it is normalized;Institute
Stating normalized formula is:N represents the collection of described electromyographic signal collection equipment
The sequence number of passage, (i j) is the discrete-time series after normalization to x.
Alternatively, for step b, may include that
Step b3, for electromyographic signal X of the many gesture motion of multichannel of same sample, (i, j), (i is j) discrete to X
Time series, i represents the sequence number of the acquisition channel of described electromyographic signal collection equipment, and i=1~N, j express time serial number,
The root mean square maximum utilizing described outer gesture N channel of waving is normalized;Described normalized formula is:
N represents passage.
Wherein, n represents the sequence number of the acquisition channel of described electromyographic signal collection equipment, and X (n, j)
It is the discrete-time series of the n-th passage, j express time sequence;(i j) is the discrete-time series after normalization to x.
Alternatively, for step b, may include that
Step b4, for electromyographic signal X of the many gesture motion of multichannel of same sample, (i j) deducts same sample and puts
(i, j), wherein, (i, j), (i is j) discrete to X ' to X to electromyographic signal X ' after the electromyographic signal formation process of pine state respective channel
Time series, i represents the sequence number of the acquisition channel of described electromyographic signal collection equipment, and i=1~N, j express time serial number;
Utilize the outer gesture maximum at the absolute value of same acquisition channel difference of waving | X ' (n)max-X′(n)min| it is normalized place
Reason;Described normalized formula is:N represents described electromyographic signal collection equipment
The sequence number of acquisition channel, (i j) is the discrete-time series after normalization to x.
For step c, including:
For different sample difference gestures, select at least one special from the plurality of feature of described N number of acquisition channel
Levy a feature combination as corresponding gesture, this feature combination form characteristic of correspondence value matrix.
The plurality of eigenvalue can be the most equal between the absolute mean (MAX) of described electromyographic signal, N number of acquisition channel
Ratio (R_MAV), root-mean-square (RMS), root-mean-square ratio (R_RMS), zero crossing (ZC), waveform length (WL) and the symbol of value are oblique
The combination in any that rate rate of change (SSC) etc. are constituted.
For step d, may include steps of::
Step d1, BP neural network algorithm is utilized to instruct the eigenvalue matrix extracted after described different normalizeds
Practice, and from M sample, determine the eigenvalue matrix that discrimination is the highest and false recognition rate is minimum, using correspondence method for normalizing as
Optimum method for normalizing.
Step d2, the electromyographic signal of different sample difference gestures uses described optimum method for normalizing process, carry
Taking the plurality of eigenvalue, combination in any forms multiple eigenvalue matrix, and the 50% of selected characteristic value matrix utilizes BP nerve net
Network algorithm carries out model training, and residue 50% is identified, and determines the eigenvalue matrix that discrimination is the highest and false recognition rate is minimum
Combine for best eigenvalue.
Step d3, described optimal characteristics value matrix is utilized BP neural network algorithm carry out model training with formed definition hands
The described BP neural network model of gesture.
It addition, step d may include steps of from BP neural network algorithm angle:
Step one, random number initialize weight matrix;
Step 2, eigenmatrix is normalized;Normalization is referenced as the maximum of multiple same feature of sample N channel
Difference;
Step 3, determine the nodes k of single hidden layer;
Rule of thumb formulaWherein a is input layer number, and b is output layer nodes,
Being 7 class gestures, σ is a constant.
When only one of which characteristic quantity, during 8 passage, k=4~14, uses root-mean-square characteristic test, k=10, i.e. 10 hidden layer joints
During point, discrimination is the highest, and false recognition rate is minimum;
When 2 characteristic quantities, during 8 passage, k=5~15;During 3 characteristic quantities, k=6~16;During 4 characteristic quantities, k=7~
17;During 5 characteristic quantities, k=7~17;During 6 characteristic quantities, k=8~18;During 7 characteristic quantities, k=8~18;By 1 feature group
Conjunction is tested, k=11, and i.e. during 11 hidden layer nodes, discrimination is the highest, and false recognition rate is minimum;
In sum, node in hidden layer is chosen as 11, and test result is optimal;
Step 4, sequentially input P learning sample, and assume that being currently entered sample is pth.
Step 5, calculate the output of each layer successively;Wherein, the input algorithm of hidden layer is:Output
Layer introduces nonlinear function:Wherein, p is the sample number being currently entered, wjiIt it is i-th
Neuron is to the weights of jth neuron, netpjIt is the input of neuron j, opjIt is the output of neuron j;
Step 6, output layer neuron are output as:
Step 7, the error performance target function of pth sample:In formula, tplIt it is neuron
The target output of l;
The sample number that step 8, recording learning are crossed, if p < P, forwards step 5 to, if p=P, forwards step 9 to;
Step 9, weights according to the modified weight each layer of formula correction;The connection weight w of output layer and hidden layerljLearning algorithm:The connection weight w of hidden layer and input layerjiLearning algorithm:
In formula, n is iterations, and η is learning rate, η ∈ [0,1], uses Variable Step Algorithm to learn;
Step 10, the weights of described each layer being added factor of momentum α, weights now are:
wlj(n+1)=wlj(n)+△wlj+α(wlj(n+1)-wlj(n));
wji(n+1)=wji(n)+△wji+α(wji(n+1)-wji(n));Wherein, the value α ∈ [0,1] of factor of momentum;
Step 11, recalculate the output of each layer according to new weights, if each sample standard deviation meets output and target
The difference of output is less than number of threshold values, or has reached the study number of times preset, then stop.Otherwise forward step 5 to.
The model schematic of a kind of BP neutral net that Fig. 2 provides for the embodiment of the present invention.Refer to Fig. 2, BP nerve net
Network is the multilayer feedforward sensing network using error back propagation (Back-propagation) algorithm.Generally by input layer, hidden
Form containing layer and output layer, the most totally interconnected.According to Kolmogorov theorem, there is a hidden layer (hidden node foot
More than enough) three layers of BP neutral net in closed set, any non-linear continuous function can be approached with arbitrary accuracy.Therefore the present invention adopts
BP neutral net with single hidden layer.
For step e, may include that
Step e1, utilize N number of acquisition channel of described electromyographic signal collection equipment that electromyographic signal is acquired;
Step e2, the electromyographic signal collected is carried out in time domain feature extraction to obtain described BP neural network model
Characteristic of correspondence value matrix;
Step e3, by described eigenvalue matrix input described BP neural network model be identified.
Wherein, step e1 gathers set parameter and includes: arranging fixing data and processing length of window is L, mobile
Step-length is L, and due to the real-time of man-machine interaction, people, when being more than 300ms time delay, just has delay sense, therefore in sampling
In the case of frequency is F, L/F should be less than 0.3, i.e. 0.3s.
The embodiment of the present invention, uses normalized technological means to solve due to individual variation, and electromyographic signal differs greatly, band
The problem that the gesture identification scheme come does not has versatility.Method for normalizing can use amplitude maximum or the width of signal
It is worth root mean square maximum and makees reference, but amplitude maximum is with reference to being probably interference, reduces the work of useful signal greatly
With, although and the situation that amplitude root-mean-square maximum can avoid interference, but either way have no idea to eliminate original letter
Offset information in number.Electromyographic signal carries out the sample space of feature extraction formation and there is information redundancy and be linearly inseparable,
Therefore, on the basis of normalization effective to signal, carry out feature selection and use the nonlinear sorting algorithm can be more preferable
Realize gesture identification.
Embodiment two
On the basis of above-described embodiment, the present embodiment possesses 8 acquisition channels with electromyographic signal collection equipment, have employed
4 kinds of methods are normalized, and extract multiple feature and include between the absolute mean (MAX) of electromyographic signal, N number of acquisition channel
The ratio (R_MAV) of absolute mean, root-mean-square (RMS), root-mean-square ratio (R_RMS), zero crossing (ZC), waveform length (WL) and
It is described in detail as a example by 7 features of symbol slope variation rate (SSC).
1, with 9 samples, 8 passages, 7 gestures, 10240 discrete-time serieses of each gesture of each sample, according to institute
State 4 kinds of methods to be normalized, the discrete-time series of each gesture composition 8*92160 size, extract absolute mean and waveform
Two characteristic quantity composition characteristic value matrixs of length, then carry out the model training of BP neutral net, then with 9 samples, each hands
The discrete-time series of gesture 8*230400 size is identified, and discrimination and false recognition rate are as shown in table 1 below.
The discrimination of 14 kinds of method for normalizing of table and false recognition rate
Therefore after pretreatment, the maximum of absolute difference is optimum method for normalizing.
2, multiple characteristic quantities are calculated as follows:
(1) the absolute mean feature of absolute mean (MAV): signal amplitude reflects the mean intensity of surface electromyogram signal,
Computing formula is as follows:
Wherein, N is total sampling number of this section surface electromyographic signal, XiAt ith sample point
The range value of surface electromyogram signal;
(2) ratio (R_MAV) of absolute mean, computing formula is as follows:
Wherein, C is port number, MAVCIt is the average of the C channel surface electromyographic signal
Value.Interchannel ratio eliminates the impact of dynamics;
(3) root-mean-square (RMS): root-mean-square is the virtual value having reacted a section surface electromyographic signal to a certain extent, meter
Calculation formula is as follows:
(4) root-mean-square ratio (R_RMS), formula is as follows:
Wherein, C is port number, RMSCIt is the average of the C channel surface electromyographic signal
Value;
(5) zero crossing (ZC);Formula is as follows:
In formula, θ is threshold value, and value is 0.025*
|Xmax-Xmin|。
(6) waveform length (WL);Formula is as follows:
(7) symbol slope variation rate (SSC);Formula is as follows:
3, discrete-time series calculating features described above value composition characteristic value matrix, row represents different types of feature, row
Represent sampled point;BP neutral net is to have the study of tutor, provides the target output matrix of classification, and row represents classification, and list is shown
Sampled point.Due to 7 features of 8 passage, characteristic dimension is relatively big, certainly exists redundancy between characteristic vector, and it is right to be therefore accomplished by
Feature selects, and eliminates redundancy.In 7 features, arbitrarily select 1,2,3 to carry out random combine, consider spy simultaneously
The order levied, altogetherPlant combination.
Eigenvalue matrix inputting BP neutral net carry out classifying and identifying, statistics discrimination and false recognition rate, storage is known
The training pattern that rate is not the highest and false recognition rate is minimum, to carry out follow-up test.Namely every kind in above-mentioned 259 kinds of combinations
Combination input BP neutral net is trained and identifies, statistics optimal characteristics combination.
In sum, specific case gesture identification method based on BP neutral net to the present invention used herein former
Reason and embodiment are set forth, and the explanation of above example is only intended to help to understand that the method for the present invention and core thereof are thought
Think;Simultaneously for one of ordinary skill in the art, according to the thought of the present invention, in specific embodiments and applications
All will change, in sum, this specification content should not be construed as limitation of the present invention, protection scope of the present invention
Should be as the criterion with appended claim.
Claims (9)
1. a gesture identification method based on BP neutral net, it is adaptable to the electromyographic signal collection equipment identification to gesture, its
It is characterised by, comprises the steps:
Gather the electromyographic signal that the multiple gesture motion of multiple sample produces;
Described electromyographic signal is normalized;
Described electromyographic signal after normalized is extracted multiple eigenvalues composition eigenvalue matrix;
Described eigenvalue matrix utilize BP neural network algorithm carry out model training to form BP neural network model;
Described BP neural network model is stored in described electromyographic signal collection equipment to carry out gesture identification.
2. the method for claim 1, it is characterised in that described electromyographic signal is normalized, including:
The electromyographic signal of the many gesture motion of multichannel of same sample is that (i, j), (i, j) is discrete-time series to X to X, and i represents
The sequence number of the acquisition channel of described electromyographic signal collection equipment, and i=1~N, j express time serial number, use outer gesture of waving to move
Make maximum | X | of the absolute value of N number of passage electromyographic signalmaxAs normalized reference standard, described electromyographic signal is carried out
Described normalized, described normalized formula is:Wherein, (i j) is manifold after normalization to x
The discrete-time series of the electromyographic signal of the many gesture motion in road.
3. method as claimed in claim 2, it is characterised in that described electromyographic signal is normalized, also includes:
For electromyographic signal X of the many gesture motion of multichannel of same sample, (i, j), (i j) is discrete-time series, i table to X
Show the sequence number of the acquisition channel of described electromyographic signal collection equipment, and i=1~N, j express time serial number, utilize outer gesture of waving
The maximum of the absolute value of same channel difference values | X (n)max-X(n)min| it is normalized;Described normalized formula
For:N represents the sequence number of the acquisition channel of described electromyographic signal collection equipment, x (i,
J) it is the discrete-time series after normalization.
4. method as claimed in claim 2, it is characterised in that described electromyographic signal is normalized, also includes:
For electromyographic signal X of the many gesture motion of multichannel of same sample, (i, j), (i j) is discrete-time series, i table to X
Show the sequence number of the acquisition channel of described electromyographic signal collection equipment, and i=1~N, j express time serial number, utilize described outer pendulum
The root mean square maximum of gesture N channel is normalized;Described normalized formula is:
Wherein, n represents the sequence number of the acquisition channel of described electromyographic signal collection equipment, and (n j) is X
The discrete-time series of the n-th passage, j express time sequence;(i j) is the discrete-time series after normalization to x.
5. method as claimed in claim 2, it is characterised in that described electromyographic signal is normalized, also includes:
For electromyographic signal X of the many gesture motion of multichannel of same sample, (i j) deducts same sample relaxation state corresponding
(i, j), wherein, (i, j), (i j) is discrete-time series, i to X ' to X to electromyographic signal X ' after the electromyographic signal formation process of passage
Represent the sequence number of the acquisition channel of described electromyographic signal collection equipment, and i=1~N, j express time serial number;Utilize outer waving
Gesture is at the maximum of the absolute value of same acquisition channel difference | X ' (n)max-X′(n)min| it is normalized;Described normalizing
Change processes formula:N represents the acquisition channel of described electromyographic signal collection equipment
Sequence number, (i j) is the discrete-time series after normalization to x.
6. the method for claim 1, it is characterised in that the described electromyographic signal after normalized is extracted multiple
Eigenvalue composition eigenvalue matrix, including:
For different sample difference gestures, from the plurality of feature of N number of acquisition channel of described electromyographic signal collection equipment
At least one feature is selected to combine as a feature of corresponding gesture, to form an eigenvalue matrix;
The plurality of eigenvalue includes the ratio of absolute mean between the absolute mean (MAX) of described electromyographic signal, N number of acquisition channel
Value (R_MAV), root-mean-square (RMS), root-mean-square ratio (R_RMS), zero crossing (ZC), waveform length (WL) and symbol slope variation
Rate (SSC).
7. the method for claim 1, it is characterised in that utilize BP neural network algorithm to carry out described eigenvalue matrix
Model training to form BP neural network model, including:
Step one, random number initialize weight matrix;
Step 2, the plurality of eigenmatrix is normalized;Normalization is referenced as multiple same feature of sample N channel
Maximum difference;
Step 3, determine the nodes k of single hidden layer;
Step 4, sequentially input P learning sample, and assume that being currently entered sample is pth.
Step 5, calculate the output of each layer successively;Wherein, the input algorithm of hidden layer is:Output layer draws
Entering nonlinear function is:Wherein, p is the sample number being currently entered, wjiIt is that i-th is neural
Unit arrives the weights of jth neuron, netpjIt is the input of neuron j, opjIt is the output of neuron j;
Step 6, output layer neuron are output as:
Step 7, the error performance target function of pth sample:In formula, tplIt is neuron l
Target exports;
The sample number that step 8, recording learning are crossed, if p < P, forwards step 5 to, if p=P, forwards step 9 to;
Step 9, weights according to the modified weight each layer of formula correction;The connection weight w of output layer and hidden layerljLearning algorithm:The connection weight w of hidden layer and input layerjiLearning algorithm:
In formula, n is iterations, and η is learning rate, η ∈ [0,1];
Step 10, the weights of described each layer being added factor of momentum α, weights now are:
wlj(n+1)=wlj(n)+Δwlj+α(wlj(n+1)-wlj(n));
wji(n+1)=wji(n)+Δwji+α(wji(n+1)-wji(n));Wherein, the value α ∈ [0,1] of factor of momentum;
Step 11, recalculate the output of each layer according to new weights, if each sample standard deviation meets output and target exports
Difference less than number of threshold values, or reached the study number of times preset, then stopped.Otherwise forward step 5 to.
8. the method for claim 1, it is characterised in that described BP neural network model is stored electromyographic signal collection
To carry out gesture identification in equipment, including:
Electromyographic signal is acquired by the N number of acquisition channel utilizing described electromyographic signal collection equipment;
The electromyographic signal collected is carried out in time domain feature extraction to obtain described BP neural network model characteristic of correspondence
Value matrix;
The described eigenvalue matrix described BP neural network model of input is identified.
9. method as claimed in claim 8, it is characterised in that utilize N number of acquisition channel of described electromyographic signal collection equipment
Electromyographic signal is acquired set parameter include:
Arranging fixing data and processing length of window is L, and moving step length is L, and sample frequency F meets condition: L/F < 0.3s.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610574817.9A CN106293057A (en) | 2016-07-20 | 2016-07-20 | Gesture identification method based on BP neutral net |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610574817.9A CN106293057A (en) | 2016-07-20 | 2016-07-20 | Gesture identification method based on BP neutral net |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106293057A true CN106293057A (en) | 2017-01-04 |
Family
ID=57651741
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610574817.9A Pending CN106293057A (en) | 2016-07-20 | 2016-07-20 | Gesture identification method based on BP neutral net |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106293057A (en) |
Cited By (25)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106682651A (en) * | 2017-02-10 | 2017-05-17 | 哈尔滨工业大学 | Monitoring method for cable insulation leather damage based on neural network |
CN107126303A (en) * | 2017-02-15 | 2017-09-05 | 上海术理智能科技有限公司 | A kind of upper and lower extremities exercising support method based on mobile phone A PP |
CN107169432A (en) * | 2017-05-09 | 2017-09-15 | 深圳市科迈爱康科技有限公司 | Biometric discrimination method, terminal and computer-readable recording medium based on myoelectricity |
CN108992066A (en) * | 2018-08-15 | 2018-12-14 | 东北大学 | Portable lower limb behavior pattern real-time identifying system and method based on electromyography signal |
CN109085918A (en) * | 2018-06-28 | 2018-12-25 | 天津大学 | Acupuncture needling manipulation training method based on myoelectricity |
CN109165587A (en) * | 2018-08-11 | 2019-01-08 | 石修英 | intelligent image information extraction method |
CN109190496A (en) * | 2018-08-09 | 2019-01-11 | 华南理工大学 | A kind of monocular static gesture identification method based on multi-feature fusion |
CN109412725A (en) * | 2018-10-15 | 2019-03-01 | 中国人民解放军战略支援部队信息工程大学 | The blind demodulation method of radio communication PCMA signal and device |
CN109613976A (en) * | 2018-11-14 | 2019-04-12 | 华东师范大学 | A kind of intelligent flexible pressure sensing hand language recognition device |
CN109635706A (en) * | 2018-12-04 | 2019-04-16 | 武汉灏存科技有限公司 | Gesture identification method, equipment, storage medium and device neural network based |
CN109976526A (en) * | 2019-03-27 | 2019-07-05 | 广东技术师范大学 | A kind of sign Language Recognition Method based on surface myoelectric sensor and nine axle sensors |
CN110378169A (en) * | 2018-04-12 | 2019-10-25 | 中移(杭州)信息技术有限公司 | The detection method and device in gesture section |
CN110413107A (en) * | 2019-06-21 | 2019-11-05 | 浙江科技学院 | Bionic mechanical hand interaction control method based on electromyography signal pattern-recognition and particle group optimizing |
CN111241982A (en) * | 2020-01-07 | 2020-06-05 | 金陵科技学院 | Robot gesture recognition method based on CAE-SVM |
CN111531537A (en) * | 2020-05-07 | 2020-08-14 | 金陵科技学院 | Mechanical arm control method based on multiple sensors |
CN111695446A (en) * | 2020-05-26 | 2020-09-22 | 浙江工业大学 | Gesture recognition method integrating sEMG and AUS |
CN111870242A (en) * | 2020-08-03 | 2020-11-03 | 南京邮电大学 | Intelligent gesture action generation method based on electromyographic signals |
CN111902847A (en) * | 2018-01-25 | 2020-11-06 | 脸谱科技有限责任公司 | Real-time processing of hand state representation model estimates |
CN111950460A (en) * | 2020-08-13 | 2020-11-17 | 电子科技大学 | Muscle strength self-adaptive stroke patient hand rehabilitation training action recognition method |
CN112101298A (en) * | 2020-10-15 | 2020-12-18 | 福州大学 | Gesture recognition system and method based on muscle electrical impedance signals |
CN112162041A (en) * | 2020-09-30 | 2021-01-01 | 陕西师范大学 | Method for identifying metal material based on Gaussian distribution of amplitude root mean square value |
CN112732090A (en) * | 2021-01-20 | 2021-04-30 | 福州大学 | Muscle cooperation-based user-independent real-time gesture recognition method |
CN113536954A (en) * | 2021-06-23 | 2021-10-22 | 厦门大学 | Gesture recognition method based on human body electromyographic signals |
CN113688802A (en) * | 2021-10-22 | 2021-11-23 | 季华实验室 | Gesture recognition method, device and equipment based on electromyographic signals and storage medium |
CN113986017A (en) * | 2021-12-27 | 2022-01-28 | 深圳市心流科技有限公司 | Myoelectric gesture template generation method and device and storage medium |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102073881A (en) * | 2011-01-17 | 2011-05-25 | 武汉理工大学 | Denoising, feature extraction and pattern recognition method for human body surface electromyography signals |
CN102169690A (en) * | 2011-04-08 | 2011-08-31 | 哈尔滨理工大学 | Voice signal recognition system and method based on surface myoelectric signal |
CN102622605A (en) * | 2012-02-17 | 2012-08-01 | 国电科学技术研究院 | Surface electromyogram signal feature extraction and action pattern recognition method |
CN103345641A (en) * | 2013-07-16 | 2013-10-09 | 杭州电子科技大学 | Hand electromyographic signal motion recognition method based on wavelet entropy and support vector machine |
CN105005383A (en) * | 2015-07-10 | 2015-10-28 | 昆山美莱来工业设备有限公司 | Wearable arm band that manipulates mobile robot by using hand gesture |
CN105608432A (en) * | 2015-12-21 | 2016-05-25 | 浙江大学 | Instantaneous myoelectricity image based gesture identification method |
CN105654037A (en) * | 2015-12-21 | 2016-06-08 | 浙江大学 | Myoelectric signal gesture recognition method based on depth learning and feature images |
-
2016
- 2016-07-20 CN CN201610574817.9A patent/CN106293057A/en active Pending
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102073881A (en) * | 2011-01-17 | 2011-05-25 | 武汉理工大学 | Denoising, feature extraction and pattern recognition method for human body surface electromyography signals |
CN102169690A (en) * | 2011-04-08 | 2011-08-31 | 哈尔滨理工大学 | Voice signal recognition system and method based on surface myoelectric signal |
CN102622605A (en) * | 2012-02-17 | 2012-08-01 | 国电科学技术研究院 | Surface electromyogram signal feature extraction and action pattern recognition method |
CN103345641A (en) * | 2013-07-16 | 2013-10-09 | 杭州电子科技大学 | Hand electromyographic signal motion recognition method based on wavelet entropy and support vector machine |
CN105005383A (en) * | 2015-07-10 | 2015-10-28 | 昆山美莱来工业设备有限公司 | Wearable arm band that manipulates mobile robot by using hand gesture |
CN105608432A (en) * | 2015-12-21 | 2016-05-25 | 浙江大学 | Instantaneous myoelectricity image based gesture identification method |
CN105654037A (en) * | 2015-12-21 | 2016-06-08 | 浙江大学 | Myoelectric signal gesture recognition method based on depth learning and feature images |
Cited By (36)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106682651A (en) * | 2017-02-10 | 2017-05-17 | 哈尔滨工业大学 | Monitoring method for cable insulation leather damage based on neural network |
CN107126303A (en) * | 2017-02-15 | 2017-09-05 | 上海术理智能科技有限公司 | A kind of upper and lower extremities exercising support method based on mobile phone A PP |
CN107169432A (en) * | 2017-05-09 | 2017-09-15 | 深圳市科迈爱康科技有限公司 | Biometric discrimination method, terminal and computer-readable recording medium based on myoelectricity |
US11587242B1 (en) | 2018-01-25 | 2023-02-21 | Meta Platforms Technologies, Llc | Real-time processing of handstate representation model estimates |
CN111902847A (en) * | 2018-01-25 | 2020-11-06 | 脸谱科技有限责任公司 | Real-time processing of hand state representation model estimates |
CN110378169B (en) * | 2018-04-12 | 2021-06-18 | 中移(杭州)信息技术有限公司 | Gesture interval detection method and device |
CN110378169A (en) * | 2018-04-12 | 2019-10-25 | 中移(杭州)信息技术有限公司 | The detection method and device in gesture section |
CN109085918A (en) * | 2018-06-28 | 2018-12-25 | 天津大学 | Acupuncture needling manipulation training method based on myoelectricity |
CN109085918B (en) * | 2018-06-28 | 2020-05-12 | 天津大学 | Myoelectricity-based acupuncture needle manipulation training method |
CN109190496A (en) * | 2018-08-09 | 2019-01-11 | 华南理工大学 | A kind of monocular static gesture identification method based on multi-feature fusion |
CN109165587A (en) * | 2018-08-11 | 2019-01-08 | 石修英 | intelligent image information extraction method |
CN108992066B (en) * | 2018-08-15 | 2021-02-26 | 东北大学 | Portable lower limb behavior pattern real-time identification method based on electromyographic signals |
CN108992066A (en) * | 2018-08-15 | 2018-12-14 | 东北大学 | Portable lower limb behavior pattern real-time identifying system and method based on electromyography signal |
CN109412725B (en) * | 2018-10-15 | 2021-02-09 | 中国人民解放军战略支援部队信息工程大学 | Radio communication PCMA signal blind demodulation method and device |
CN109412725A (en) * | 2018-10-15 | 2019-03-01 | 中国人民解放军战略支援部队信息工程大学 | The blind demodulation method of radio communication PCMA signal and device |
CN109613976B (en) * | 2018-11-14 | 2023-08-22 | 华东师范大学 | Intelligent flexible pressure sensing sign language recognition device |
CN109613976A (en) * | 2018-11-14 | 2019-04-12 | 华东师范大学 | A kind of intelligent flexible pressure sensing hand language recognition device |
CN109635706A (en) * | 2018-12-04 | 2019-04-16 | 武汉灏存科技有限公司 | Gesture identification method, equipment, storage medium and device neural network based |
CN109635706B (en) * | 2018-12-04 | 2020-09-01 | 武汉灏存科技有限公司 | Gesture recognition method, device, storage medium and device based on neural network |
CN109976526A (en) * | 2019-03-27 | 2019-07-05 | 广东技术师范大学 | A kind of sign Language Recognition Method based on surface myoelectric sensor and nine axle sensors |
CN110413107A (en) * | 2019-06-21 | 2019-11-05 | 浙江科技学院 | Bionic mechanical hand interaction control method based on electromyography signal pattern-recognition and particle group optimizing |
CN110413107B (en) * | 2019-06-21 | 2023-04-25 | 浙江科技学院 | Bionic manipulator interaction control method based on electromyographic signal pattern recognition and particle swarm optimization |
CN111241982A (en) * | 2020-01-07 | 2020-06-05 | 金陵科技学院 | Robot gesture recognition method based on CAE-SVM |
CN111531537B (en) * | 2020-05-07 | 2022-11-01 | 金陵科技学院 | Mechanical arm control method based on multiple sensors |
CN111531537A (en) * | 2020-05-07 | 2020-08-14 | 金陵科技学院 | Mechanical arm control method based on multiple sensors |
CN111695446B (en) * | 2020-05-26 | 2021-07-27 | 浙江工业大学 | Gesture recognition method integrating sEMG and AUS |
CN111695446A (en) * | 2020-05-26 | 2020-09-22 | 浙江工业大学 | Gesture recognition method integrating sEMG and AUS |
CN111870242A (en) * | 2020-08-03 | 2020-11-03 | 南京邮电大学 | Intelligent gesture action generation method based on electromyographic signals |
CN111950460A (en) * | 2020-08-13 | 2020-11-17 | 电子科技大学 | Muscle strength self-adaptive stroke patient hand rehabilitation training action recognition method |
CN112162041A (en) * | 2020-09-30 | 2021-01-01 | 陕西师范大学 | Method for identifying metal material based on Gaussian distribution of amplitude root mean square value |
CN112101298A (en) * | 2020-10-15 | 2020-12-18 | 福州大学 | Gesture recognition system and method based on muscle electrical impedance signals |
CN112732090A (en) * | 2021-01-20 | 2021-04-30 | 福州大学 | Muscle cooperation-based user-independent real-time gesture recognition method |
CN113536954A (en) * | 2021-06-23 | 2021-10-22 | 厦门大学 | Gesture recognition method based on human body electromyographic signals |
CN113688802A (en) * | 2021-10-22 | 2021-11-23 | 季华实验室 | Gesture recognition method, device and equipment based on electromyographic signals and storage medium |
CN113986017A (en) * | 2021-12-27 | 2022-01-28 | 深圳市心流科技有限公司 | Myoelectric gesture template generation method and device and storage medium |
CN113986017B (en) * | 2021-12-27 | 2022-05-17 | 深圳市心流科技有限公司 | Myoelectric gesture template generation method and device and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106293057A (en) | Gesture identification method based on BP neutral net | |
Al-Shawwa et al. | Predicting birth weight using artificial neural network | |
Elamvazuthi et al. | Electromyography (EMG) based classification of neuromuscular disorders using multi-layer perceptron | |
Xia et al. | EMG‐based estimation of limb movement using deep learning with recurrent convolutional neural networks | |
Naser et al. | Predicting student performance using artificial neural network: In the faculty of engineering and information technology | |
CN101317794B (en) | Myoelectric control ability detecting and training method for hand-prosthesis with multiple fingers and multiple degrees of freedom | |
CN108491077A (en) | A kind of surface electromyogram signal gesture identification method for convolutional neural networks of being divided and ruled based on multithread | |
CN106897738A (en) | A kind of pedestrian detection method based on semi-supervised learning | |
He et al. | Hand gesture recognition using MYO armband | |
Dai et al. | A new approach of intelligent physical health evaluation based on GRNN and BPNN by using a wearable smart bracelet system | |
CN110826437A (en) | Intelligent robot control method, system and device based on biological neural network | |
Precup et al. | Experiments in incremental online identification of fuzzy models of finger dynamics | |
CN104809230A (en) | Cigarette sensory quality evaluation method based on multi-classifier integration | |
Oweis et al. | ANN-based EMG classification for myoelectric control | |
CN110413107A (en) | Bionic mechanical hand interaction control method based on electromyography signal pattern-recognition and particle group optimizing | |
Rahimian et al. | Few-shot learning for decoding surface electromyography for hand gesture recognition | |
Hu et al. | Robust continuous hand motion recognition using wearable array myoelectric sensor | |
Ahsan et al. | The use of artificial neural network in the classification of EMG signals | |
Li et al. | Transfer learning-based muscle activity decoding scheme by low-frequency sEMG for wearable low-cost application | |
Caesarendra et al. | EMG based classification of hand gestures using PCA and ANFIS | |
Nazemi et al. | Artificial neural network classifier in comparison with LDA and LS-SVM classifiers to recognize 52 hand postures and movements | |
Peng et al. | Gesture recognition by ensemble extreme learning machine based on surface electromyography signals | |
Millar et al. | LSTM Network Classification of Dexterous Individual Finger Movements | |
Veer et al. | Electromyographic classification of effort in muscle strength assessment | |
Wang et al. | Study on intelligent syndrome differentiation in traditional Chinese medicine based on information fusion technology |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
TA01 | Transfer of patent application right |
Effective date of registration: 20181211 Address after: 10 000 07 Group B 7, 3rd floor, Zhangzizhong Road, Dongcheng District, Beijing Applicant after: Zhang Wendong Address before: 710000 Room 204, Arc Tower, 60 West Avenue, New Industrial Park, Xi'an High-tech Zone, Shaanxi Province Applicant before: XI'AN ZHONGKE BIQI INNOVATION TECHNOLOGY CO., LTD. |
|
TA01 | Transfer of patent application right | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170104 |
|
RJ01 | Rejection of invention patent application after publication |