CN110473603A - A kind of body-building householder method based on electromyography signal - Google Patents
A kind of body-building householder method based on electromyography signal Download PDFInfo
- Publication number
- CN110473603A CN110473603A CN201910614764.2A CN201910614764A CN110473603A CN 110473603 A CN110473603 A CN 110473603A CN 201910614764 A CN201910614764 A CN 201910614764A CN 110473603 A CN110473603 A CN 110473603A
- Authority
- CN
- China
- Prior art keywords
- model
- function
- value
- follows
- training
- 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.)
- Granted
Links
- 238000002567 electromyography Methods 0.000 title claims abstract description 33
- 238000000034 method Methods 0.000 title claims abstract description 31
- 230000006870 function Effects 0.000 claims abstract description 41
- 238000012549 training Methods 0.000 claims abstract description 32
- 238000013528 artificial neural network Methods 0.000 claims abstract description 19
- 238000000605 extraction Methods 0.000 claims abstract description 14
- 230000009471 action Effects 0.000 claims abstract description 13
- 238000012706 support-vector machine Methods 0.000 claims abstract description 13
- 230000004913 activation Effects 0.000 claims abstract description 10
- 238000010801 machine learning Methods 0.000 claims abstract description 9
- 230000000694 effects Effects 0.000 claims abstract description 8
- 238000013507 mapping Methods 0.000 claims abstract description 7
- 230000003183 myoelectrical effect Effects 0.000 claims abstract description 7
- 210000002364 input neuron Anatomy 0.000 claims abstract description 6
- 210000003205 muscle Anatomy 0.000 claims description 11
- 230000008569 process Effects 0.000 claims description 7
- 238000009795 derivation Methods 0.000 claims description 5
- 238000003062 neural network model Methods 0.000 claims description 5
- 238000003199 nucleic acid amplification method Methods 0.000 claims description 5
- 238000006243 chemical reaction Methods 0.000 claims description 4
- 239000011159 matrix material Substances 0.000 claims description 4
- 210000002569 neuron Anatomy 0.000 claims description 4
- 238000012360 testing method Methods 0.000 claims description 4
- 239000006185 dispersion Substances 0.000 claims description 2
- 230000009977 dual effect Effects 0.000 claims description 2
- 230000003387 muscular Effects 0.000 claims description 2
- 238000005457 optimization Methods 0.000 claims description 2
- 238000005070 sampling Methods 0.000 claims description 2
- 230000009466 transformation Effects 0.000 claims description 2
- 241000208340 Araliaceae Species 0.000 claims 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 claims 1
- 235000003140 Panax quinquefolius Nutrition 0.000 claims 1
- 230000005611 electricity Effects 0.000 claims 1
- 235000008434 ginseng Nutrition 0.000 claims 1
- 230000001537 neural effect Effects 0.000 claims 1
- 238000003909 pattern recognition Methods 0.000 abstract description 3
- 238000013473 artificial intelligence Methods 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 11
- 238000012545 processing Methods 0.000 description 5
- 230000003321 amplification Effects 0.000 description 4
- 238000001914 filtration Methods 0.000 description 2
- 206010050031 Muscle strain Diseases 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 239000012141 concentrate Substances 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 230000002265 prevention Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
- A61B5/316—Modalities, i.e. specific diagnostic methods
- A61B5/389—Electromyography [EMG]
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/10—Machine learning using kernel methods, e.g. support vector machines [SVM]
-
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield 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
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H20/00—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance
- G16H20/30—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance relating to physical therapies or activities, e.g. physiotherapy, acupressure or exercising
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Biophysics (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Mathematical Physics (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Public Health (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Veterinary Medicine (AREA)
- Animal Behavior & Ethology (AREA)
- Surgery (AREA)
- Computational Linguistics (AREA)
- Heart & Thoracic Surgery (AREA)
- Pathology (AREA)
- Primary Health Care (AREA)
- Signal Processing (AREA)
- Psychiatry (AREA)
- Physiology (AREA)
- Fuzzy Systems (AREA)
- Epidemiology (AREA)
- Physical Education & Sports Medicine (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
A kind of body-building householder method based on electromyography signal of the present invention belongs to pattern-recognition and field of artificial intelligence, is related to a kind of body-building householder method based on electromyography signal.This method is acquired human body surface myoelectric signal by electrode slice, and is handled by a variety of Acquisition Circuits human body surface myoelectric signal, and by treated, electromyography signal reaches host computer, carries out characteristics extraction to signal.By the SVM support vector machines of machine learning as acts of determination whether the sorter model of standard, use the Gaussian kernel of linear kernel and different lambda parameters respectively, carry out SVM model training respectively, select the model body-building action criteria judgement that training effect is best.By three layers of BP neural network as to the sorter model classified of movement, using the characteristic value extracted as the input neuron of neural network.Use RELU function as activation primitive, realize the Nonlinear Mapping of input information, sorter model is obtained after training and realizes action recognition, there is very high quasi- not rate.
Description
Technical field
The invention belongs to pattern-recognitions and field of artificial intelligence, are related to a kind of body-building auxiliary square based on electromyography signal
Method.
Background technique
Body-building facilitates prevention of various diseases, and in recent years, more and more people recognize the importance of body-building, exercise industry
It quickly grows, penetrates into the crowd of each age level.However nonstandard body-building movement can make body-building effect bad, when serious
It also will cause pulled muscle.In addition, further investigations have shown that, if carrying out body-building in the case where there is supervision, user in training more
It is dynamic.So having great significance to the research of body-building householder method.
Current body-building householder method is based primarily upon body posture, in Depari A, Ferrari P, Flammini A,
Rinaldi S.Lightweight Machine Learning-Based Approach for Supervision of
Fitness Workout[C]//IEEE Conference Sensors Applications Symposium.Institute
Of Electrical and Electronics Engineers Inc in 2019. this articles, passes through x, the appearance of tri- axis of y, z
State carries out body-building auxiliary, can assist under certain error tolerance body-building, but since posture information information content is larger, disturbs
Data are more, are easy to happen erroneous judgement, and can not directly reflect the particular state of muscle by posture information, auxiliary confidence level compared with
It is low.
With the development of machine learning and universal, many complicated algorithms are all achieved, and are also had very in pattern-recognition
Good application scenarios.Machine learning, which is applied to body-building standard determination and body-building action recognition, very high quasi- not rate, but needs
The model of machine learning is compared, most suitable algorithm is selected.
Summary of the invention
The present invention to overcome the shortcomings of existing technologies, has invented a kind of body-building householder method based on electromyography signal, the party
Method acquires the original electromyography signal of body-building personnel by electrode slice, by reaching PC after the filtering and enhanced processing of hardware circuit
End, to treated, electromyography signal carries out integral myoelectricity value (iEMG), average absolute value (MAV), variance (VAR), root-mean-square value
(RMS) four characteristics extractions.By the SVM support vector machines of machine learning as acts of determination whether the classifier mould of standard
Type does training set by the lower standard operation of fitness's guidance, using linear kernel and the Gaussian kernel of different lambda parameters and divides respectively
It carry out not SVM model training.The SVM model for selecting training effect best carries out body-building movement in personal fitness training alone
Standard determination, standard are then determined as 1, otherwise are determined as 0；By three layers of BP neural network as the classification classified to movement
Device model, hidden layer neuron number are 5；Characteristic value to extract is used as the input neuron of neural network
RELU function realizes the Nonlinear Mapping of input information, obtains sorter model after training 1000 times, realize as activation primitive
Action recognition.Method has the characteristics that passband ripple is low, stopband fall off rate is fast, and body-building standard determination has with body-building action recognition
Very high quasi- not rate.
The technical scheme adopted by the invention is as follows a kind of body-building householder method based on electromyography signal, which is characterized in that auxiliary
Aid method is handled human body surface myoelectric signal by electrode slice and a variety of Acquisition Circuits, after reaching host computer, to signal
Carry out characteristics extraction；By the SVM support vector machines of machine learning as acts of determination whether the sorter model of standard, point
Not Shi Yong linear kernel and different lambda parameters Gaussian kernel, carry out SVM model training respectively, select the best model of training effect strong
Body action criteria determines；By three layers of BP neural network as the sorter model classified to movement, with the spy extracted
Input neuron of the value indicative as neural network；Use RELU function as activation primitive, realizes that input the non-linear of information reflects
It penetrates, sorter model is obtained after training, realize action recognition；Specific step is as follows for method:
Electrode slice is attached on muscle to be measured by step 1, original electromyography signal is connected on hardware circuit, original myoelectricity
Signal passes sequentially through pre-amplification circuit, Butterworth high and low frequency filter circuit, trap circuit, second amplifying circuit, is put
Greatly, Hz noise is filtered out treated electromyography signal；Since electromyography signal has positive negativity, then by AD Acquisition Circuit to hardware
Electromyography signal after processing of circuit carries out analog-to-digital conversion, then treated electromyography signal is reached host computer；
Step 2 carries out the characteristics extraction in time domain, the spy of extraction to the host computer electromyography signal that receives that treated
Value indicative are as follows: integral myoelectricity value (iEMG), average absolute value (MAV), variance (VAR), root-mean-square value (RMS)；Wherein, myoelectricity is integrated
Value iEMG is the absolute summation of all sampled points, reflects the gross energy information of muscle in a sampling period:
Average absolute value (MAV) can reflect the average energy information of muscle,
Variance (VAR) reflects muscular energy dispersion degree,
Root-mean-square value (RMS) reflects muscle effective energy information,
Aforementioned four characteristic value is input in SVM supporting vector machine model by step 3, carries out model training；SVM is supported
Vector machine hyperplane equation are as follows:
wTX+b=0 (5)
Wherein, w is the weight vector of each dimension, and b is the bias vector of each dimension, and x is sample data in each dimension
The feature value vector of degree；Classifier functions only need to export two as a result, simultaneously created symbol function sign, when functional value is when being greater than
It is 1 when zero, is -1 when less than 0, classifier functions formula are as follows:
F (x)=sign (wTxi+b) (6)
According to support vector machines principle, the process for selecting hyperplane is to be made apart from hyperplane most by changing w and b vector
Close point distance maximizes, and objective function Equation is as follows:
Constraint condition formula is as follows:
It is as follows that objective function Equation is obtained by deformation:
Constraint condition are as follows:
The convex function that function becomes under N number of Linear Constraints seeks Constrained and Unconstrained Optimization, and is strong dual problem, passes through glug
Bright day multiplier solves, formula are as follows:
Former objective function becomes formula 8:
G (x)=minw,bmaxλ Lw,b,λ (12)
It, can be using dualistic transformation as formula 9 since the function has strong duality:
G (x)=maxλ minw,bLw,b,λ (13)
Constraint condition are as follows:
λi≥0 (14)
Since convex function extreme value is most worth, local derviation can be asked to w, b, λ by L respectively, 0 solution is equal to by partial derivative
The expression formula of w out, w are as follows:
Since function has strong duality, meet KKT condition, KKT condition is 0 and λ except three local derviationsi>=0 condition
Additional conditions there are two outer, additional conditions are formula 16, formula 17:
λi×(1-f(xi)×(wTxi+ b))=0 (16)
1-f(xi)×(wTxi+b)≤0 (17)
W is brought into KKT condition, is solved:
Wherein, xkIt is the value for meeting KKT inequality condition；
Above-mentioned formula is SVM in the derivation process using linear kernel function, if data are non-linear, selection rbf Gaussian kernel
Function is calculated, and derivation process is similarly；Gaussian kernel function are as follows:
k(xi,xj)=e-γ||xi-xj||2 (19)
SVM support vector machines decision model is trained by the above method, whether standard determines to body-building；
Step 4, using aforementioned four characteristic value as the input neuron of BP neural network model, carry out model training；It is logical
Three layers of BP neural network are crossed as the sorter model classified to movement, hidden layer neuron number is 5；Use RELU
Function realizes the Nonlinear Mapping of input information as activation primitive；It need to first be inputted when the training of BP neural network model initial
Change weight coefficient matrix w, initialization intercept coefficient matrix b and test set；Wherein, test set include feature value vector x and with
The corresponding ideal output vector y of every group of feature value vector.Use RELU function as activation primitive, realizes the non-of input information
Linear Mapping obtains model output vector by forward-propagatingJudged by error functionWith the error of y, as amendment w
With the parameter of b.By backpropagation, error function corrects weight coefficient w and intercept parameter b step by step, positive again after amendment to pass
It broadcasts, by the repetition training to model, obtains ideal BP neural network disaggregated model, by inputting four kinds of characteristic values, to strong
Body movement is classified.
The beneficial effects of the invention are as follows this method to be acquired human body surface myoelectric signal by electrode slice, and by more
Kind of Acquisition Circuit handles human body surface myoelectric signal, and surface dynamoelectric signal can intuitively reflect muscle information, data compared with
Benefit reason can be determined and be classified to body-building movement in the case where serious forgiveness is very low.Ten rank Butterworths are selected to filter
Device has the characteristics that passband ripple is low, stopband fall off rate is fast.Will treated that electromyography signal reaches host computer, to signal into
Row characteristics extraction.By the SVM support vector machines of machine learning as acts of determination whether the sorter model of standard, respectively
Using linear kernel and the Gaussian kernel of different lambda parameters and progress SVM model training respectively, the model body-building for selecting training effect best
Action criteria determines.Body builder can carry out SVM model training under coach directed, pass through trained mould when training alone
Type determine itself movement whether standard.By three layers of BP neural network as the sorter model classified to movement, to mention
Input neuron of the characteristic value got as neural network.Use RELU function as activation primitive, realizes input information
Nonlinear Mapping obtains sorter model after training, realize action recognition.Machine learning is applied to body-building standard determination and is good for
In body action recognition, there is very high quasi- not rate.Body builder can carry out the plan in a body-building period by body-building disaggregated model
It arranges and executes, reach setting target when counting each body-building movement by classifier, body-building terminates.There is the progress of target
Body-building can improve body-building effect with the raising body-building enthusiasm of high degree.
Detailed description of the invention
Fig. 1 is that electrode slice pastes schematic diagram and body-building acts implementation diagram, a) is chest-developer ordinary grip, b) it is chest-developer
Reverse grip, c) it is attachment of electrodes position.
Fig. 2 is the flow chart of householder method of the present invention.
Fig. 3 is ten rank Butterworth filter circuit schematic diagrames of the invention.
Fig. 4 is characteristics extraction figure of the invention.
Fig. 5 is the whole hardware schematic of the present invention.
Fig. 6 is software schematic diagram of the invention.
Fig. 7 is SVM support vector machines schematic diagram of the invention.
Fig. 8 is BP neural network schematic diagram of the invention,
Fig. 9 is classifying quality figure of the invention.Wherein, Fig. 9 a) be ideal sort as a result, Fig. 9 b) it is category of model result.
Specific embodiment
The present invention is described in further details with technical solution with reference to the accompanying drawing:
Attached drawing 1 is that electrode slice pastes schematic diagram and body-building acts implementation diagram, a) is chest-developer ordinary grip, b) it is arm strength
Device reverse grip, c) it is attachment of electrodes position.It is illustrated by taking the body-building auxiliary to upper chest as an example.Hardware and software of the invention with it is upper
Machine schematic diagram is shown in attached drawing 5 and attached drawing 6 respectively.
Attached drawing 3 is ten rank Butterworth circuit diagrams, using connection type even after preceding surprise.Wherein, 50Hz trap circuit: by
Have a great impact in Hz noise to signal, 50Hz trap circuit carries out at trap the interference of 50Hz using double T trappers
Reason.Second amplifying circuit: via front stage circuits, treated that electromyography signal is relatively pure, after needing to continue to be exaggerated in order to
Phase data processing selects operational amplifier circuit to amplify.AD Acquisition Circuit: electromyography signal has positive negativity, selects 14 AD conversion cores
Piece carries out analog-to-digital conversion to the electromyography signal after hardware circuits which process.
Attached drawing 2 is the flow chart of householder method of the present invention.Dotted portion is hardware circuit to original electromyography signal in figure
Processing, is divided into six parts, respectively pre-amplification circuit, high-pass filtering circuit, low-pass filter circuit, 50Hz trap circuit,
Second amplifying circuit, AD Acquisition Circuit.Specific step is as follows for householder method:
Electrode slice is attached on muscle to be measured by step 1, and original electromyography signal is connected on hardware circuit.Preposition amplification
Circuit: original electromyography signal is very faint, needs the preposition enhanced processing by high input impedance, high cmrr, preposition
Amplifying circuit selects AD8221 amplifier chip, and 100 times of amplification is carried out to signal.High pass low-pass filter circuit: surface myoelectric letter
Number main energetic concentrates within the scope of 10Hz-1000Hz, there are High-frequency Interference, low-frequency disturbance in environment, needs to be filtered place
Reason；High pass low-pass filter circuit selects five rank high pass Butterworth filters and five rank low pass Butterworth filters, group respectively
Cheng Shijie Butterworth filter has the characteristics that passband ripple is low, stopband fall off rate is fast.
Step 2 carries out the characteristics extraction in time domain, the spy of extraction to the host computer electromyography signal that receives that treated
Value indicative are as follows: integral myoelectricity value (iEMG), average absolute value (MAV), variance (VAR), root-mean-square value (RMS).Using formula
(1)-(4) obtain.Attached drawing 4 is characterized value extraction figure, in host computer extract real-time and draws curve.
The characteristic value extracted is input in SVM supporting vector machine model by step 3, carries out model training；SVM is supported
Vector machine hyperplane equation is formula (5), and classifier functions export two as a result, being 1, when less than 0 when functional value is greater than zero
It is -1.By formula (6)-(19) derivation, SVM support vector machines decision model is trained, whether standard is sentenced to body-building
It is fixed.
Attached drawing 7 is SVM support vector cassification two-dimensional surface plan view, and support vector machines is a kind of based on classification boundaries
Algorithm, core concept are to find so that the maximized hyperplane of isolated degree, under the premise of guaranteeing nicety of grading with two dimension
For plane, upper left side represents positive sample, and lower right represents negative sample, and H is hyperplane, HPIt is nearest from hyperplane in positive sample
Parallel lines, HNIt is parallel lines nearest from hyperplane in negative sample, hyperplane, which takes, arrives HPWith HNThe equal position of two plan ranges
It sets.Each characteristic value represents a dimension, by the selection of constantly more various hyperplane, so that HPWith HNThe distance between most
Greatly, so that distance maximizes between positive sample and the most marginal point of negative sample, SVM support vector machines is obtained with this and determines mould
Type improves SVM model by selecting different core, and the formula reasoning that this patent provides is linear kernel, other core reasonings are similarly.
Step 4, using aforementioned four characteristic value as the input layer of BP neural network model, carry out model training.
Attached drawing 8 is BP neural network disaggregated model schematic diagram, using 4 characteristic values as the input of three layers of BP neural network
Layer, hidden layer neuron number are 5, and the activation primitive between every layer selects RELU function, are passed by forward-propagating with reversed
Constantly amendment weight coefficient w and intercept parameter b is broadcast, BP neural network disaggregated model is obtained with this.
Embodiment 1, using BP neural network to kg sitting posture concentration curl, 10kg sitting posture concentration curl, 30kg
Chest-developer ordinary grip three act the classification results figure classified, as shown in Figure 9.Three movements are respectively marked as 1,2,3.By three
Kind of movement, which is respectively done 20 groups and is input in trained BP neural network disaggregated model, obtains output valve.Fig. 9 a) it is ideal sort knot
Fruit, Fig. 9 b) it is category of model as a result, it can be seen in figure 9 that the classifying quality that acts of label 1 is worst, but also completely can be with
Other two kinds movements are opened respectively.
Embodiment, 2, it is acted using chest-developer ordinary grip mock standard body-building, as shown in Fig. 1 a)；Use chest-developer reverse grip mould
Intend non-standard body-building movement, as shown in Fig. 1 b).10 groups of movements are respectively done into two kinds of movements, are input under the SVM model under different IPs
Judgement as a result, following table is judgement result of the invention.
Determine that result is best using linear kernel, is only once judged by accident in 20 times.
Claims (1)
1. a kind of body-building householder method based on electromyography signal, which is characterized in that householder method passes through electrode slice and a variety of acquisitions
Circuit handles human body surface myoelectric signal, after reaching host computer, carries out characteristics extraction to signal；Pass through machine learning
SVM support vector machines as acts of determination whether the sorter model of standard, respectively use linear kernel and different lambda parameters height
This core carries out SVM model training respectively, and the model body-building action criteria for selecting training effect best determines；Pass through three layers of BP mind
It is neural using the input of characteristic value as the neural network extracted through network as the sorter model classified to movement
Member；Use RELU function as activation primitive, realizes the Nonlinear Mapping of input information, obtain sorter model after training, it is real
Existing action recognition；Specific step is as follows for method:
Electrode slice is attached on muscle to be measured by step 1, original electromyography signal is connected on hardware circuit, original electromyography signal
Pre-amplification circuit, Butterworth high and low frequency filter circuit, trap circuit, second amplifying circuit are passed sequentially through, amplified,
Filter out Hz noise treated electromyography signal；Since electromyography signal has positive negativity, then by AD Acquisition Circuit to hardware electricity
Electromyography signal that treated on road carries out analog-to-digital conversion, then treated electromyography signal is reached host computer, to electromyography signal
Carry out the characteristics extraction in time domain；
Step 2 carries out the characteristics extraction in time domain, the characteristic value of extraction to the host computer electromyography signal that receives that treated
Are as follows: integral myoelectricity value (iEMG), average absolute value (MAV), variance (VAR), root-mean-square value (RMS)；Wherein, myoelectricity value is integrated
IEMG is the absolute summation of all sampled points, reflects the gross energy information of muscle in a sampling period:
Average absolute value (MAV) can reflect the average energy information of muscle,
Variance (VAR) reflects muscular energy dispersion degree,
Root-mean-square value (RMS) reflects muscle effective energy information,
Aforementioned four characteristic value is input in SVM supporting vector machine model by step 3, carries out model training；SVM supporting vector
Machine hyperplane equation are as follows:
wTX+b=0 (5)
Wherein, w is the weight vector of each dimension, and b is the bias vector of each dimension, and x is sample data in each dimension
Feature value vector；Classifier functions only need to export two as a result, simultaneously created symbol function sign, when functional value is when being greater than zero
It is 1, is -1 when less than 0；Classifier functions formula are as follows:
F (x)=sign (wTxi+b) (6)
According to support vector machines principle, the process for selecting hyperplane is to be made nearest apart from hyperplane by changing w and b vector
Point distance maximizes, and objective function Equation is as follows:
Constraint condition formula is as follows:
It is as follows that objective function Equation is obtained by deformation:
Constraint condition are as follows:
The convex function that function becomes under N number of Linear Constraints seeks Constrained and Unconstrained Optimization, and is strong dual problem, passes through Lagrange
Multiplier solves, formula are as follows:
Former objective function becomes formula 8:
G (x)=minw,bmaxλLw,b,λ (12)
It, can be using dualistic transformation as formula 9 since the function has strong duality:
G (x)=maxλminw,bLw,b,λ (13)
Constraint condition are as follows:
λi≥0 (14)
Since convex function extreme value is most worth, local derviation can be asked to w, b, λ by L respectively, be equal to 0 by partial derivative and solve w,
The expression formula of w are as follows:
Since function has strong duality, meet KKT condition, KKT condition is 0 and λ except three local derviationsiOutside >=0 condition also
Two additional conditions, additional conditions are formula 16, formula 17:
λi×(1-f(xi)×(wTxi+ b))=0 (16)
1-f(xi)×(wTxi+b)≤0 (17)
W is brought into KKT condition, is solved:
Wherein, xkIt is the value for meeting KKT inequality condition；
Above-mentioned formula is SVM in the derivation process using linear kernel function, if data are non-linear, selection rbf gaussian kernel function
It is calculated, derivation process is similarly；Gaussian kernel function are as follows:
SVM support vector machines decision model is trained by the above method, the SVM model for selecting training effect best, to body-building
Whether standard is determined；Carry out the judgement of body-building action criteria in personal fitness training alone, standard is then determined as 1, on the contrary
It is determined as 0；
Step 4, using aforementioned four characteristic value as the input neuron of BP neural network model, carry out model training；Pass through three
For layer BP neural network as the sorter model classified to movement, hidden layer neuron number is 5；Use RELU function
As activation primitive, the Nonlinear Mapping of input information is realized；First input initialization power is needed when the training of BP neural network model
Weight coefficient matrix w, initialization intercept coefficient matrix b and test set, wherein test set include feature value vector x and with every group of spy
The corresponding ideal output vector y of value indicative vector；Use RELU function as activation primitive, realizes that input the non-linear of information reflects
It penetrates through forward-propagating, obtains model output vectorJudged by error functionWith the error of y, as amendment w and b ginseng
Number；By backpropagation, error function corrects weight coefficient w and intercept parameter b step by step, and forward-propagating again, passes through after amendment
Repetition training to model obtains ideal BP neural network disaggregated model, by input four kinds of characteristic values, to body-building act into
Row classification.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910614764.2A CN110473603B (en) | 2019-07-09 | 2019-07-09 | Body-building assisting method based on electromyographic signals |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910614764.2A CN110473603B (en) | 2019-07-09 | 2019-07-09 | Body-building assisting method based on electromyographic signals |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110473603A true CN110473603A (en) | 2019-11-19 |
| CN110473603B CN110473603B (en) | 2022-04-12 |
Family
ID=68507153
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910614764.2A Expired - Fee Related CN110473603B (en) | 2019-07-09 | 2019-07-09 | Body-building assisting method based on electromyographic signals |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110473603B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115349876A (en) * | 2022-09-22 | 2022-11-18 | 北京市神经外科研究所 | Wearable wireless facial myoelectricity collection system and myoelectricity collection system |
| CN117612669A (en) * | 2023-10-17 | 2024-02-27 | 广东东软学院 | Rehabilitation training safety assessment method based on wearable equipment |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107341351A (en) * | 2017-07-06 | 2017-11-10 | 京东方科技集团股份有限公司 | Intelligent body-building method, apparatus and system |
| CN108209910A (en) * | 2017-05-25 | 2018-06-29 | 深圳市未来健身衣科技有限公司 | The feedback method and device of body building data |
| CN109213305A (en) * | 2017-06-29 | 2019-01-15 | 沈阳新松机器人自动化股份有限公司 | A kind of gesture identification method based on surface electromyogram signal |
| CN109934111A (en) * | 2019-02-12 | 2019-06-25 | 清华大学深圳研究生院 | A kind of body-building Attitude estimation method and system based on key point |
-
2019
- 2019-07-09 CN CN201910614764.2A patent/CN110473603B/en not_active Expired - Fee Related
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108209910A (en) * | 2017-05-25 | 2018-06-29 | 深圳市未来健身衣科技有限公司 | The feedback method and device of body building data |
| CN109213305A (en) * | 2017-06-29 | 2019-01-15 | 沈阳新松机器人自动化股份有限公司 | A kind of gesture identification method based on surface electromyogram signal |
| CN107341351A (en) * | 2017-07-06 | 2017-11-10 | 京东方科技集团股份有限公司 | Intelligent body-building method, apparatus and system |
| CN109934111A (en) * | 2019-02-12 | 2019-06-25 | 清华大学深圳研究生院 | A kind of body-building Attitude estimation method and system based on key point |
Non-Patent Citations (2)
| Title |
|---|
| 丁其川等: "基于表面肌电的运动意图识别方法研究及应用综述", 《自动化学报》 * |
| 佟丽娜等: "基于多路 sEMG 时序分析的人体运动模式识别方法", 《自动化学报》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115349876A (en) * | 2022-09-22 | 2022-11-18 | 北京市神经外科研究所 | Wearable wireless facial myoelectricity collection system and myoelectricity collection system |
| CN115349876B (en) * | 2022-09-22 | 2023-09-15 | 北京市神经外科研究所 | Myoelectricity acquisition system |
| CN117612669A (en) * | 2023-10-17 | 2024-02-27 | 广东东软学院 | Rehabilitation training safety assessment method based on wearable equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110473603B (en) | 2022-04-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Pham et al. | Improving skin-disease classification based on customized loss function combined with balanced mini-batch logic and real-time image augmentation | |
| CN105654037B (en) | A kind of electromyography signal gesture identification method based on deep learning and characteristic image | |
| CN108764207A (en) | A kind of facial expression recognizing method based on multitask convolutional neural networks | |
| CN101317794B (en) | Myoelectric control ability detecting and training method for hand-prosthesis with multiple fingers and multiple degrees of freedom | |
| Xia et al. | A novel wearable electrocardiogram classification system using convolutional neural networks and active learning | |
| CN104573630B (en) | Multiclass brain power mode ONLINE RECOGNITION method based on double SVMs probability outputs | |
| CN108615010A (en) | Facial expression recognizing method based on the fusion of parallel convolutional neural networks characteristic pattern | |
| CN109711383A (en) | Convolutional neural networks Mental imagery EEG signal identification method based on time-frequency domain | |
| CN110473603A (en) | A kind of body-building householder method based on electromyography signal | |
| WO2020042511A1 (en) | Motion potential brain-machine interface encoding and decoding method based on spatial filtering and template matching | |
| CN101859377A (en) | Electromyographic signal classification method based on multi-kernel support vector machine | |
| CN106527716A (en) | Wearable equipment based on electromyographic signals and interactive method between wearable equipment and terminal | |
| CN108171318A (en) | One kind is based on the convolutional neural networks integrated approach of simulated annealing-Gaussian function | |
| CN109907753B (en) | Multi-dimensional ECG signal intelligent diagnosis system | |
| CN109344856B (en) | Offline signature identification method based on multilayer discriminant feature learning | |
| Wan et al. | A neural network to identify human subjects with electrocardiogram signals | |
| CN113111831A (en) | Gesture recognition technology based on multi-mode information fusion | |
| Yao et al. | Multi-feature gait recognition with DNN based on sEMG signals | |
| Meng et al. | A motor imagery EEG signal classification algorithm based on recurrence plot convolution neural network | |
| Caesarendra et al. | EMG based classification of hand gestures using PCA and ANFIS | |
| CN113116361A (en) | Sleep staging method based on single-lead electroencephalogram | |
| Kumar et al. | A critical review on hand gesture recognition using semg: Challenges, application, process and techniques | |
| Xue et al. | Late potential recognition by artificial neural networks | |
| CN110738093A (en) | Classification method based on improved small world echo state network electromyography | |
| Lv et al. | A novel interval type-2 fuzzy classifier based on explainable neural network for surface electromyogram gesture recognition |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20220412 |