CN108564105A - Online gesture recognition method for myoelectric individual difference problem - Google Patents
Online gesture recognition method for myoelectric individual difference problem Download PDFInfo
- Publication number
- CN108564105A CN108564105A CN201810165350.1A CN201810165350A CN108564105A CN 108564105 A CN108564105 A CN 108564105A CN 201810165350 A CN201810165350 A CN 201810165350A CN 108564105 A CN108564105 A CN 108564105A
- Authority
- CN
- China
- Prior art keywords
- sample
- initial
- collection
- kkt
- newly
- 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
- 238000000034 method Methods 0.000 title claims abstract description 40
- 230000003183 myoelectrical effect Effects 0.000 title abstract 3
- 238000012549 training Methods 0.000 claims abstract description 77
- 230000013016 learning Effects 0.000 claims abstract description 24
- 238000012216 screening Methods 0.000 claims abstract description 18
- 238000013145 classification model Methods 0.000 claims abstract description 12
- 210000000707 wrist Anatomy 0.000 claims description 12
- 238000000605 extraction Methods 0.000 claims description 3
- 230000033001 locomotion Effects 0.000 claims description 3
- 230000002123 temporal effect Effects 0.000 claims description 3
- 238000002567 electromyography Methods 0.000 abstract description 2
- 238000012706 support-vector machine Methods 0.000 description 10
- 206010068052 Mosaicism Diseases 0.000 description 4
- 210000003765 sex chromosome Anatomy 0.000 description 4
- 238000012986 modification Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 238000013459 approach Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 210000003205 muscle Anatomy 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- LFQSCWFLJHTTHZ-UHFFFAOYSA-N Ethanol Chemical compound CCO LFQSCWFLJHTTHZ-UHFFFAOYSA-N 0.000 description 1
- 101000739754 Homo sapiens Semenogelin-1 Proteins 0.000 description 1
- 235000008331 Pinus X rigitaeda Nutrition 0.000 description 1
- 235000011613 Pinus brutia Nutrition 0.000 description 1
- 241000018646 Pinus brutia Species 0.000 description 1
- 241000411545 Punargentus Species 0.000 description 1
- 102100037550 Semenogelin-1 Human genes 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000004070 electrodeposition Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 210000003414 extremity Anatomy 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 210000001087 myotubule Anatomy 0.000 description 1
- 230000000399 orthopedic effect Effects 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000002035 prolonged effect Effects 0.000 description 1
- 210000002374 sebum Anatomy 0.000 description 1
- 230000001568 sexual effect Effects 0.000 description 1
Classifications
-
- 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
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
An online gesture recognition method aiming at the problem of myoelectric personal difference comprises the following steps: (1) establishing a gesture electromyography data set to obtain initial data; (2) training an initial classification model; (3) newly adding a sample KKT for judgment; (4) obtaining a preliminary training set by using DBSCAN density clustering; (5) and (5) obtaining a final training set through secondary screening. The invention provides an online gesture recognition method aiming at the problem of myoelectricity individual difference. The method is combined with a DBSCAN density clustering algorithm, improves the original KKT-SVM increment learning method, is used for online myoelectric gesture recognition, and solves the problem of individual difference.
Description
Technical field
The present invention relates to electromyography signal processing and incremental learning fields, are asked for myoelectricity individual difference more particularly to one kind
The online gesture identification method of topic.
Background technology
Surface electromyogram signal (sEMG) is used as a kind of interactive medium means, is widely used in artificial limb, ectoskeleton, orthopedic
In the control of the peripheral hardwares such as device.SEMG can reflect muscle activity degree in real time, be ideal voltage input between people-machine.But by
In the individual differences sexual factor such as sebum, force method, muscle fiber, carried out it is difficult to obtain a general common classification model
Myoelectricity pattern-recognition, therefore generally require to carry out prolonged training early period for individual consumer to obtain exact classification identification mould
Type, relatively time-consuming effort.It is or longer and in practical application, grader generally there will be no change after training for the first time
A period of time in immobilize.
Have many scholars both at home and abroad and uses the adaptive approach based on incremental learning (Incremental Learning)
Solve the problems, such as the time-varying (such as environmental change, electrode position change, fatigue) of sEMG.By incremental learning, classifier system can
On the basis of having learnt to obtain knowledge, to learn to new knowledge constantly from new samples.But still rare research is by increment
The thought of study is trained the early period for individual consumer's disaggregated model, to solve the individual difference sex chromosome mosaicism of sEMG.
In numerous Incremental Learning Algorithms, based on the incremental learning of support vector machines (SVM) due to its peculiar advantage, closely
The hot issue of research is increasingly becoming over year.It is by this earliest by the SVM Incremental Learning Algorithm Batch-ISVM of the propositions such as Syed
Training learns supporting vector collection (SV collection) after secondary increment together in next incremental learning with all newly-increased samples, gives up non-
SV samples.The problem of this method, is in newly-increased sample may to include useless or bad sample, not only influence training speed
Accuracy of identification can be reduced.Based on this, there is scholar to propose the SVM increments based on KKT conditions (Karush-Kuhn-Tucker)
Algorithm (KKT-ISVM) is practised, to obtain effectively newly-increased sample.Although KKT-ISVM ratios Batch-ISVM is further, there are still
Problems with：(1) the non-SV samples given up may include useful classification information；(2) that gives up because meeting KKT conditions is newly-increased
Sample may also include useful classification information.
Invention content
In order to overcome existing gesture identification mode can not to solve individual difference sex chromosome mosaicism, recognition accuracy lower not
Foot, the present invention provide a kind of online gesture identification method for myoelectricity individual difference problem.This method combination DBSCAN density
Clustering algorithm improves original KKT-SVM Increment Learning Algorithms, online myoelectricity gesture identification is used for, to solve individual difference
Sex chromosome mosaicism.
The technical solution adopted by the present invention to solve the technical problems is：
A kind of online gesture identification method for myoelectricity individual difference problem includes the following steps：
(1) gesture myoelectricity data set is established, and obtains primary data；
(2) training preliminary classification model；
(3) sample KKT is increased newly to differentiate；
(4) DBSCAN Density Clusterings are used to obtain initial training collection；
(5) postsearch screening obtains final training set.
Further, in the step (1), the primary data is to build the initial model of incremental learning.It adopts altogether
Collect 8 kinds of common gesture motions of 20 males subject, respectively：Hand opens, hold fist, hand crawl, hand loosen, receive in wrist,
Wrist abduction, wrist outward turning and wrist inward turning.The sample rate of myoelectricity is 1024Hz, is opened a window by the way of sliding window, length of window is set as
200ms.3 kinds of temporal signatures extractions, respectively root-mean-square value (RMS), average absolute value (MAV) and integral are carried out to each window
Myoelectricity (iEMG).One window (sample) can get one 3 (characteristic value) × 4 (channel) totally 12 the feature vector [{ RMS tieed upk,
MAVk,iEMGk}Ch], wherein k is sample number, and Ch is port number.Each subject follows the prompt in screen, keeps one at random
Kind gesture 5s, each gesture occur 30 times, then the data set being each tested isGesture × 30 time, wherein
twindowFor windowing length (200ms), then the data set total sample number being each tested is 6000
Further, in the step (2), for each subject, by all instructions of 19 subjects of remaining in primary data
Practice the SVM preliminary classification models Ψ that sample is used for training the subjectoldWith initial SV collection Dold-SV。
Further, in the step (3), an incremental learning is carried out within every 120 seconds, it is each to be tested in the period
The prompt in screen is followed, keeps a kind of gesture 5s, each gesture 3 times (i.e. 120 seconds) occur at random, increment sample isGesture × 3 time totally 600.Use preliminary classification model ΨoldTo detect newly-increased sample set Dnew, by newly-increased sample set
DnewIt is divided into the newly-increased training sample D for meeting KKT conditionsnew-KKTWith the newly-increased training sample D for being unsatisfactory for KKT conditionsnew-NKKT.Area
Divide method：If sample is located on class interval or outside class interval, to meet the newly-increased training sample of KKT conditions；If sample bit
In in class interval, to be unsatisfactory for the newly-increased training sample of KKT conditions.
In the step (4), the newly-increased training sample D of KKT conditions will be metnew-KKT, initial SV collection Dold-SVWith it is initial non-
SV collection Dold-NSVIt puts together, carries out Density Clustering using DBSCAN, obtain multiple clustering cluster Cm.It is initial to select kernel object
SV collection Dold-SVThe clustering cluster C of middle samplem, these clustering clusters CmIn include the initial non-SV samples D ' near initial SV collectionold-NSV
With the newly-increased training sample D ' for meeting KKT conditions near initial SV collectionnew-KKT.Initial SV collection Dold-SV, near initial SV collection
Initial non-SV samples D 'old-NSV, the newly-increased training sample D ' for meeting KKT conditions near initial SV collectionnew-KKTBe unsatisfactory for KKT
The newly-increased training sample D of conditionnew-NKKTConstitute initial training collection Dnew1。
In the step (4), the DBSCAN Density Clusterings method is determined poly- by the tightness degree of sample distribution
Class formation.It is based on one group of Neighbourhood parameter (ε, MINPts) to analyze the connectivity between sample, and not open close extend excessively is gathered
Class cluster obtains cluster result.Take ε=0.11, MinPts=4.The DBSCAN Density Clustering method and steps are as follows：
(4.1) kernel object collection is found out based on Neighbourhood parameter；
(4.2) " seed " kernel object is randomly selected from kernel object concentration, finds out the reachable sample of all density,
Constitute first clustering cluster；
(4.3) core sample for including in clustering cluster is concentrated from kernel object and is removed.
In the step (5), to the initial non-SV samples D ' near initial SV collectionold-NSVWith the satisfaction near initial SV collection
The newly-increased training sample D ' of KKT conditionsnew-KKTPostsearch screening is carried out, to reduce training set number of samples, reduces the training time.Cause
It is too far with a distance from Optimal Separating Hyperplane, becomes that SV probabilities are smaller, and postsearch screening rule is：It calculates initial non-near initial SV collection
SV samples D 'old-NSVWith the newly-increased training sample D ' for meeting KKT conditions near initial SV collectionnew-KKTIn each sample to classification
The distance of hyperplane, the maximum distance from hyperplane is δ, chooses the sample that distance in the two sample sets is less than 1/2 δ, obtains
Initial non-SV samples D " near initial SV collection after postsearch screeningold-NSVMeet the new of KKT conditions near initial SV collection
Increase training sample D "new-KKT.By initial SV collection Dold-SV, initial non-SV samples near initial SV collection after postsearch screening
D″old-NSV, the newly-increased training sample D " for meeting KKT conditions near initial SV collection after postsearch screeningnew-KKTBe unsatisfactory for KKT
The newly-increased training sample D of conditionnew-NKKTConstitute final training set Dnew2, for training the disaggregated model after the secondary incremental learning
Ψnew。
The beneficial effects of the invention are as follows：
(1) DBSCAN density clustering algorithms are combined, original KKT-SVM Increment Learning Algorithms are improved, are used for online flesh
Electric gesture identification, to solve individual difference sex chromosome mosaicism.
(2) this method can keep higher recognition accuracy.The average classification discrimination of preceding 10 incremental learnings is calculated, this
Inventive method (92.12%) and tradition SVM (69.23%), Batch-ISVM (86.41%), (90.03%) three kind of KKT-ISVM
Algorithm is compared, and has highest average discrimination.Wherein, Batch-ISVM and KKT-ISVM is traditional Increment Learning Algorithm；It passes
System SVM algorithm has no incremental learning process, directly by preliminary classification model increase newly the classification of sample below, for describing
The individual difference of myoelectricity (discrimination is very low, that is, embodies the problem of individual difference influences common model Classification and Identification rate).
(3) this method can further increase disaggregated model training pace of learning.Calculate the average instruction of preceding 10 incremental learnings
Practice time, two kinds of algorithm phases of the method for the present invention (30.43 seconds) and Batch-ISVM (37.88 seconds) and KKT-ISVM (32.54 seconds)
Than having the faster training time.
Description of the drawings
Fig. 1 is holistic approach flow chart of the present invention；
Fig. 2 is DBSCAN Density Clustering method flow diagrams.
Specific implementation mode
The invention will be further described with reference to the accompanying drawings and examples.
Referring to Figures 1 and 2, a kind of online gesture identification method for myoelectricity individual difference problem, includes the following steps：
(1) gesture myoelectricity data set is established, and obtains primary data.The primary data is to build the initial of incremental learning
Model.8 kinds of common gesture motions of 20 males subject are acquired altogether, respectively：Hand opens, holds fist, hand captures, hand is put
Receipts, wrist abduction, wrist outward turning and wrist inward turning in pine, wrist.The MP150 type multichannel physiological signals of BIOPAC companies of the U.S. record and analyze
System is used to the myoelectricity letter of acquisition subject 4 pieces of target muscles of the right hand (musculus palmaris longus, extensor digitorum, flexor digitorum superficialis and flexor carpi ulnaris)
Number.It needs, using alcohol removal dead skin and skin oil and fat, to reduce impedance before sticking electrode.The sample rate of myoelectricity is 1024Hz, is used
The mode of sliding window opens a window, and length of window is set as 200ms.3 kinds of temporal signatures extractions, respectively root mean square are carried out to each window
It is worth (RMS), average absolute value (MAV) and integral myoelectricity (iEMG), formula is as follows：
Wherein, N is sampling number, XiFor the emg amplitude of sampled point in window, N1To integrate starting point, N2To integrate terminal,
X (t) is myoelectricity curve, and d (t) is the time interval of sampling.It is (logical that one window (sample) can get one 3 (characteristic value) × 4
Road) totally 12 dimension feature vector [{ RMSk,MAVk,iEMGk}Ch], wherein k is sample number, and Ch is port number.Each subject with
With the prompt in screen, a kind of gesture 5s, each gesture is kept to occur at random 30 times, the data set being each tested is
Wherein, twindowFor windowing length (200ms), then the data set total sample number being each tested is 6000.
(2) training preliminary classification model.For each subject, by all trained samples of 19 subjects of remaining in primary data
This is used for training the SVM preliminary classification models Ψ of the subjectold.The model training method：Cores of the SVM in classification problem
Thought wants to find the division hyperplane with largest interval, to obtain the correct classification of training sample；If training set is D=
{(x1,y1),...(xm,ym), yi∈ { -1 ,+1 } divides hyperplane and is represented by ωTX+b=0, apart from nearest several of hyperplane
A training sample is referred to as supporting vector (SV), and the sum of the distance of SV to the hyperplane of two classes is
Largest interval to be found finds parameter ω and b and so that γ is maximum, is equivalent to minimize | | ω | |2, then
s.t.yi(ωTxi+ b) >=1, i=1,2 ..., m
The formula introduces method of Lagrange multipliers, disaggregated model can be obtained as a convex quadratic programming problem
Wherein αiFor Lagrange multiplier.Since SVM is a kind of two graders, the present embodiment uses one-to-one method (one-
Versus-one more classification problems) are handled.Obtain preliminary classification model ΨoldAfterwards, while initial SV collection D is obtainedold-SV。
(3) sample KKT is increased newly to differentiate.Carry out an incremental learning within every 120 seconds, in the period, each subject follows
Prompt in screen keeps a kind of gesture 5s, each gesture 3 times (i.e. 120 seconds) occur at random, and increment sample isHand
Gesture × 3 time totally 600.Use preliminary classification model ΨoldTo detect newly-increased sample set Dnew, by newly-increased sample set DnewIt is divided into full
The newly-increased training sample D of sufficient KKT conditionsnew-KKTWith the newly-increased training sample D for being unsatisfactory for KKT conditionsnew-NKKT.Differentiating method：If
Sample is located on class interval or outside class interval, to meet the newly-increased training sample of KKT conditions；If sample is located at class interval
It is interior, to be unsatisfactory for the newly-increased training sample of KKT conditions.
(4) DBSCAN Density Clusterings are used to obtain initial training collection.The newly-increased training sample of KKT conditions will be met
Dnew-KKT, initial SV collection Dold-SVWith initial non-SV collection Dold-NSVIt puts together, carries out Density Clustering using DBSCAN, obtain multiple
Clustering cluster Cm.It is initial SV collection D to select kernel objectold-SVThe clustering cluster C of middle samplem, these clustering clusters CmIn include initial SV
Initial non-SV samples D ' near collectionold-NSVWith the newly-increased training sample D ' for meeting KKT conditions near initial SV collectionnew-KKT.Just
Beginning SV collection Dold-SV, initial non-SV samples D ' near initial SV collectionold-NSV, meet the newly-increased of KKT conditions near initial SV collection
Training sample D 'new-KKTWith the newly-increased training sample D for being unsatisfactory for KKT conditionsnew-NKKTConstitute initial training collection Dnew1。
Further, the DBSCAN Density Clusterings method determines cluster knot by the tightness degree of sample distribution
Structure.It analyzes the connectivity between sample based on one group of Neighbourhood parameter (ε, MINPts), not open close to cross extended clustering cluster
Obtain cluster result.ε indicates that the radius of neighbourhood, MINPts are that set point is counted in neighborhood as the minimum neighborhood of kernel object.
In the present embodiment, ε=0.11, MinPts=4.As shown in Fig. 2, the DBSCAN Density Clustering method and steps are as follows：
(4.1) kernel object collection is found out based on Neighbourhood parameter.Data-oriented collection E={ z1,z2,...zm, if distance is wherein
One sample zjMinPts sample is included at least in field no more than ε, i.e., | Nε(zj) | >=MinPts, then zjFor a core
Heart object.Kernel object collection Ω={ z can be found outa,zb,zc,...zn, a, b, c ... n ∈ m.
(4.2) " seed " kernel object is randomly selected from kernel object concentration, finds out the reachable sample of all density,
Constitute first clustering cluster.Assuming that selecting zaAs seed, based on this, then first clustering cluster is C1={ za,zc,...zp,
a,c,...p∈m。
(4.3) core sample for including in clustering cluster is concentrated from kernel object and is removed.I.e. by C1In include core sample
This is removed from Ω：
Ω '=Ω C1={ zb,...,zm}
Again from new set omega ' in randomly select another core sample generate second clustering cluster, constantly repeat, until
Core sample collection is sky.
(5) postsearch screening obtains final training set.To the initial non-SV samples D ' near initial SV collectionold-NSVWith initial SV
The newly-increased training sample D ' for meeting KKT conditions near collectionnew-KKTPostsearch screening is carried out to reduce training set number of samples to subtract
Few training time.Because too far with a distance from Optimal Separating Hyperplane, become that SV probabilities are smaller, and postsearch screening rule is：Calculate initial SV
Initial non-SV samples D ' near collectionold-NSVWith the newly-increased training sample D ' for meeting KKT conditions near initial SV collectionnew-KKTIn
For each sample to the distance of Optimal Separating Hyperplane, the maximum distance from hyperplane is δ, chooses distance in the two sample sets and is less than 1/
The sample of 2 δ obtains the initial non-SV samples D " near the initial SV collection after postsearch screeningold-NSVWith expiring near initial SV collection
The newly-increased training sample D " of sufficient KKT conditionsnew-KKT.By initial SV collection Dold-SV, it is initial near initial SV collection after postsearch screening
Non- SV samples D "old-NSV, the newly-increased training sample D " for meeting KKT conditions near initial SV collection after postsearch screeningnew-KKTNo
Meet the newly-increased training sample D of KKT conditionsnew-NKKTConstitute final training set Dnew2, for training point after the secondary incremental learning
Class model Ψnew。
The above is only a preferred embodiment of the present invention, it should be understood that the present invention is not limited to described herein
Form is not to be taken as excluding other embodiments, and can be used for other combinations, modifications, and environments, and can be at this
In the text contemplated scope, modifications can be made through the above teachings or related fields of technology or knowledge.And those skilled in the art institute into
Capable modifications and changes do not depart from the spirit and scope of the present invention, then all should be in the protection domain of appended claims of the present invention
It is interior.
Claims (8)
1. a kind of online gesture identification method for myoelectricity individual difference problem, which is characterized in that the method includes following
Step：
(1) gesture myoelectricity data set is established, and obtains primary data；
(2) training preliminary classification model；
(3) sample KKT is increased newly to differentiate；
(4) DBSCAN Density Clusterings are used to obtain initial training collection；
(5) postsearch screening obtains final training set.
2. a kind of online gesture identification method for myoelectricity individual difference problem as described in claim 1, which is characterized in that
In the step (1), the primary data is to acquire what 20 males were tested altogether to build the initial model of incremental learning
8 kinds of common gesture motions, respectively：Hand opens, holds fist, hand captures, hand loosens, receipts, wrist abduction, wrist outward turning and wrist in wrist
Inward turning.
3. a kind of online gesture identification method for myoelectricity individual difference problem as claimed in claim 2, which is characterized in that
In the step (1), the sample rate of myoelectricity is 1024Hz, is opened a window by the way of sliding window, length of window is set as 200ms, right
Each window carries out 3 kinds of temporal signatures extractions, respectively root-mean-square value RMS, average absolute value MAV and integral myoelectricity iEMG；One
A window can get one 3 × 4 totally 12 the feature vector [{ RMS tieed upk,MAVk,iEMGk}Ch], wherein k is sample number, and Ch is
Port number；Each subject follows the prompt in screen, keeps a kind of gesture 5s at random, each gesture occurs 30 times, then each quilt
The data set of examination isGesture × 30 time, wherein twindowFor the length that opens a window, then the data set being each tested
Total sample number is 6000.
4. a kind of online gesture identification method for myoelectricity individual difference problem as claimed in claim 3, which is characterized in that
In the step (2), for each subject, all training samples of 19 subjects of remaining in primary data are used for training the quilt
The SVM preliminary classification models Ψ of examinationoldWith initial SV collection Dold-SV。
5. a kind of online gesture identification method for myoelectricity individual difference problem as claimed in claim 4, which is characterized in that
In the step (3), an incremental learning is carried out within every 120 seconds, in the period, each subject follows the prompt in screen,
A kind of gesture 5s, each gesture is kept to occur at random 3 times, increment sample isGesture × 3 time totally 600, using initial
Disaggregated model ΨoldTo detect newly-increased sample set Dnew, by newly-increased sample set DnewIt is divided into the newly-increased training sample for meeting KKT conditions
Dnew-KKTWith the newly-increased training sample D for being unsatisfactory for KKT conditionsnew-NKKT；Differentiating method：If sample is located on class interval or classifies
Interval is outer, to meet the newly-increased training sample of KKT conditions；If sample is located in class interval, to be unsatisfactory for the newly-increased of KKT conditions
Training sample.
6. a kind of online gesture identification method for myoelectricity individual difference problem as claimed in claim 5, which is characterized in that
In the step (4), the newly-increased training sample D of KKT conditions will be metnew-KKT, initial SV collection Dold-SVWith initial non-SV collection
Dold-NSVIt puts together, carries out Density Clustering using DBSCAN, obtain multiple clustering cluster Cm；It is initial SV collection to select kernel object
Dold-SVThe clustering cluster C of middle samplem, these clustering clusters CmIn include the initial non-SV samples D ' near initial SV collectionold-NSVWith it is first
The newly-increased training sample D ' for meeting KKT conditions near beginning SV collectionnew-KKT；Initial SV collection Dold-SV, it is initial near initial SV collection
Non- SV samples D 'old-NSV, the newly-increased training sample D ' for meeting KKT conditions near initial SV collectionnew-KKTBe unsatisfactory for KKT conditions
Newly-increased training sample Dnew-NKKTConstitute initial training collection Dnew1。
7. a kind of online gesture identification method for myoelectricity individual difference problem as claimed in claim 6, which is characterized in that
In the step (4), the DBSCAN Density Clusterings method determines cluster structure by the tightness degree of sample distribution, it
The connectivity between sample is analyzed based on one group of Neighbourhood parameter (ε, MINPts), not open close extended clustering cluster of crossing is gathered
Class is as a result, take ε=0.11, MinPts=4；The DBSCAN Density Clustering method and steps are as follows：
(4.1) kernel object collection is found out based on Neighbourhood parameter；
(4.2) " seed " kernel object is randomly selected from kernel object concentration, finds out the reachable sample of all density, constituted
First clustering cluster；
(4.3) core sample for including in clustering cluster is concentrated from kernel object and is removed.
8. a kind of online gesture identification method for myoelectricity individual difference problem as claimed in claim 7, which is characterized in that
In the step (5), to the initial non-SV samples D ' near initial SV collectionold-NSVMeet KKT conditions near initial SV collection
Newly-increased training sample D 'new-KKTCarry out postsearch screening；Because too far with a distance from Optimal Separating Hyperplane, it is smaller to become SV probabilities, two
Secondary screening rule is：Calculate the initial non-SV samples D ' near initial SV collectionold-NSVMeet KKT conditions near initial SV collection
Newly-increased training sample D 'new-KKTIn each sample to the distance of Optimal Separating Hyperplane, maximum distance from hyperplane is δ, is chosen
Distance is less than the sample of 1/2 δ in the two sample sets, obtains the initial non-SV samples near the initial SV collection after postsearch screening
D″old-NSVWith the newly-increased training sample D " for meeting KKT conditions near initial SV collectionnew-KKT；By initial SV collection Dold-SV, secondary sieve
Initial non-SV samples D " near initial SV collection after choosingold-NSV, meet KKT conditions near initial SV collection after postsearch screening
Newly-increased training sample D "new-KKTWith the newly-increased training sample D for being unsatisfactory for KKT conditionsnew-NKKTConstitute final training set Dnew2, use
To train the disaggregated model Ψ after the secondary incremental learningnew。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810165350.1A CN108564105A (en) | 2018-02-28 | 2018-02-28 | Online gesture recognition method for myoelectric individual difference problem |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810165350.1A CN108564105A (en) | 2018-02-28 | 2018-02-28 | Online gesture recognition method for myoelectric individual difference problem |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108564105A true CN108564105A (en) | 2018-09-21 |
Family
ID=63531213
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810165350.1A Pending CN108564105A (en) | 2018-02-28 | 2018-02-28 | Online gesture recognition method for myoelectric individual difference problem |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108564105A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110547806A (en) * | 2019-09-11 | 2019-12-10 | 湖北工业大学 | gesture action online recognition method and system based on surface electromyographic signals |
| CN111178679A (en) * | 2019-12-06 | 2020-05-19 | 中能瑞通（北京）科技有限公司 | Phase identification method based on clustering algorithm and network search |
| CN118171117A (en) * | 2024-05-13 | 2024-06-11 | 浙江强脑科技有限公司 | Gesture training method and device for bionic hand, storage medium and bionic hand |
Citations (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101944122A (en) * | 2010-09-17 | 2011-01-12 | 浙江工商大学 | Incremental learning-fused support vector machine multi-class classification method |
| CN103605631A (en) * | 2013-11-20 | 2014-02-26 | 温州大学 | Increment learning method on the basis of supporting vector geometrical significance |
| CN103645804A (en) * | 2013-12-18 | 2014-03-19 | 三星电子（中国）研发中心 | Method and device for identifying human body gestures as well as watch using device |
| CN103861290A (en) * | 2014-02-27 | 2014-06-18 | 常州大学 | Myoelectricity remote control toy aircraft and control method thereof |
| CN104134060A (en) * | 2014-08-03 | 2014-11-05 | 上海威璞电子科技有限公司 | Sign language interpreting, displaying and sound producing system based on electromyographic signals and motion sensors |
| CN104537382A (en) * | 2015-01-12 | 2015-04-22 | 杭州电子科技大学 | Electromyographic signal gait recognition method for optimizing support vector machine based on genetic algorithm |
| CN105208746A (en) * | 2015-10-26 | 2015-12-30 | 刘鸿飞 | Cubic LED desk lamp based on motion and muscle current signal control and method |
| CN105243312A (en) * | 2015-11-13 | 2016-01-13 | 上海傲意信息科技有限公司 | Password system and encryption and decryption method |
| CN105426842A (en) * | 2015-11-19 | 2016-03-23 | 浙江大学 | Support vector machine based surface electromyogram signal multi-hand action identification method |
| CN105446484A (en) * | 2015-11-19 | 2016-03-30 | 浙江大学 | Electromyographic signal gesture recognition method based on hidden markov model |
| CN105893959A (en) * | 2016-03-30 | 2016-08-24 | 北京奇艺世纪科技有限公司 | Gesture identifying method and device |
| CN105919591A (en) * | 2016-04-12 | 2016-09-07 | 东北大学 | Surface myoelectrical signal based sign language recognition vocal system and method |
| CN106383579A (en) * | 2016-09-14 | 2017-02-08 | 西安电子科技大学 | EMG and FSR-based refined gesture recognition system and method |
| CN106980367A (en) * | 2017-02-27 | 2017-07-25 | 浙江工业大学 | A kind of gesture identification method based on myoelectricity topographic map |
| CN107273798A (en) * | 2017-05-11 | 2017-10-20 | 华南理工大学 | A kind of gesture identification method based on surface electromyogram signal |
| CN107480697A (en) * | 2017-07-12 | 2017-12-15 | 中国科学院计算技术研究所 | A kind of myoelectricity gesture identification method and system |
-
2018
- 2018-02-28 CN CN201810165350.1A patent/CN108564105A/en active Pending
Patent Citations (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101944122A (en) * | 2010-09-17 | 2011-01-12 | 浙江工商大学 | Incremental learning-fused support vector machine multi-class classification method |
| CN103605631A (en) * | 2013-11-20 | 2014-02-26 | 温州大学 | Increment learning method on the basis of supporting vector geometrical significance |
| CN103645804A (en) * | 2013-12-18 | 2014-03-19 | 三星电子（中国）研发中心 | Method and device for identifying human body gestures as well as watch using device |
| CN103861290A (en) * | 2014-02-27 | 2014-06-18 | 常州大学 | Myoelectricity remote control toy aircraft and control method thereof |
| CN104134060A (en) * | 2014-08-03 | 2014-11-05 | 上海威璞电子科技有限公司 | Sign language interpreting, displaying and sound producing system based on electromyographic signals and motion sensors |
| CN104537382A (en) * | 2015-01-12 | 2015-04-22 | 杭州电子科技大学 | Electromyographic signal gait recognition method for optimizing support vector machine based on genetic algorithm |
| CN105208746A (en) * | 2015-10-26 | 2015-12-30 | 刘鸿飞 | Cubic LED desk lamp based on motion and muscle current signal control and method |
| CN105243312A (en) * | 2015-11-13 | 2016-01-13 | 上海傲意信息科技有限公司 | Password system and encryption and decryption method |
| CN105426842A (en) * | 2015-11-19 | 2016-03-23 | 浙江大学 | Support vector machine based surface electromyogram signal multi-hand action identification method |
| CN105446484A (en) * | 2015-11-19 | 2016-03-30 | 浙江大学 | Electromyographic signal gesture recognition method based on hidden markov model |
| CN105893959A (en) * | 2016-03-30 | 2016-08-24 | 北京奇艺世纪科技有限公司 | Gesture identifying method and device |
| CN105919591A (en) * | 2016-04-12 | 2016-09-07 | 东北大学 | Surface myoelectrical signal based sign language recognition vocal system and method |
| CN106383579A (en) * | 2016-09-14 | 2017-02-08 | 西安电子科技大学 | EMG and FSR-based refined gesture recognition system and method |
| CN106980367A (en) * | 2017-02-27 | 2017-07-25 | 浙江工业大学 | A kind of gesture identification method based on myoelectricity topographic map |
| CN107273798A (en) * | 2017-05-11 | 2017-10-20 | 华南理工大学 | A kind of gesture identification method based on surface electromyogram signal |
| CN107480697A (en) * | 2017-07-12 | 2017-12-15 | 中国科学院计算技术研究所 | A kind of myoelectricity gesture identification method and system |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110547806A (en) * | 2019-09-11 | 2019-12-10 | 湖北工业大学 | gesture action online recognition method and system based on surface electromyographic signals |
| CN111178679A (en) * | 2019-12-06 | 2020-05-19 | 中能瑞通（北京）科技有限公司 | Phase identification method based on clustering algorithm and network search |
| CN118171117A (en) * | 2024-05-13 | 2024-06-11 | 浙江强脑科技有限公司 | Gesture training method and device for bionic hand, storage medium and bionic hand |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110070105B (en) | Electroencephalogram emotion recognition method and system based on meta-learning example rapid screening | |
| Li et al. | Boosting-based EMG patterns classification scheme for robustness enhancement | |
| CN110353673B (en) | Electroencephalogram channel selection method based on standard mutual information | |
| CN106980367A (en) | A kind of gesture identification method based on myoelectricity topographic map | |
| CN109934089A (en) | Multistage epileptic EEG Signal automatic identifying method based on supervision gradient lifter | |
| CN109480833A (en) | The pretreatment and recognition methods of epileptic's EEG signals based on artificial intelligence | |
| CN108960182A (en) | A kind of P300 event related potential classifying identification method based on deep learning | |
| CN108564105A (en) | Online gesture recognition method for myoelectric individual difference problem | |
| CN109583331B (en) | Deep learning-based accurate positioning method for positions of wrist vein and mouth of person | |
| CN114532993B (en) | Automatic detection method for electroencephalogram high-frequency oscillation signals of epileptic patients | |
| WO2024016590A1 (en) | Online neuron classification method based on neuromorphic computing | |
| CN113392733B (en) | Multi-source domain self-adaptive cross-tested EEG cognitive state evaluation method based on label alignment | |
| CN108256579A (en) | A kind of multi-modal sense of national identity quantization measuring method based on priori | |
| Li et al. | EEG signal classification method based on feature priority analysis and CNN | |
| CN109009098B (en) | Electroencephalogram signal feature identification method under motor imagery state | |
| CN113010013A (en) | Wasserstein distance-based motor imagery electroencephalogram migration learning method | |
| Yuan et al. | Chinese sign language alphabet recognition based on random forest algorithm | |
| CN114548165B (en) | Myoelectricity mode classification method capable of crossing users | |
| CN114159079B (en) | Multi-type muscle fatigue detection method based on feature extraction and GRU deep learning model | |
| CN113729738B (en) | Construction method of multichannel myoelectricity characteristic image | |
| CN114384999B (en) | User-independent myoelectric gesture recognition system based on self-adaptive learning | |
| CN103488995A (en) | Method for identifying rotation of neck | |
| CN109662710A (en) | A kind of EMG Feature Extraction based on convolutional neural networks | |
| CN110547806B (en) | Gesture action online recognition method and system based on surface electromyographic signals | |
| CN110738093B (en) | Classification method based on improved small world echo state network electromyography |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180921 |
|
| RJ01 | Rejection of invention patent application after publication |