CN107198509A - Feature extracting method and system based on surface myoelectric - Google Patents
Feature extracting method and system based on surface myoelectric Download PDFInfo
- Publication number
- CN107198509A CN107198509A CN201610741979.7A CN201610741979A CN107198509A CN 107198509 A CN107198509 A CN 107198509A CN 201610741979 A CN201610741979 A CN 201610741979A CN 107198509 A CN107198509 A CN 107198509A
- Authority
- CN
- China
- Prior art keywords
- mrow
- semg
- signal sequences
- feature
- active signal
- 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
Classifications
-
- 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
-
- 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
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Pathology (AREA)
- Heart & Thoracic Surgery (AREA)
- Psychiatry (AREA)
- Signal Processing (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biophysics (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Physiology (AREA)
- Medical Informatics (AREA)
- Molecular Biology (AREA)
- Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- General Health & Medical Sciences (AREA)
- Public Health (AREA)
- Veterinary Medicine (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The present invention provides a kind of feature extracting method and system based on surface myoelectric, and feature extracting method includes：SEMG signal sequences are received, the sEMG signal sequences include at least one active signal section；Signal threshold value is calculated according to the TKE operators for removing equalization signal of the sEMG signal sequences, the initial time of active signal section described in the sEMG signal sequences is detected according to the signal threshold value；The initial time of active signal section according to time threshold amendment obtains the beginning and ending time of the active signal section；The active signal section is extracted according to the beginning and ending time of active signal section；According to active signal section, at least one feature of the sEMG signal sequences is obtained.Feature extracting method based on surface myoelectric that the present invention is provided and system are come can be using gathering the feature that sEMG signal sequences obtain sEMG signal sequences exactly.
Description
Technical field
The present invention relates to field of signal processing, more particularly to a kind of feature extracting method and system based on surface myoelectric.
Background technology
Neuromuscular fatigue mechanism and the focus that forecasting research is domestic and international sports medical science research, while being also movement human
The emphasis of scientific research.In motion process, due to the keeping amount in blood, or nutriment lacks etc. can all make muscle
A series of changes occur for structure, metabolism and energy etc., can decline the efficiency of neuromuscular system, so that muscle can not continue
Completion task, causes muscular fatigue.Muscular fatigue may cause the muscular fatigue under muscle damage, serious conditions can not be extensive
It is multiple.The research of muscular fatigue is wide in the field such as ergonomics, man-machine interface, rehabilitation medical, injury gained in sports, artificial limb application prospect
It is general.
At present, the clinical detection instrument of muscular fatigue mainly have electromyographic signal (sEMG, surface electromyography),
Flesh sound (MMG, Mechanomyogram), vocal muscles figure (SMG, Sonomyography), near infrared spectrum (NIRS, Near-
Infrared spectroscopy), sound wave graphy figure (AMG, Acoustic myography), angular measurement sensor etc..Wherein,
It is the method commonly used in work physiology to be recorded using sEMG, study muscle, and as a kind of simple, hurtless measure, section is quantitative grinds
Study carefully method, it can study the variation characteristic during Local muscle fatigue, be a kind of accurate detection instrument.
In actual applications, many noises can be produced because the sEMG signal sequences of collection are affected by various factors, it is impossible to
It is accurate that feature is obtained according to the sEMG signal sequences gathered, and then detected.
The content of the invention
There is provided a kind of feature extraction side based on surface myoelectric in order to overcome the defect that above-mentioned prior art is present by the present invention
Method and system, the interactive training system based on surface myoelectric and the rehabilitation degree sort method based on surface myoelectric, to utilize
Collection sEMG signal sequences obtain the feature of sEMG signal sequences exactly.
The present invention provides a kind of feature extracting method based on surface myoelectric, including：SEMG signal sequences are received, it is described
SEMG signal sequences include at least one active signal section；Calculated according to the TKE for removing equalization signal of the sEMG signal sequences
Son calculates signal threshold value, when detecting the starting of active signal section described in the sEMG signal sequences according to the signal threshold value
Between；The initial time of active signal section according to time threshold amendment obtains the beginning and ending time of the active signal section；According to
The beginning and ending time of the active signal section extracts the active signal section；According to active signal section, the sEMG letters are obtained
At least one feature of number sequence.
Preferably, removing equalization signal and described going equalization to believe for the sEMG signal sequences is calculated according to equation below
Number TKE operators：
Wherein,Equalization signal is removed for the sEMG signal sequences, x (n) is the letter of the sEMG signal sequences
Number, ψ (n) removes equalization signal to be describedTKE operators, N be the sEMG signal sequences total length, M is described
The ambient noise length of sEMG signal sequences.
Preferably, the signal threshold value is calculated according to equation below：
Th=u0+j·δ0,
Wherein, μ0For ambient noise average of the sEMG signal sequences in TKE domains, δ0For the sEMG signal sequences
Background noise standard difference in TKE domains is used to calculate signal threshold value Th, j to be less than or equal to 30 parameter more than or equal to 5.
Preferably, judge whether the signal of the sEMG signal sequences belongs to the active signal section according to equation below：
Wherein, s (n) represents to judge sequence；
The initial time of the active signal section is calculated according to equation below：
Wherein,Represent the initial time of the active signal section.
Preferably, the initial time of active signal section obtains rising for the active signal section according to time threshold amendment
Only the step of time includes：If the time interval of adjacent s (n)=1 be less than very first time threshold value, by adjacent s (n)=1 it
Between value be all set to 1；And if the time interval of adjacent s (n)=0 is less than the second time threshold, by the adjacent s (n)
Value between=0 is all set to 0.
Preferably, at least one described feature includes：Maximum amplitude, energy, approximate entropy, integration myoelectricity value, mean power
One or more of frequency, median frequency.
According to another aspect of the invention, a kind of Feature Extraction System based on surface myoelectric is also provided, using as above institute
The feature extracting method based on surface myoelectric stated, the Feature Extraction System includes：Harvester, for gathering the sEMG
Signal sequence, the sEMG signal sequences include at least one active signal section；Processor, communicates with the harvester, institute
Stating processor includes：Reception device, for receiving the sEMG signal sequences；First computing device, for according to the sEMG
The TKE operators for removing equalization signal of signal sequence calculate signal threshold value, and detect that the sEMG believes according to the signal threshold value
The initial time of active signal section described in number sequence；Correcting device, for the active signal section according to time threshold amendment
Initial time obtain beginning and ending time of active signal section；Active signal section extraction element, for being believed according to the activity
The beginning and ending time of number section extracts the active signal section；Feature deriving means, for according to active signal section, obtaining described
At least one feature of sEMG signal sequences.
Preferably, the harvester is one or more electrode patch.
According to another aspect of the invention, a kind of interactive training system based on surface myoelectric is also provided, including：It is multiple to adopt
Acquisition means, the sEMG signal sequences for gathering multiple objects respectively, each sEMG signal sequences include at least one activity
Signal segment；At least one processor, including：Extraction element, using the feature extracting method based on surface myoelectric as described above
Extract at least one feature of each sEMG signal sequences；Interactive device, for providing artificial scene to multiple objects respectively,
And at least one feature of each sEMG signal sequences extracted according to the extraction element, carried out in the artificial scene
Interaction, the multiple object is located in same artificial scene.
Preferably, each harvester gathers the sEMG signal sequences of an object.
Preferably, the multiple harvester and a processor communication, the interactive training system also include：It is many
Individual display device, with the processor communication, shows the artificial scene to multiple objects respectively.
Preferably, the multiple harvester, multiple display devices and the processor are located in the same space.
Preferably, a part for the multiple harvester and the multiple display device is located at the first space, described many
Individual harvester and another part of the multiple display device are located at second space, and it is empty that the processor is located at described first
Between, the second space or the 3rd space.
Multiple harvesters, multiple display devices and the processor for being preferably located at different spaces lead to
Wireless mode is crossed to be communicated.
Preferably, the processor also includes：Role's distributor, for distributing different roles to each object.
Preferably, the interactive device provides multiple artificial scenes, each artificial scene correspondence sEMG signals sequence
One or more features of row.
According to another aspect of the invention, a kind of rehabilitation degree sort method based on surface myoelectric is also provided, including：Using
Feature extracting method as described above extracts described at least one feature of sEMG signal sequences；Will at least one feature input
Housebroken Z weak regression models, Z is the integer more than 1；Correspondence institute is obtained according to the output valve of the Z weak regression models
State the rehabilitation degree of sEMG signal sequences.
Preferably, the Z weak regression models are trained according to following manner：Obtain multiple sEMG signal sequences and right
Answer the known rehabilitation degree of each sEMG signal sequences；Extract multiple active signals section of multiple sEMG signal sequences；Root
Feature is extracted according to each active signal section；It regard the Q active signal section in the multiple active signal section as training set, Q
For the integer more than 1；Using the feature of each active signal section in the training set as the Z weak regression models input；Will
The known rehabilitation degree of the sEMG signal sequences where each active signal section of correspondence is used as the defeated of the individual weak regression models of the Z
Go out；And train the Z weak regression model.
Preferably, the Z weak regression model is tested according to following manner：By P in the multiple active signal section
Active signal section is as test set, and P is the integer more than 1；Using the feature of each active signal section in the test set as through instruction
The input of the Z experienced weak regression models；The rehabilitation of the sEMG signal sequences where each active signal section of output correspondence
Degree, the rehabilitation degree is compared with the known rehabilitation degree.
Preferably, the rehabilitation degree of the correspondence sEMG signal sequences is obtained according to the output valve of the Z weak regression models
Including：Output valve to the Z weak regression models is weighted the average rehabilitation to draw the correspondence sEMG signal sequences
Degree.
Preferably, in addition to：The sEMG signal sequences are ranked up by the rehabilitation degree.
According to another aspect of the invention, a kind of combined type interactive training system based on surface myoelectric is also provided, including：
Multiple harvesters, the sEMG signal sequences for gathering multiple objects respectively, each sEMG signal sequences include at least one
Individual active signal section；At least one processor, including：Extraction element, is carried using the feature based on surface myoelectric as described above
Method is taken to extract at least one feature of each sEMG signal sequences；Interactive device, for providing empty to multiple objects respectively
Intend scene, and at least one feature of each sEMG signal sequences extracted according to the extraction element, in the virtual feelings
Interaction is carried out in scape, the multiple object is located in same artificial scene；Rehabilitation degree computing device, for by the extraction element
At least one feature of each sEMG signal sequences extracted inputs housebroken Z weak regression model, and according to the Z
The output valve of weak regression model obtains the rehabilitation degree of the correspondence sEMG signal sequences.
Preferably, in addition to：Adjusting apparatus, for described in the rehabilitation degree adjustment that is calculated according to the rehabilitation degree computing device
The artificial scene that interactive device is provided.
Compared with prior art, the present invention has following advantage：
1) the TKE operators of equalization signal are gone to obtain the active segment of sEMG signal sequences by sEMG signal sequences, with this
The motor segment of acquisition target is represented, and then removes the sEMG signal sequences of non-athletic section exactly, obtains and accurately represents object fortune
Dynamic sEMG signal sequences, and then the feature for accurately representing object motion parameter is extracted from sEMG signal sequences；
2) interactive training system based on surface myoelectric is provided, it is special according to the sEMG signal sequences for representing object motion parameter
Levy there is provided the interactive training under artificial scene, and support the interaction that many objects are long-range or not remote is under same artificial scene
Training；
3) the rehabilitation degree ordering system based on surface myoelectric is provided, the sEMG signal sequences for representing object motion parameter is special
The housebroken weak regression model of input is levied, accurate object rehabilitation degree can be exported automatically；
4) combine above-mentioned interactive training system and rehabilitation degree is calculated, interactive training system is controlled by rehabilitation degree result of calculation
The scene provided, and then the interactive training of the current rehabilitation degree of suitable object can be provided.
Brief description of the drawings
Its example embodiment is described in detail by referring to accompanying drawing, above and other feature of the invention and advantage will become
It is more obvious.
Fig. 1 shows the flow chart of the feature extracting method according to embodiments of the present invention based on surface myoelectric.
Fig. 2 shows the oscillogram of initial sEMG signal sequences according to embodiments of the present invention.
Fig. 3 shows the process obtained according to the active signal section for sEMG signal sequences in part in Fig. 2.
Fig. 4 shows that sEMG signal sequences according to embodiments of the present invention extract the oscillogram after active signal section.
Fig. 5 shows the schematic diagram of the Feature Extraction System according to embodiments of the present invention based on surface myoelectric.
Fig. 6 shows the schematic diagram of the interactive training system according to an embodiment of the invention based on surface myoelectric.
Fig. 7 shows the schematic diagram of the interactive training system in accordance with another embodiment of the present invention based on surface myoelectric.
Fig. 8 shows the flow chart of the interactive training according to embodiments of the present invention based on surface myoelectric.
Fig. 9 shows the oscillogram of sEMG signal sequences according to embodiments of the present invention.
Figure 10 shows the oscillogram of sEMG signal sequences according to embodiments of the present invention.
Figure 11 shows the flow chart of the rehabilitation degree sort method according to embodiments of the present invention based on surface myoelectric.
Figure 12 shows the flow chart of the weak regression model of training according to embodiments of the present invention.
Figure 13 shows the schematic diagram of weak regression model according to embodiments of the present invention.
Figure 14 shows the flow chart of the weak regression model of test according to embodiments of the present invention.
Figure 15 shows the flow chart of the rehabilitation degree sort method according to embodiments of the present invention based on surface myoelectric.
Figure 16 shows the schematic diagram of the interactive training system based on surface myoelectric according to yet another embodiment of the invention.
Embodiment
Example embodiment is described more fully with referring now to accompanying drawing.However, example embodiment can be with a variety of shapes
Formula is implemented, and is not understood as limited to embodiment set forth herein；On the contrary, thesing embodiments are provided so that the present invention will
Fully and completely, and by the design of example embodiment those skilled in the art is comprehensively conveyed to.Identical is attached in figure
Icon note represents same or similar structure, thus will omit repetition thereof.
Described feature, structure or characteristic can be combined in one or more embodiments in any suitable manner
In.Embodiments of the present invention are fully understood so as to provide there is provided many details in the following description.However,
One of ordinary skill in the art would recognize that, without one or more in specific detail, or using other methods, constituent element, material
Material etc., can also put into practice technical scheme.In some cases, be not shown in detail or describe known features, material or
Person's operation is fuzzy of the invention to avoid.
The accompanying drawing of the present invention is only used for illustrating the size of element in relative position relation, accompanying drawing not represent actual size
Proportionate relationship.
In order to solve problem of the prior art, the present invention provides a kind of feature extracting method based on surface myoelectric.First
Referring to Fig. 1, Fig. 1 shows the flow chart of the feature extracting method according to embodiments of the present invention based on surface myoelectric.In Fig. 1
In, for showing 5 steps：
S110：SEMG signal sequences are received, sEMG signal sequences include at least one active signal section.
Specifically, sEMG signal sequences are obtained using the harvester of such as electrode patch.In further embodiments,
After sEMG signal sequences are obtained by harvester, it is stored in storage device, when carrying out feature extraction, from storage device
Obtain sEMG signal sequences.
The oscillogram of the sEMG signal sequences gathered is as shown in Figure 2.One section of sEMG signal sequence can represent to be gathered
The motion of object for a period of time.Motion of the object within a period of time would generally include at least one action.When there is multiple repetitions
Pause is had during action, between action.Therefore, represent that the sEMG signal sequences of object motion can include at least one active signal
Section.Each active signal segment table shows the one-off of object.
When sEMG signal sequences are gathered, when object is not acted, noise generation is also had.In order to more accurately extract
The features of sEMG signal sequences does not do making an uproar of acting, it is necessary to extract the active signal section in sEMG signal sequences to remove object
Sound.Following step is performed to carry out the extraction of active signal section.
S120：Signal threshold value is calculated according to the TKE operators for removing equalization signal of the sEMG signal sequences, according to described
Signal threshold value detects the initial time of active signal section described in the sEMG signal sequences.For TKE operators, its predecessor TE is calculated
Sub (Teager energy operator) is proposed by Teager in nineteen eighty-three earliest, for the voice based on linear theory at that time
Model, non-linear process is included with experimental verification in voice signal.In nineteen ninety, Kaiser is exported on the basis of TE operators
TKE operators calculate the energy of the audio signal of Disgrete Time Domain and continuous time.TKE operators are mainly used to detect rising for signal
Stop.
Specifically, removing equalization signal and removing equalization signal for sEMG signal sequences is calculated according to equation below
TKE operators：
Wherein,Equalization signal is removed for sEMG signal sequences, x (n) is the signal of sEMG signal sequences, and ψ (n) is
Remove equalization signalTKE operators, for the total lengths of the sEMG signal sequences, (length is multiplied by N for the sampling time herein
Sample rate, such as sample rate are 100Hz, and the sampling time is 3 seconds, then total length is exactly that 300), M is the back of the body of sEMG signal sequences
Scape noise length (generally select muscle activity initiate before 450-500ms time in sEMG signals as background noise).
Then, signal threshold value is calculated according to equation below：
Th=u0+j·δ0,
Wherein, μ0The ambient noise average for being sEMG signal sequences in TKE domains, δ0It is sEMG signal sequences in TKE domains
Background noise standard difference be used to calculating signal threshold value Th, j is is less than or equal to 30 parameter more than or equal to 5.
Next, judging whether the signal of sEMG signal sequences belongs to active signal section according to equation below, that is, sentence
The signal of disconnected sEMG signal sequences indicates whether that object is acted：
Wherein, s (n) represents to judge sequence, and s (n)=1 represents that the signal of sEMG signal sequences at n belongs to active signal section,
S (n)=0 represents that the signal of sEMG signal sequences at n is not belonging to active signal section, in other words, and s (n)=1 represents that object is just at n
Acted, s (n)=0 represents that object is without action at n.
Afterwards, the initial time of active signal section is calculated according to equation below：
Wherein,Represent the initial time of active signal section.
It for details, reference can be made to the waveform of the top first in Fig. 3, Fig. 3 and illustrate the portion irised out in Fig. 2 sEMG signal sequences
Sub-signal sequence.The sEMG signal sequences of the part are performed after step S120, second oscillogram on Fig. 3 is can obtain.Second
Individual oscillogram represents to correspond to the judgement sequence s (n) of the sEMG signal sequences that first waveform is illustrated in Fig. 3.
S130：The beginning and ending time of active signal section is obtained according to the initial time of time threshold amendment active signal section.
Because the noise included in sEMG signal sequences is relatively more, it is possible to by the fluctuating of signal in normal muscle activity
It is mistaken for no muscle activity to occur, normally muscle activity will be mistaken for s (n)=0；Simultaneously, it is also possible to when will be inactive
Spike noise be mistaken for muscle activity, i.e., be mistaken for s (n)=1 by inactive.In order to realize to once lasting muscle activity
Accurate judgement, it is necessary to " 0 " and " 1 " judged in s (n) sequences is carried out some merge with rejecting handle.
Specifically, it should be handled by following two steps：
If the time interval of adjacent s (n)=1 is less than very first time threshold value (such as 300ms), by adjacent s (n)=1
Between value be all set to 1.For example, in Fig. 3 in second oscillogram, the time interval Ta of s (n)=1 is less than the very first time
Threshold value, then be set to 1 by the value of all s (n) in time interval Ta.After being performed by the step, reference can be made to the 3rd ripple in Fig. 3
Shape figure.Pass through the step, it is to avoid the non-activity that the sEMG signal sequences during muscle activity rise and fall is judged by accident.
If the time interval of adjacent s (n)=0 is less than the second time threshold (such as 100ms), by adjacent s (n)=0
Between value be all set to 0.For example, in Fig. 3 in the 3rd oscillogram, the time interval Tb of s (n)=0 was less than for the second time
Threshold value, then be set to 0 by the value of all s (n) in time interval Tb.After being performed by the step, reference can be made to the 4th ripple in Fig. 3
Shape figure.The step is crossed, the muscle activity erroneous judgement of the spike noise of sEMG signal sequences is eliminated.
S140：The active signal section is extracted according to the beginning and ending time of active signal section.
The judgement sequence s (n) of the whole sEMG signal sequences of correspondence oscillogram can be obtained after above-mentioned steps S130, referring to
An oscillogram on Fig. 4.According to the value for judging sequence s (n), retain correspondence s (n)=1 sEMG signal sequences, remove correspondence
The sEMG signal sequences of s (n)=0.To obtain the active signal section of sEMG signal sequences.
S150：According to active signal section, at least one feature of sEMG signal sequences is obtained.
For the movable function state of accurate, comprehensive reaction muscle, the redundancy of information is reduced, then is needed to sEMG signals
Sequence carries out feature extraction.The feature that can be extracted includes maximum amplitude, energy, approximate entropy, integration myoelectricity value, mean power frequency
One or more of rate, median frequency.
Specifically, maximum amplitude (AMP) can be extracted according to equation below：
AMP=max (xi) (i=1,2 ..., N),
Wherein, N represents the length of sEMG signal sequences, xiRepresent the signal value at sEMG signal sequence length i.
Energy (E) can be calculated according to equation below：
Wherein, N represents the length of sEMG signal sequences, xiRepresent the signal value at sEMG signal sequence length i.
Approximate entropy (Approximation Entropy, ApEn) is one and is used for quantitative description signal sequence (time sequence
Row) complexity nonlinear kinetics parameter.It represents the complexity of time series with a nonnegative number, with it is irregular when
Between the corresponding approximate entropy of sequence it is bigger.The calculation procedure of approximate entropy is as follows：
(1) reconstruct m ties up phase space in order：
X (i)=[x (i), x (i+1) ..., x (i+m-1)],
Wherein, X (i) is m dimension phase space vectors, and x (i) represents the signal value at sEMG signal sequence length i, 1≤i≤N-
M+1, m are the integer between 5 to 30.
(2) the distance between vector X (i) and X (j) d is calculatedij：
dij=max | x (i+k)-x (j+k) |,
Wherein, 0≤k≤m-1,1≤i, j≤N-m+1.
(3) similar tolerance limit r ＞ 0 are selected, to each X (i), statistical distance dij≤ r number, and calculate the number and vector
The ratio of sum
(4) by ratioTake the logarithm, then ask it for all i average value φm(r)：
(5) dimension m is increased by 1, repeats the above steps (1) to (4), try to achieveAnd φm+1(r)。
(6) approximate entropy ApEn is calculated according to equation below：
(7) if N is finite value, approximate entropy ApEn is obtained by statistical value estimation, i.e.,：
ApEn (m, r, N)=Φm(r)-Φm+1(r)。
In addition to the features such as maximum amplitude, energy, approximate entropy, integration myoelectricity value (iEMG), mean power can also be extracted
The feature such as frequency (MPF) and median frequency (MF).
Integration myoelectricity value iEMG can characterize the strength size produced during muscle activity, be calculated as follows：
Wherein, x (i) (i=0,1,2 ..., N) is the sEMG signal sequences in the active signal section that length is N.
Median frequency (MF) and frequency of average power (MPF) can characterize the fatigue conditions in muscle contraction, with
Muscular fatigue develops, and the two is on a declining curve.
Median frequency fmfIt is calculated as follows：
Frequency of average power fmeanIt is calculated as follows：
Wherein, it please provide in above-mentioned two formula, the implication of each symbol.
SEMG signal sequences can be extracted extremely according to active signal section exactly by above-mentioned steps S110 to step S150
A few feature.
Correspondence features described above extracting method, the present invention also provides a kind of Feature Extraction System, referring to Fig. 5.Fig. 5 shows root
According to the schematic diagram of the Feature Extraction System based on surface myoelectric of the embodiment of the present invention.In Feature Extraction System shown in Fig. 5, bag
Include harvester 210 and processor 220.Harvester 210 is preferably the electrode patch of acquisition target muscle activity information.Adopt
Acquisition means 210 are used to gather sEMG signal sequences.SEMG signal sequences include at least one activity for representing that object is acted
Signal segment.
Processor 220 communicates with harvester 210.Preferably, processor 220 is filled by way of wired connection with collection
210 communications are put, in some change case, processor 220 wirelessly can also communicate with harvester 210.It can adopt
Wireless communication technology includes but is not limited to bluetooth, ZigBee, LAN, internet, RFID etc..
Processor 220 includes reception device 221, the first computing device 222, correcting device 223, active signal section and extracts dress
Put 224 and feature deriving means 225.
Reception device 221 communicates with harvester 210, and for receiving sEMG signal sequences.First computing device 222
The sEMG signal sequences received according to reception device 221, are calculated.First computing device 222 performs step S120 shown in Fig. 1.
That is the first computing device 222 is used to calculate signal threshold according to the TKE operators for removing equalization signal of sEMG signal sequences
Value, and according to the initial time of active signal section in signal threshold value detection sEMG signal sequences.Correcting device 223 is according to the first meter
Calculate the beginning and ending time of the initial time amendment active signal section for the active signal section that device 222 is calculated.Correcting device 223 is performed
Step S130 shown in Fig. 1.That is correcting device 223 is used to obtain according to the initial time of time threshold amendment active signal section
Obtain the beginning and ending time of active signal section.Active signal section extraction element 224 performs the step S140 shown in Fig. 1, is repaiied for basis
The beginning and ending time for the active signal section that equipment 223 is corrected extracts active signal section.Feature deriving means 225 are performed shown in Fig. 1
Step S150, for the active signal section extracted according to active signal section extraction element 224, obtain sEMG signal sequences extremely
A few feature.
Fig. 5 only shows the partial devices of processor 220, in some change case, and processor 220 can also include performing
The device of other functions.In addition, each device of processor can be independent device in Fig. 5, or it is integrated in same device,
Those skilled in the art can realize different change case, will not be described here.
The present invention also puies forward a kind of interactive training system based on surface myoelectric, referring to Fig. 6 and Fig. 7.Fig. 6 and Fig. 7 show this
Two embodiments of the interactive training system based on surface myoelectric provided are provided.
Referring first to Fig. 6, Fig. 6 shows the interactive training system 300 based on surface myoelectric of first embodiment of the invention.Hand over
Mutual training system 300 includes harvester 310 and processor 320.
Harvester 310 is preferably the electrode patch of acquisition target muscle activity information, for gathering some object
SEMG signal sequences.Alternatively, harvester 310 is integrated in a rehabilitation training equipment.Processor 320 and harvester
310 communications.Preferably, processor 320 can receive the sEMG that harvester 310 is gathered by built-in or peripheral hardware reception device
Signal sequence.
Processor 320 includes extraction element 321 and interactive device 322.Extraction element 321 is carried using the feature shown in Fig. 1
Method is taken to extract at least one feature of the sEMG signal sequences.Interactive device 322 is used to provide artificial scene to object, and
At least one feature of the sEMG signal sequences extracted according to extraction element, is carried out interactive in artificial scene.
Referring then to Fig. 7, Fig. 7 shows the interactive training system 400 based on surface myoelectric of second embodiment of the invention.Hand over
Mutual training system 400 include multiple harvesters (including harvester 410A, 410B and 410C), multiple display devices (including
Display device 430A, 430B and 430C) and processor 420.
Multiple harvesters are used for the sEMG signal sequences for gathering multiple objects respectively.Each harvester and a display are filled
Put correspondence.For example, harvester 410A correspondence display devices 430A；Harvester 410B correspondence display devices 430B；Collection dress
Put 410C correspondence display devices 430C.In other words, for each object, a harvester gathers the sEMG signal sequences of the object
Row, a display device is to the object display image.Preferably, corresponding harvester and display device are integrated in a rehabilitation
Train in equipment.Processor 420 is communicated with multiple harvesters respectively.Preferably, processor 420 can pass through built-in or peripheral hardware
Reception device receive harvester collection sEMG signal sequences.
Processor 420 includes extraction element 421, role's distributor 422 and interactive device 323.Extraction element 421 is used
Feature extracting method shown in Fig. 1 extracts at least one feature of the sEMG signal sequences.Role's distributor 422 is not according to
Same harvester or the feature of different sEMG signal sequences distributes different roles to object.Interactive device 423 is with showing
Showing device is communicated, and provides artificial scene, and the sEMG signals extracted according to extraction element to object by display device
At least one feature of sequence, carries out interaction, multiple objects are with different roles in same artificial scene in artificial scene
Carry out interactive.
In a second embodiment, multiple harvesters, multiple display devices and processor are located in the same space.So
Embodiment in, harvester, display device and processor can near radio connection or wired connection by way of enter
Row communication.
In a change case of second embodiment, harvester 410A and display device 430A are located in the first space；
Harvester 410B and display device 430B is located in second space；It is empty that harvester 410C and display device 430C is located at the 3rd
Between in.And processor can be located at above-mentioned first space, second space, the 3rd space or different from any of the above-described space the
In four spaces.Above-mentioned first space, second space, the 3rd space and the 4th space are different spaces.Harvester, display dress
Put and processor can be communicated by way of the connection of the remote-wireless such as internet.
Above-mentioned Fig. 6 and Fig. 7 diagrammatically only show the schematic diagram of the interactive training system based on surface myoelectric, this area
Technical staff can also realize more change case, for example, the quantity of harvester, the quantity of display device, each object institute
Other elements in the quantity of harvester that needs, processor etc..These changes are all within the scope of the present invention.
Specifically, the flow chart for the interactive training that the present invention is provided may refer to Fig. 8.Fig. 8 shows 4 steps altogether：
It is step S510 first, harvester gathers the sEMG signal sequences of multiple objects.For example, collection acquisition sEMG1,
sEMG2、...、sEMGN.Gather the multiple sEMG signal sequences sEMG1 obtained, sEMG2 ..., sEMGN entered by step S520
Row pretreatment, pre-processes the signal transacting such as the extraction for including but is not limited to active signal section, denoising, smooth.After pretreatment, perform
Step S530, step S530 is similar with step S150, and feature extraction is carried out according to the sEMG signal sequences after processing.Perform afterwards
Step, S540 distributes different roles for different objects, and carries out artificial scene interaction.For example, dividing to sEMG1 object
With role 1；Role 2 is distributed to sEMG2 object；To sEMGN object distribution role N.Multiple roles are in same artificial scene
In interact.
Further, the interactive device of the interactive training system based on surface myoelectric can provide multiple artificial scenes, respectively
One or more features of artificial scene correspondence sEMG signal sequences.
By taking inflating ball as an example, it is assumed that have multiple virtual roles (virtual arm) in virtual feelings, while there is multiple float in the air
Balloon, the height of balloon is different, and the corresponding score value for breaking balloon is also different.Electrode patch is attached to object arm
On, object lifts the height of arm to realize impact balloon by side, and target was competed in the defined time (for example in one minute),
To compare the height (number of the fraction=impact balloon × corresponding score value of each balloon) for obtaining fraction.
In certain experiment, sEMG signal sequences such as Fig. 9 when two objects complete sides lift arm action and as indicated by 10.From
It can visually see in Fig. 9, the sEMG signal sequence maximum amplitudes of the object are about 100uV or so, and object in Figure 10
SEMG signal maximum amplitudes are up to 250-300uV.It therefore, it can control in artificial scene not by this feature of maximum amplitude
Whether level balloon is breached.
The present invention also provides a kind of rehabilitation degree sort method based on surface myoelectric, referring to Figure 11.Figure 11 shows basis
The flow chart of the rehabilitation degree sort method based on surface myoelectric of the embodiment of the present invention.3 steps are shown altogether in fig. 11：
Step S610：At least one feature of sEMG signal sequences is extracted using feature extracting method as shown in Figure 1.
Step S620：At least one feature of sEMG signal sequences is inputted into housebroken Z weak regression model.Z can be with
It is the integer between 5 to 100.Specifically, the Z weak regression models are the extreme learning machines based on random forests algorithm to build
Vertical.
Step S630：The rehabilitation degree of correspondence sEMG signal sequences is obtained according to the output valve of the Z weak regression models.Tool
For body, because Z weak regression models export multiple rehabilitation degree, it is therefore desirable to the output valve to the Z weak regression models
It is weighted the average rehabilitation degree to draw the correspondence sEMG signal sequences.
The step of can also including being ranked up sEMG signal sequences according to rehabilitation degree after step S630.
Specifically, above-mentioned Z weak regression models are trained according to mode as shown in figure 12：
Step S601：Obtain the known rehabilitation degree of multiple sEMG signal sequences and each sEMG signal sequences of correspondence.
Specifically, above-mentioned known rehabilitation degree is the rehabilitation degree that doctor judges according to patient (object) motion conditions.Correspondence
In an object, its sEMG signal sequence one known rehabilitation degree of correspondence.
Step S602：Extract multiple active signals section of multiple sEMG signal sequences.
Active signal section can be extracted according to step S120 shown in Fig. 1 to step S140.
Step S603：Feature is extracted according to each active signal section.
The feature that can be extracted includes maximum amplitude, energy, approximate entropy, integration myoelectricity value, frequency of average power, intermediate value frequency
One or more of rate.
Step S604：Using the Q active signal section in multiple active signals section as training set, Q, which is less than or equal to activity, to be believed
The sum of number section.In general, Q can be the integer between 50 to 1000.
Step S605：Using the feature of each active signal section in training set as Z weak regression models input.
Step S606：The known rehabilitation degree of sEMG signal sequences where each active signal section of correspondence is individual weak time as Z
Return the output of model.
Step S607：Train Z weak regression models.
The training of one weak regression model may refer to Figure 13.
Assuming that between the neuron 720,731,740 of input layer 710 and hidden layer 730, hidden layer 730 and output layer 750
By the way of connecting entirely.Assuming that input layer 710 has D neuron, D input variable of correspondence be (feature namely inputted
Quantity), hidden layer 730 has R neuron, and output layer 750 has E neuron 740, and E output variable of correspondence is (in the present embodiment
In, E=1, the D feature of one sEMG signal sequence of correspondence exports a rehabilitation degree).Without loss of generality, it is assumed that input layer 710
Connection weight W between hidden layer 730 is：
Wherein, W is the matrix that R rows D is arranged, wjiRepresent j-th of god of i-th of neuron of input layer 710 720 and hidden layer 730
Through the connection weight between member 731, i is 1 to the integer between D, and j is 1 to the integer between R.
Assuming that the connection weight β between hidden layer 730 and output layer is：
Wherein, β is the matrix that R rows E is arranged, βjkRepresent k-th of god of j-th of neuron of hidden layer 730 731 and output layer 750
Through the connection weight between member 740, k is 1 to the integer between E.
Assuming that the threshold value b of hidden layer neuron 731 is：
Wherein, b is the matrix that R rows 1 are arranged.
Assuming that having Q active signal section (Q sample) in training set, then input matrix X and output matrix Y are respectively：
Wherein, X is the input matrix that D rows Q is arranged, and Y is the output matrix that E rows Q is arranged.
Assuming that the activation primitive of hidden layer neuron is g (x), activation primitive is used to transmit signal, then can be obtained according to Figure 13,
The output Y of output layer neuron is：
Y=[y1, y2..., yQ],
Wherein, Y is the matrix that E rows Q is arranged,
Wherein, wj=[wj1, wj2..., wjD], xl=[x1l, x2l..., xDl]T, herein subscript T represent transposed matrix.
Above formula is represented by：
H β=Y ',
Wherein, Y ' is matrix Y transposed matrix, and H is referred to as the output matrix (output of hidden layer of the hidden layer of neutral net
Matrix refers to that the output of each neuron of hidden layer is stitched together the matrix to be formed, and hidden layer output matrix is multiplied with β,
Can obtain the output of output layer neuron), specifically it is represented by：
When activation primitive g (x) infinitely can be micro-, (SLFN parameter refers to input layer and hidden layer god to SLFN parameter
Through the connection weight between the connection weight between member, the threshold value of hidden layer neuron, hidden layer and output layer neuron) not
Need all to be adjusted, W and b can be randomly choosed before training, and be kept in the training process constant.And hidden layer 730
Connection weight β between output layer 750 can be by solvingThe least square solution of equation group is obtained,
Its solution is：Wherein, H+For the output matrix H of hidden layer 730 Moore-penrose generalized inverses.
With reference to Figure 13, a weak regression model as shown in fig. 13 that can be trained in the manner described above, accordingly can profit
Z weak regression models are trained in a like fashion.
In a change case of the rehabilitation degree sort method based on surface myoelectric, after Z weak regression models of training, also wrap
The step of including test Z weak regression models of test, with specific reference to Figure 14：
Step S611：It regard the P active signal section in the multiple active signals extracted in step S603 section as test set.
Preferably, P subtracts Q for the sum of active signal section.For example, P can be the integer between 10 to 200.
Step S612：Using the feature of each active signal section in test set as housebroken Z weak regression models input.
Step S613：The rehabilitation degree of sEMG signal sequences where each active signal section of output correspondence, by rehabilitation degree and
Know that rehabilitation degree is compared.If the rehabilitation degree of Z weak regression model outputs of same sEMG signal sequences connects with known rehabilitation degree
Closely, then it can be determined that Z weak regression models complete training, if the health of Z weak regression model output of same sEMG signal sequences
Multiplicity differs greatly with known rehabilitation degree, then can continue Z weak regression models of training.
With reference next to Figure 15, Figure 15 shows the rehabilitation degree sequence side according to embodiments of the present invention based on surface myoelectric
The flow chart of method.
In the weak regression model training stage 910, step 911, sEMG signal sequences are obtained.Believe in step S912 for sEMG
Number sequence is pre-processed.The feature of sEMG signal sequences after step S913 extraction processs.By the feature of sEMG signal sequences
As the input of Z weak regression models, the known rehabilitation degree of sEMG signal sequences will be corresponded to as the output of Z weak regression models
(step S914).Step S915 is trained to Z weak regression models.
In the specific embodiment of weak regression model training stage 910, have 20 and be in the different rehabilitation stages
Object, allows them to complete side lift arm action 10 times, while recording the sEMG signal sequences in action executing process respectively.It is first
The active signal section of each action is first extracted, i.e. totally 200 sections of sEMG active signals section.And it is randomly divided into training set and test
Collection, wherein training set include 150 sections of sEMG active signals sections, and test set includes remaining 50 sections of sEMG active signals section.Then, pin
To every section of sEMG active signals section, corresponding feature (maximum amplitude, energy, approximate entropy, integration myoelectricity value, mean power are extracted
Frequency and median frequency etc.).Then, 10 weak regression models are set up using random forest method, in each weak regression model, instruction
Practice each feature of collection as the input of weak regression model, clinician is to each object assessment result (known rehabilitation degree) by stages
It is trained as the output of weak regression model.
The weak regression model training stage 910 can also include the model measurement stage, be lived for 50 sections of sEMG in test set
Dynamic signal segment, extracts the feature of each active signal section, then feature is inputted to 10 weak regression models respectively first, then will
The output of 10 weak regression models is averaged, so as to obtain testing rehabilitation degree.Test rehabilitation degree can be entered with known rehabilitation degree
Row contrast, to complete the test of weak regression model.
The stage 920 is obtained in rehabilitation degree, step 921, sEMG signal sequences are obtained.In step S922 for sEMG signal sequences
Row are pre-processed.The feature of sEMG signal sequences after step S923 extraction processs.The feature of sEMG signal sequences is exported into Z
Individual weak regression model, in step S924, rehabilitation degree is exported by Z weak regression models.
It will be appreciated by those skilled in the art that multiple weak regression model Z once enter in the above-mentioned weak regression model training stage 910
After training, the follow-up training that rehabilitation degree obtains stage 920 and weak regression model training stage 910 can be carried out synchronously, with weak
During the constantly training of regression model training stage 910 is ripe, more accurately rehabilitation degree is obtained.
The present invention also provides a kind of interactive training system of the combined type based on surface myoelectric, as shown in figure 16.Figure 16 institutes
The interactive training system 800 shown includes multiple harvesters, multiple display devices and at least one processor 820.
Multiple harvesters are used for the sEMG signal sequences for gathering multiple objects respectively.Each harvester and a display are filled
Put correspondence.For example, harvester 810A correspondence display devices 830A；Harvester 810B correspondence display devices 830B.In other words,
For each object, a harvester gathers the sEMG signal sequences of the object, and a display device is shown to the object to be schemed
Picture.Preferably, corresponding harvester and display device are integrated in a rehabilitation training equipment.Processor 820 respectively with it is many
Individual harvester communication.Preferably, processor 820 can receive what harvester was gathered by built-in or peripheral hardware reception device
SEMG signal sequences.
Processor 820 includes extraction element 821, interactive device 822, rehabilitation degree computing device 823 and adjusting apparatus 824.
Extraction element 821 extracts at least one feature of the sEMG signal sequences using the feature extracting method shown in Fig. 1.Interaction dress
822 are put to be communicated with display device, and by display device to object offer artificial scene, and extracted according to extraction element
At least one feature of sEMG signal sequences, carries out interaction, multiple objects are with different roles in same void in artificial scene
Intend carrying out interaction in scene.Rehabilitation degree computing device 823 performs the step shown in Figure 11, for extract extraction element 821
At least one feature of each sEMG signal sequences inputs housebroken Z weak regression model, and according to the defeated of the individual weak regression models of Z
Go out the rehabilitation degree that value obtains correspondence sEMG signal sequences.Adjusting apparatus 824 is used for the rehabilitation calculated according to rehabilitation degree computing device
The artificial scene that degree adjustment interactive device 822 is provided.For example, adjusting apparatus 824 is used for what is calculated according to rehabilitation degree computing device
Degree-of-difficulty factor and other specification in the artificial scene that rehabilitation degree adjustment interactive device 822 is provided.
Figure 16 only shows a kind of embodiment of the interactive training system of the combined type based on surface myoelectric, those skilled in the art
More change case can also be realized, for example, integrated, other function element additions of the quantity of each device, each device etc., this
A little changes are all within the scope of the present invention.
Compared with prior art, the present invention has following advantage：
1) the TKE operators of equalization signal are gone to obtain the active segment of sEMG signal sequences by sEMG signal sequences, with this
The motor segment of acquisition target is represented, and then removes the sEMG signal sequences of non-athletic section exactly, obtains and accurately represents object fortune
Dynamic sEMG signal sequences, and then the feature for accurately representing object motion parameter is extracted from sEMG signal sequences；
2) interactive training system based on surface myoelectric is provided, it is special according to the sEMG signal sequences for representing object motion parameter
Levy there is provided the interactive training under artificial scene, and support the interaction that many objects are long-range or not remote is under same artificial scene
Training；
3) the rehabilitation degree ordering system based on surface myoelectric is provided, the sEMG signal sequences for representing object motion parameter is special
The housebroken weak regression model of input is levied, accurate object rehabilitation degree can be exported automatically；
4) combine above-mentioned interactive training system and rehabilitation degree is calculated, interactive training system is controlled by rehabilitation degree result of calculation
The scene provided, and then the interactive training of the current rehabilitation degree of suitable object can be provided.
The illustrative embodiments of the present invention are particularly shown and described above.It should be understood that the invention is not restricted to institute
Disclosed embodiment, on the contrary, it is intended to cover comprising various modifications within the scope of the appended claims and equivalent put
Change.
Claims (23)
1. a kind of feature extracting method based on surface myoelectric, it is characterised in that including：
SEMG signal sequences are received, the sEMG signal sequences include at least one active signal section；
Signal threshold value is calculated according to the TKE operators for removing equalization signal of the sEMG signal sequences, according to the signal threshold value
Detect the initial time of the section of active signal described in the sEMG signal sequences；
The initial time of active signal section according to time threshold amendment obtains the beginning and ending time of the active signal section；
The active signal section is extracted according to the beginning and ending time of active signal section；
According to active signal section, at least one feature of the sEMG signal sequences is obtained.
2. feature extracting method as claimed in claim 1, it is characterised in that the sEMG signals sequence is calculated according to equation below
Row remove equalization signal and the TKE operators for removing equalization signal：
<mrow>
<mover>
<mi>x</mi>
<mo>~</mo>
</mover>
<mrow>
<mo>(</mo>
<mi>n</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mi>x</mi>
<mrow>
<mo>(</mo>
<mi>n</mi>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mfrac>
<mn>1</mn>
<mi>N</mi>
</mfrac>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>N</mi>
</munderover>
<mi>x</mi>
<mrow>
<mo>(</mo>
<mi>i</mi>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<mi>&psi;</mi>
<mrow>
<mo>(</mo>
<mi>n</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<msup>
<mover>
<mi>x</mi>
<mo>~</mo>
</mover>
<mn>2</mn>
</msup>
<mrow>
<mo>(</mo>
<mi>n</mi>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mover>
<mi>x</mi>
<mo>~</mo>
</mover>
<mrow>
<mo>(</mo>
<mi>n</mi>
<mo>+</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
<mover>
<mi>x</mi>
<mo>~</mo>
</mover>
<mrow>
<mo>(</mo>
<mi>n</mi>
<mo>-</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
<mo>,</mo>
<mi>n</mi>
<mo>=</mo>
<mn>1</mn>
<mo>,</mo>
<mn>2</mn>
<mo>,</mo>
<mo>...</mo>
<mo>,</mo>
<mi>M</mi>
<mo>,</mo>
<mo>...</mo>
<mo>,</mo>
<mi>N</mi>
<mo>,</mo>
</mrow>
Wherein,Equalization signal is removed for the sEMG signal sequences, x (n) is the signal of the sEMG signal sequences, ψ
(n) equalization signal is removed to be describedTKE operators, N be the sEMG signal sequences total length, M be the sEMG letter
The ambient noise length of number sequence.
3. feature extracting method as claimed in claim 2, it is characterised in that the signal threshold value is calculated according to equation below：
<mrow>
<msub>
<mi>&mu;</mi>
<mn>0</mn>
</msub>
<mo>=</mo>
<mfrac>
<mn>1</mn>
<mi>M</mi>
</mfrac>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>M</mi>
</munderover>
<mi>&psi;</mi>
<mrow>
<mo>(</mo>
<mi>i</mi>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<mtable>
<mtr>
<mtd>
<mrow>
<msub>
<mi>&delta;</mi>
<mn>0</mn>
</msub>
<mo>=</mo>
<msqrt>
<mrow>
<mfrac>
<mn>1</mn>
<mrow>
<mi>M</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</mfrac>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>M</mi>
</munderover>
<msup>
<mrow>
<mo>(</mo>
<mi>&psi;</mi>
<mo>(</mo>
<mi>i</mi>
<mo>)</mo>
<mo>-</mo>
<msub>
<mi>&mu;</mi>
<mn>0</mn>
</msub>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
</mrow>
</msqrt>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mi>T</mi>
<mi>h</mi>
<mo>=</mo>
<msub>
<mi>u</mi>
<mn>0</mn>
</msub>
<mo>+</mo>
<mi>j</mi>
<mo>&CenterDot;</mo>
<msub>
<mi>&delta;</mi>
<mn>0</mn>
</msub>
</mrow>
</mtd>
</mtr>
</mtable>
<mo>,</mo>
</mrow>
Wherein, μ0For ambient noise average of the sEMG signal sequences in TKE domains, δ0It is the sEMG signal sequences in TKE
Background noise standard difference in domain is used to calculate signal threshold value Th, j to be less than or equal to 30 parameter more than or equal to 5.
4. feature extracting method as claimed in claim 3, it is characterised in that the sEMG signals sequence is judged according to equation below
Whether the signal of row belongs to the active signal section：
<mrow>
<mtable>
<mtr>
<mtd>
<mrow>
<mi>s</mi>
<mrow>
<mo>(</mo>
<mi>n</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mi>s</mi>
<mrow>
<mo>(</mo>
<mi>&psi;</mi>
<mo>(</mo>
<mi>n</mi>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mi>T</mi>
<mi>h</mi>
<mo>)</mo>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mo>=</mo>
<mfenced open = "{" close = "">
<mtable>
<mtr>
<mtd>
<mn>1</mn>
</mtd>
<mtd>
<mrow>
<mi>&psi;</mi>
<mrow>
<mo>(</mo>
<mi>n</mi>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mi>T</mi>
<mi>h</mi>
<mo>></mo>
<mn>0</mn>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mn>0</mn>
</mtd>
<mtd>
<mrow>
<mi>&psi;</mi>
<mrow>
<mo>(</mo>
<mi>n</mi>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mi>T</mi>
<mi>h</mi>
<mo>&le;</mo>
<mn>0</mn>
</mrow>
</mtd>
</mtr>
</mtable>
</mfenced>
</mrow>
</mtd>
</mtr>
</mtable>
<mo>,</mo>
</mrow>
Wherein, s (n) represents to judge sequence；
The initial time of the active signal section is calculated according to equation below：
<mrow>
<msub>
<mover>
<mi>t</mi>
<mo>^</mo>
</mover>
<mn>0</mn>
</msub>
<mo>=</mo>
<mi>m</mi>
<mi>i</mi>
<mi>n</mi>
<mrow>
<mo>(</mo>
<mi>n</mi>
<mo>|</mo>
<mi>s</mi>
<mo>(</mo>
<mi>n</mi>
<mo>)</mo>
<mo>=</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
<mo>,</mo>
<mi>n</mi>
<mo>=</mo>
<mn>1</mn>
<mo>,</mo>
<mn>2</mn>
<mo>,</mo>
<mo>...</mo>
<mo>,</mo>
<mi>M</mi>
<mo>,</mo>
<mo>...</mo>
<mo>,</mo>
<mi>N</mi>
<mo>,</mo>
</mrow>
Wherein,Represent the initial time of the active signal section.
5. feature extracting method as claimed in claim 4, it is characterised in that the active signal section according to time threshold amendment
Initial time include the step of obtain the beginning and ending time of active signal section：
If the time interval of adjacent s (n)=1 is less than very first time threshold value, the value between adjacent s (n)=1 is all set
For 1；And
If the time interval of adjacent s (n)=0 is less than the second time threshold, the value between adjacent s (n)=0 is all set
For 0.
6. the feature extracting method as described in any one of claim 1 to 5, it is characterised in that at least one described feature includes：
One or more of maximum amplitude, energy, approximate entropy, integration myoelectricity value, frequency of average power, median frequency.
7. a kind of Feature Extraction System based on surface myoelectric, it is characterised in that using as described in any one of claim 1 to 6
The feature extracting method based on surface myoelectric, the Feature Extraction System includes：
Harvester, for gathering the sEMG signal sequences, the sEMG signal sequences include at least one active signal section；
Processor, communicates with the harvester, and the processor includes：
Reception device, for receiving the sEMG signal sequences；
First computing device, signal threshold value is calculated for the TKE operators for removing equalization signal according to the sEMG signal sequences,
And the initial time of active signal section described in the sEMG signal sequences is detected according to the signal threshold value；
Correcting device, the initial time for the active signal section according to time threshold amendment obtains the active signal section
Beginning and ending time；
Active signal section extraction element, for extracting the active signal section according to the beginning and ending time of active signal section；
Feature deriving means, for according to active signal section, obtaining at least one feature of the sEMG signal sequences.
8. Feature Extraction System as claimed in claim 7, it is characterised in that the harvester is one or more electrode pastes
Piece.
9. a kind of interactive training system based on surface myoelectric, it is characterised in that including：
Multiple harvesters, the sEMG signal sequences for gathering multiple objects respectively, each sEMG signal sequences are included extremely
Few active signal section；
At least one processor, including：
Extraction element, each institute is extracted using the feature extracting method based on surface myoelectric as described in any one of claim 1 to 6
State at least one feature of sEMG signal sequences；
Interactive device, for respectively to multiple objects provide artificial scene, and according to the extraction element extract it is each described in
At least one feature of sEMG signal sequences, carries out interaction in the artificial scene, and the multiple object is located at same virtual
In scene.
10. interactive training system as claimed in claim 9, it is characterised in that each harvester is gathered described in one
The sEMG signal sequences of object.
11. interactive training system as claimed in claim 10, it is characterised in that the multiple harvester and a place
Device communication is managed, the interactive training system also includes：
Multiple display devices, with the processor communication, show the artificial scene to multiple objects respectively.
12. interactive training system as claimed in claim 11, it is characterised in that the multiple harvester, multiple displays dress
Put and the processor is located in the same space.
13. interactive training system as claimed in claim 11, it is characterised in that the multiple harvester and the multiple aobvious
A part for showing device is located at the first space, and another part of the multiple harvester and the multiple display device is located at the
Two spaces, the processor is located at first space, the second space or the 3rd space.
14. interactive training system as claimed in claim 13, it is characterised in that positioned at multiple collection dresses of different spaces
Put, multiple display devices and the processor are wirelessly communicated.
15. the interactive training system as described in any one of claim 9 to 14, it is characterised in that the processor also includes：
Role's distributor, for distributing different roles to each object.
16. the interactive training system as described in any one of claim 9 to 14, it is characterised in that the interactive device provides many
Individual artificial scene, one or more features of each artificial scene correspondence sEMG signal sequences.
17. a kind of rehabilitation degree sort method based on surface myoelectric, it is characterised in that including：
Using at least one of sEMG signal sequences as described in the feature extracting method extraction as described in any one of claim 1 to 6
Feature；
At least one feature of the sEMG signal sequences is inputted into housebroken Z weak regression model, Z is the integer more than 1；
The rehabilitation degree of the correspondence sEMG signal sequences is obtained according to the output valve of the Z weak regression models.
18. rehabilitation degree sort method as claimed in claim 17, it is characterised in that the Z weak regression models are according to as follows
Mode is trained：
Obtain the known rehabilitation degree of multiple sEMG signal sequences and each sEMG signal sequences of correspondence；
Multiple active signals section of multiple sEMG signal sequences will be extracted；
Feature is extracted according to each active signal section；
Using the Q active signal section in the multiple active signal section as training set, Q is the integer more than 1；
Using the feature of each active signal section in the training set as the Z weak regression models input；
It regard the known rehabilitation degree of the sEMG signal sequences where each active signal section of correspondence as the Z weak regression models
Output；And
Train the Z weak regression models.
19. rehabilitation degree sort method as claimed in claim 18, it is characterised in that the Z is tested according to following manner individual weak
Regression model：
Using the P active signal section in the multiple active signal section as test set, P is the integer more than 1；
Using the feature of each active signal section in the test set as the housebroken Z weak regression models input；
The rehabilitation degree of the sEMG signal sequences where output correspondence each active signal section, by the rehabilitation degree with it is described known
Rehabilitation degree is compared.
20. rehabilitation degree sort method as claimed in claim 17, it is characterised in that according to the defeated of the individual weak regression models of the Z
Going out the rehabilitation degree of the value acquisition correspondence sEMG signal sequences includes：
Output valve to the Z weak regression models is weighted the average rehabilitation to draw the correspondence sEMG signal sequences
Degree.
21. rehabilitation degree sort method as claimed in claim 17, it is characterised in that also include：
The sEMG signal sequences are ranked up by the rehabilitation degree.
22. a kind of combined type interactive training system based on surface myoelectric, it is characterised in that including：
Multiple harvesters, the sEMG signal sequences for gathering multiple objects respectively, each sEMG signal sequences are included extremely
Few active signal section；
At least one processor, including：
Extraction element, each institute is extracted using the feature extracting method based on surface myoelectric as described in any one of claim 1 to 6
State at least one feature of sEMG signal sequences；
Interactive device, for respectively to multiple objects provide artificial scene, and according to the extraction element extract it is each described in
At least one feature of sEMG signal sequences, carries out interaction in the artificial scene, and the multiple object is located at same virtual
In scene；
Rehabilitation degree computing device, at least one feature of each sEMG signal sequences for the extraction element to be extracted is defeated
Enter housebroken Z weak regression models, and the correspondence sEMG signals sequence is obtained according to the output valve of the Z weak regression models
The rehabilitation degree of row.
23. combined type interactive training system as claimed in claim 22, it is characterised in that also include：
Adjusting apparatus, the rehabilitation degree for being calculated according to the rehabilitation degree computing device adjusts the virtual of the interactive device offer
Scene.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610741979.7A CN107198509B (en) | 2016-08-26 | 2016-08-26 | Feature extraction method and system based on surface myoelectricity |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610741979.7A CN107198509B (en) | 2016-08-26 | 2016-08-26 | Feature extraction method and system based on surface myoelectricity |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107198509A true CN107198509A (en) | 2017-09-26 |
| CN107198509B CN107198509B (en) | 2021-01-22 |
Family
ID=59904790
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610741979.7A Active CN107198509B (en) | 2016-08-26 | 2016-08-26 | Feature extraction method and system based on surface myoelectricity |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107198509B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110236538A (en) * | 2019-06-21 | 2019-09-17 | 西安交通大学 | A kind of automatic real-time detection method of movement starting point based on electromyography signal |
| CN111696645A (en) * | 2020-06-07 | 2020-09-22 | 长春理工大学 | Hand exoskeleton rehabilitation training device and method based on surface electromyographic signals |
| CN111723717A (en) * | 2020-06-11 | 2020-09-29 | 天津大学 | Silent voice recognition method and system |
| CN111973184A (en) * | 2019-05-22 | 2020-11-24 | 中国科学院沈阳自动化研究所 | Model training data optimization method for nonideal sEMG signals |
| CN112690808A (en) * | 2020-12-22 | 2021-04-23 | 浙江大学 | Human muscle fatigue identification method and system based on surface electromyographic signals and phase space reconstruction method |
| CN112773380A (en) * | 2019-11-07 | 2021-05-11 | 深圳市理邦精密仪器股份有限公司 | Electromyographic signal processing method, processing device and storage medium |
| CN114010453A (en) * | 2021-09-24 | 2022-02-08 | 阿凡达康复科技(东莞)有限公司 | Feedback wearing system based on motion signal and preprocessing method |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102846317A (en) * | 2012-07-18 | 2013-01-02 | 上海交通大学 | Wearing-type electromyographic signal acquisition system |
| CN103777752A (en) * | 2013-11-02 | 2014-05-07 | 上海威璞电子科技有限公司 | Gesture recognition device based on arm muscle current detection and motion sensor |
| KR20140107741A (en) * | 2013-02-28 | 2014-09-05 | 주식회사메디아나 | Information inputting device using an electromyogram signal, and method thereof |
| CN105769202A (en) * | 2016-04-29 | 2016-07-20 | 中国科学院苏州生物医学工程技术研究所 | Multi-body-position pressure balance evaluating system |
-
2016
- 2016-08-26 CN CN201610741979.7A patent/CN107198509B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102846317A (en) * | 2012-07-18 | 2013-01-02 | 上海交通大学 | Wearing-type electromyographic signal acquisition system |
| KR20140107741A (en) * | 2013-02-28 | 2014-09-05 | 주식회사메디아나 | Information inputting device using an electromyogram signal, and method thereof |
| CN103777752A (en) * | 2013-11-02 | 2014-05-07 | 上海威璞电子科技有限公司 | Gesture recognition device based on arm muscle current detection and motion sensor |
| CN105769202A (en) * | 2016-04-29 | 2016-07-20 | 中国科学院苏州生物医学工程技术研究所 | Multi-body-position pressure balance evaluating system |
Non-Patent Citations (2)
| Title |
|---|
| 万泽明: "脑卒中患者上肢肌肉表面肌电信号分解与疲劳研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
| 赵鹏: "基于肌电反馈的下肢康复机器人控制策略研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111973184A (en) * | 2019-05-22 | 2020-11-24 | 中国科学院沈阳自动化研究所 | Model training data optimization method for nonideal sEMG signals |
| CN111973184B (en) * | 2019-05-22 | 2021-09-03 | 中国科学院沈阳自动化研究所 | Model training data optimization method for nonideal sEMG signals |
| CN110236538A (en) * | 2019-06-21 | 2019-09-17 | 西安交通大学 | A kind of automatic real-time detection method of movement starting point based on electromyography signal |
| CN110236538B (en) * | 2019-06-21 | 2020-08-04 | 西安交通大学 | Automatic real-time detection method for motion starting point based on electromyographic signal |
| CN112773380A (en) * | 2019-11-07 | 2021-05-11 | 深圳市理邦精密仪器股份有限公司 | Electromyographic signal processing method, processing device and storage medium |
| CN112773380B (en) * | 2019-11-07 | 2023-09-22 | 深圳市理邦精密仪器股份有限公司 | Myoelectric signal processing method, processing equipment and storage medium |
| CN111696645A (en) * | 2020-06-07 | 2020-09-22 | 长春理工大学 | Hand exoskeleton rehabilitation training device and method based on surface electromyographic signals |
| CN111723717A (en) * | 2020-06-11 | 2020-09-29 | 天津大学 | Silent voice recognition method and system |
| CN112690808A (en) * | 2020-12-22 | 2021-04-23 | 浙江大学 | Human muscle fatigue identification method and system based on surface electromyographic signals and phase space reconstruction method |
| CN112690808B (en) * | 2020-12-22 | 2022-03-15 | 浙江大学 | Human muscle fatigue identification method and system based on surface electromyographic signals and phase space reconstruction method |
| CN114010453A (en) * | 2021-09-24 | 2022-02-08 | 阿凡达康复科技(东莞)有限公司 | Feedback wearing system based on motion signal and preprocessing method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107198509B (en) | 2021-01-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107198509A (en) | Feature extracting method and system based on surface myoelectric | |
| CN106984027B (en) | A kind of movement comparative analysis method, apparatus and a kind of display | |
| CN104883966B (en) | Method and system for determining ventilation threshold | |
| US10993651B2 (en) | Exercise guidance method and exercise guidance device | |
| CN104107134B (en) | Upper limbs training method and system based on EMG feedback | |
| US6411841B2 (en) | Human-related measuring assessment | |
| US11547344B2 (en) | System and method for post-stroke rehabilitation and recovery using adaptive surface electromyographic sensing and visualization | |
| Mai et al. | ECG biometric using multilayer perceptron and radial basis function neural networks | |
| CN110179643A (en) | A kind of neck rehabilitation training system and training method based on annulus sensor | |
| US20080009759A1 (en) | Cardiac monitoring system | |
| CN107137071A (en) | It is a kind of to analyze the method that heart impact signal is used for calculating short-term heart beat value | |
| CN106980746A (en) | A kind of general Woundless blood sugar Forecasting Methodology based on Time-Series analysis | |
| CN106725376B (en) | Sign detection method and device | |
| CN107198508A (en) | Rehabilitation degree sort method and combined type interactive training system | |
| CN110464347A (en) | A kind of Pelvic floor treatment method, system and device | |
| CN107518896B (en) | A kind of myoelectricity armlet wearing position prediction technique and system | |
| Zhou et al. | Qualitative and quantitative assessment of technical skills in percutaneous coronary intervention: In vivo porcine studies | |
| CN108354610A (en) | A kind of Falls Among Old People detection method and detecting system based on three-axis sensor and EGC sensor | |
| CN103886184A (en) | Construction method for heart pathology recognition model | |
| Wan et al. | A neural network to identify human subjects with electrocardiogram signals | |
| CN114298089A (en) | Multi-mode strength training assisting method and system | |
| CN111437573A (en) | Grip strength measuring device for palm muscle rehabilitation and training rehabilitation method | |
| KR100994408B1 (en) | Method and device for deducting pinch force, method and device for discriminating muscle to deduct pinch force | |
| CN113057617B (en) | Non-invasive monitoring system for cardiac output | |
| Kurniawan et al. | Arrhythmia Classification on Electrocardiogram Signal Using Convolution Neural Network Based on Frequency Spectrum |
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 | ||
| PE01 | Entry into force of the registration of the contract for pledge of patent right |
Denomination of invention: Feature extraction method and system based on surface electromyography Effective date of registration: 20230227 Granted publication date: 20210122 Pledgee: Bank of Nanjing Co.,Ltd. Changzhou Branch Pledgor: CHANGZHOU QIANJING REHABILITATION Co.,Ltd. Registration number: Y2023320000104 |
|
| PE01 | Entry into force of the registration of the contract for pledge of patent right |