CN109800792A - Electromyography signal feature selection approach based on fuzzy C-mean algorithm and DBI - Google Patents
Electromyography signal feature selection approach based on fuzzy C-mean algorithm and DBI Download PDFInfo
- Publication number
- CN109800792A CN109800792A CN201811606346.0A CN201811606346A CN109800792A CN 109800792 A CN109800792 A CN 109800792A CN 201811606346 A CN201811606346 A CN 201811606346A CN 109800792 A CN109800792 A CN 109800792A
- Authority
- CN
- China
- Prior art keywords
- fuzzy
- dbi
- coefficient
- class
- feature
- 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
Landscapes
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The invention discloses a kind of electromyography signal feature selection approach based on fuzzy C-mean algorithm and DBI, the present invention acquires 4 tunnel electromyography signals when human body does daily behavior movement first, then the average amplitude of original signal is extracted, variance, Wilson's amplitude, autoregressive coefficient, median frequency, frequency of average power, wavelet energy coefficient, wavelet-packet energy coefficient, fuzzy entropy, arranging entropy, totally 10 myoelectricity features form myoelectricity feature pool, these features are divided with fuzzy C-mean algorithm again, n vector is divided into c ambiguity group, and seek every group of cluster centre, so that the cost function of non-similarity index reaches minimum.Finally, the DBI value after computation partition, selects the smallest 4 features of DBI as the feature for being suitble to identification daily behavior movement.Electromyography signal feature selection approach based on fuzzy C-mean algorithm and DBI can choose effective reasonable myoelectricity characteristic value.
Description
Technical field
The invention belongs to feature selecting fields, are related to a kind of electromyography signal feature selecting side based on fuzzy C-mean algorithm and DBI
Method.
Background technique
Surface electromyogram signal (sEMG) electric signal adjoint when being contraction of muscle, can by muscle surface using electrode come
Guidance and record.SEMG signal is widely used in muscle disease diagnosis, fatigue study, sport biomechanics, rehabilitation instruction at present
White silk, prosthesis control, biofeedback training control and other relevant controls etc..During electromyography signal feature extraction,
Higher want is proposed to the storage of computer and the speed of service due to data volume of electromyography signal is big, intrinsic dimensionality is high the problems such as
It asks, so how to guarantee the validity of signal and the intrinsic dimensionality for reducing signal is particularly important.Feature selecting is that mode is known
Not, the important research content in the fields such as data mining, it passes through important feature constitutive characteristic in selection primitive character set
Collection reaches reduction data dimension, while keeping or improving the purpose of genealogical classification performance.It is different from feature extraction, feature selecting
What is retained is original physical feature, therefore, can veritably reduce storage needs, measurement demand, computing cost etc..
Feature selecting using relatively broad, can be divided into filtering (Filter) in pattern-recognition and big data processing field
Type and winding the two machine-independent learning algorithm of class algorithm .Filter type algorithm of (Wrapper) type, have operation time it is short, effect
The features such as rate is high；And Wrapper type algorithm needs to be embedded in machine learning algorithm, has the characteristics that dimensionality reduction effect is good.
Denatious et al. proposes the Combined Mining method that intrusion behavior is used for using cluster, classification and correlation rule etc..
Chitrakar et al. combined application K central cluster algorithm and algorithm of support vector machine construct abnormality detection model.Srinoy etc.
People realizes the reduction of data capacity according to different degree of membership integrated use Rough Fuzzy clustering method and disappears superfluous, and can be used for more
The object of cluster especially portrays overlapping region cluster and borderline region etc., can more objectively realize to normal row
For the fuzzy partitioning with abnormal behaviour.
SEMG signal is widely used in muscle disease diagnosis, fatigue study, sport biomechanics, rehabilitation training, artificial limb
Control, biofeedback training control and other relevant controls etc..During electromyography signal feature extraction, because myoelectricity is believed
Number data volume is big, the problems such as intrinsic dimensionality is high and storage and the speed of service to computer propose higher requirement, so
How to guarantee the validity of signal and the intrinsic dimensionality for reducing signal is particularly important.
Summary of the invention
It is selected in view of the deficiencies of the prior art, the present invention proposes a kind of based on the electromyography signal feature of fuzzy C-mean algorithm and DBI
Selection method.
The present invention devises a kind of electromyography signal feature selection approach based on fuzzy C-mean algorithm and DBI for electromyography signal.
Firstly, acquisition human body gastrocnemius when doing daily behavior movement, tibialis anterior, vastus medials, musculus vastus lateralis this 4 tunnel electromyography signal, then
Extract the average amplitude of original signal, variance, Wilson's amplitude, autoregressive coefficient, median frequency, frequency of average power, small echo
Energy coefficient, wavelet-packet energy coefficient, fuzzy entropy, totally 10 myoelectricity features form myoelectricity feature pool to arrangement entropy, then equal with Fuzzy C
Value divides these features, n vector is divided into c ambiguity group, and seek every group of cluster centre, so that non-similarity refers to
Target cost function reaches minimum.Finally, DBI (Davies-Bouldin index) value after computation partition, selects DBI minimum
4 features as be suitble to identification daily behavior movement feature.Fuzzy C-Means Cluster Algorithm is because of the simple fast convergence rate of algorithm
And large data sets can be handled.DBI is one for non-supervisory fuzzy clustering Validity Index, can be with by calculating this index
Determine the most reasonable value of cluster.Electromyography signal feature selection approach based on fuzzy C-mean algorithm and DBI, it is effective reasonable to can choose
Myoelectricity characteristic value.
In order to achieve the goal above, the method for the present invention mainly comprises the steps that
1. the electromyography signal feature selection approach based on fuzzy C-mean algorithm and DBI, this method comprises the following steps:
It is total that step (1) acquires human body gastrocnemius, tibialis anterior, vastus medials, musculus vastus lateralis when human body does daily behavior movement
4 tunnel electromyography signals；Test daily behavior act include level land walk, go upstairs, going downstairs, running, standing-seat, seat-stand, stand-crouching,
Crouching-is stood, sat, and-lie, lie-sits, falls；
Step (2) extracts every road electromyography signal x that sample number is NiAverage amplitude MA,
Variance VAR,Wilson's amplitude WAMP,Wherein
U (x) indicates jump function, and T is threshold value；Autoregressive coefficient AR；Median frequency MF,fiWith
hiRespectively frequency and spectrum intensity；Frequency of average power MPF,P(fi) be basic point signal function
Rate spectrum；Wavelet energy coefficient EWT,FjIt is the coefficient of wavelet energy, K is jth layer decomposition coefficient, Wj,k
It is k-th of coefficient of jth layer decomposition coefficient；Wavelet-packet energy coefficient EWP；Fuzzy entropy FE,WhereinM defines data
Dimension, DijIt is the similarity of two samples, r is DijThe width of middle exponential function, referred to as average similarity；Entropy PE is arranged,N refers to sample points, and totally 10 myoelectricity features form myoelectricity feature pool；
Step (3) is respectively adopted Fuzzy C-Means Cluster Algorithm and walks to level land, upstairs to 10 electromyography signal feature vectors
Ladder, go downstairs, run, standing-seat, seat-stand, stand-crouching, crouching-are stood, are sat-lie, lie-sit, totally 13 class daily behaviors movement progress of falling
Clustering；It is specific as follows:
FCM is n vector xi(i=1,2 ..., n) is divided into c ambiguity group, and seeks every group of cluster centre, so that non-phase
Reach minimum like the cost function of property index；FCM is represented by
In formula, c is cluster centre number, and n is number of samples, and b is Weighted Index；aijAnd dijRespectively j-th of data point
To the degree of membership and Euclidean distance at ith cluster center, A=[aij]c×nFor fuzzy classified matrix, m=[m1,m2,...,mc]TFor
Cluster centre, mi(i=1,2 ... it is c) Blur center of i-th of ambiguity group；
Objective function is constructed using Lagrangian Arithmetic, introduces λj, j=1,2 ... n,
The smallest necessary condition of objective function is made to all input parameter derivations:
By above-mentioned two necessary condition, Fuzzy C-Means Cluster Algorithm is a simple iterative process；In batch processing mode
When operation, FCM determines cluster centre m and fuzzy classified matrix A with the following steps；Its step are as follows:
1) random number with value between 0,1 initializes fuzzy classified matrix, meets it conditional (2), b=0, iteration time
Number l=0；
2) c cluster centre m is calculated with formula (4)i, i=1,2 ... c；
3) new fuzzy classified matrix A is calculated with formula (5)；
4) for giving discrimination precision ε > 0, if | | Al+1-Al| | < ε then stops iteration, otherwise sets l=l+1, returns to step
It is rapid 2), until meet condition；
Step (4) calculates DBI to the recognition result of step (3):
Wherein SiIndicate the dispersion degree of data object in the i-th class, SjIndicate the dispersion degree of data object in jth class,
di,jIndicate the distance between the i-th class and jth class, K indicates the number of class, the i.e. quantity of cluster；
Step (5) is according to clustering as a result, and the smallest 4 feature selectings of DBI are to be suitble to identification daily behavior movement
Feature.
The present invention has a characteristic that compared with the feature selecting algorithm of existing many electromyography signals
Fuzzy C-Means Cluster Algorithm is because of the simple fast convergence rate of algorithm and can handle large data sets, and solution problem scope is wide,
It is easy to the features such as appliance computer is realized and receives the concern of more and more people, and is applied to every field.DBI is one and is directed to
Non-supervisory fuzzy clustering Validity Index determines the auxiliary tool of number of clusters for determining that test set is most suitable for being divided into several classes
Index.By calculating this index, the most reasonable value of cluster can be determined.It is selected based on the electromyography signal feature of fuzzy C-mean algorithm and DBI
Selection method can choose effective reasonable myoelectricity characteristic value.
Detailed description of the invention
Fig. 1 is implementation flow chart of the invention；
Fig. 2 is that static conversion acts sub-clustering DBI；
Fig. 3 is that gait acts sub-clustering DBI；
Fig. 4 is ADLs sub-clustering DBI.
Specific embodiment
Elaborate with reference to the accompanying drawing to the embodiment of the present invention: the present embodiment before being with technical solution of the present invention
It puts and is implemented, the detailed implementation method and specific operation process are given, but protection scope of the present invention is not limited to down
The embodiment stated.
As shown in Figure 1, the present embodiment includes the following steps:
Step 1 acquires human body gastrocnemius, tibialis anterior, vastus medials, musculus vastus lateralis totally 4 when human body does daily behavior movement
Road electromyography signal；Test daily behavior act include level land walk, go upstairs, going downstairs, running, standing-seat, seat-stand, stand-crouching,
Crouching-is stood, sat, and-lie, lie-sits, falls.
Step 2 extracts the average amplitude MA of every road electromyography signal, variance VAR, Wilson's amplitude WAMP, autoregressive coefficient
AR, median frequency MF, frequency of average power MPF, wavelet energy coefficient EWT, wavelet-packet energy coefficient EWP, fuzzy entropy FE, arrangement
Entropy PE, totally 10 myoelectricity features form myoelectricity feature pool；
Step 3, to 10 electromyography signal feature vectors, be respectively adopted Fuzzy C-Means Cluster Algorithm (FCM) level land is walked,
Go upstairs, go downstairs, running, standing-seat (chair) sits (chair)-stand, stand-seat (horizontal plane), seat (horizontal plane)-stand, stand-crouching,
Crouching-, which is stood, sat ,-lie, lie-sits, totally 13 class daily behaviors movement of falling carries out clustering.
Step 4 calculates DBI to the clustering of step 3.
Step 5, according to clustering as a result, and the smallest 4 feature selectings of DBI are to be suitble to identification daily behavior movement
Feature.
As shown in figure 4, the present invention is by calculating static conversion movement (8 class), gait movement (4 class) and comprising falling
The ADLs (13 classes: the movement of 8 class static conversions, the movement of 4 class gaits, tumble divide 1 class into) of movement, the DBI coefficient under different number of clusters,
Wherein, runic data are the smallest data of DBI (i.e. optimal classification number of clusters) in same feature group.
1 static conversion of table acts the DBI (5~10 cluster) for being divided into different number of clusters
2 gait of table acts the DBI (2~7 cluster) for being divided into different number of clusters
Fig. 1,2,3 respectively show for static conversion movement, gait movement and ADLs be respectively divided into certain number of clusters
DBI coefficient.Fig. 1 is the DBI coefficient situation that static conversion movement is classified as 5~10 classes respectively.Although frequency domain character group and entropy domain
The DBI value of feature group be always maintained at it is minimum but of slight difference to the effect for being divided into any number of clusters, for non-supervisory cluster
For be not characteristic set.Practical static conversion movement of the invention is 8 classes, thus is reached most when abscissa is 8
Small value, and there is higher DBI value to be only most reliable feature set other number of clusters.Thus, temporal signatures group and Static-
Convert feature group is preferable characteristic set.Likewise, Gait is made practical to be 4 classes, thus is being when number of clusters for table 2
Reach minimum value when 4, and having higher DBI value for other number of clusters is most reliable feature set.In Fig. 2, temporal signatures group,
ADL-Act feature group and Gait-Move feature group reach lesser DBI when number of clusters reaches 4, but only Gait-Move is special
Sign group reaches the smallest DBI when number of clusters is 4, and is above its minimum value in other number of clusters.For ADLs, due in Fig. 3
Each feature is not just like Fig. 1, and apparent DBI relationship in 2, this also demonstrates the present invention to ADLs with hierarchical classification, first to quiet
The major class of state, static conversion, gait and tumble carries out a subseries, then carries out the reasonability of respective secondary classification.
Claims (1)
1. the electromyography signal feature selection approach based on fuzzy C-mean algorithm and DBI, which is characterized in that this method comprises the following steps:
Step (1) acquires human body gastrocnemius, tibialis anterior, vastus medials, musculus vastus lateralis totally 4 tunnel when human body does daily behavior movement
Electromyography signal；Test daily behavior act include level land walk, go upstairs, going downstairs, running, standing-seat, seat-stand, stand-crouching, crouching-
- lie, lie-is stood, sits to sit, fall；
Step (2) extracts every road electromyography signal x that sample number is NiAverage amplitude MA,Variance
VAR,Wilson's amplitude WAMP,Wherein
U (x) indicates jump function, and T is threshold value；Autoregressive coefficient AR；Median frequency MF,fiAnd hiRespectively
For frequency and spectrum intensity；Frequency of average power MPF,P(fi) be basic point signal power spectrum；It is small
Wave energy coefficient of discharge EWT,FjIt is the coefficient of wavelet energy, K is jth layer decomposition coefficient, Wj,kIt is jth layer
K-th of coefficient of decomposition coefficient；Wavelet-packet energy coefficient EWP；Fuzzy entropy FE,WhereinM defines data
Dimension, DijIt is the similarity of two samples, r is DijThe width of middle exponential function, referred to as average similarity；Entropy PE is arranged,N refers to sample points, and totally 10 myoelectricity features form myoelectricity feature pool；
Step (3) to 10 electromyography signal feature vectors, be respectively adopted Fuzzy C-Means Cluster Algorithm level land is walked, is gone upstairs,
Go downstairs, run, standing-seat, seat-stand, stand, and-crouching, crouching-are stood, sat-lies, lies-sit, totally 13 class daily behaviors movement of falling is gathered
Class divides；It is specific as follows:
FCM is n vector xi(i=1,2 ..., n) is divided into c ambiguity group, and seeks every group of cluster centre, so that non-similarity
The cost function of index reaches minimum；FCM is represented by
In formula, c is cluster centre number, and n is number of samples, and b is Weighted Index；aijAnd dijRespectively j-th of data point is to i-th
The degree of membership and Euclidean distance of a cluster centre, A=[aij]c×nFor fuzzy classified matrix, m=[m1,m2,...,mc]TFor cluster
Center, mi(i=1,2 ... it is c) Blur center of i-th of ambiguity group；
Objective function is constructed using Lagrangian Arithmetic, introduces λj, j=1,2 ... n,
The smallest necessary condition of objective function is made to all input parameter derivations:
By above-mentioned two necessary condition, Fuzzy C-Means Cluster Algorithm is a simple iterative process；It is run in batch processing mode
When, FCM determines cluster centre m and fuzzy classified matrix A with the following steps；Its step are as follows:
1) random number with value between 0,1 initializes fuzzy classified matrix, it is made to meet conditional (2), b=0, the number of iterations l
=0；
2) c cluster centre m is calculated with formula (4)i, i=1,2 ... c；
3) new fuzzy classified matrix A is calculated with formula (5)；
4) for giving discrimination precision ε > 0, if | | Al+1-Al| | < ε then stops iteration, otherwise sets l=l+1, return step 2),
Until meeting condition；
Step (4) calculates DBI to the recognition result of step (3):
Wherein SiIndicate the dispersion degree of data object in the i-th class, SjIndicate the dispersion degree of data object in jth class, di,jTable
Show the distance between the i-th class and jth class, K indicates the number of class, the i.e. quantity of cluster；
Step (5) is according to clustering as a result, and the smallest 4 feature selectings of DBI are to be suitble to the spy of identification daily behavior movement
Sign.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811606346.0A CN109800792A (en) | 2018-12-26 | 2018-12-26 | Electromyography signal feature selection approach based on fuzzy C-mean algorithm and DBI |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811606346.0A CN109800792A (en) | 2018-12-26 | 2018-12-26 | Electromyography signal feature selection approach based on fuzzy C-mean algorithm and DBI |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109800792A true CN109800792A (en) | 2019-05-24 |
Family
ID=66557657
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811606346.0A Pending CN109800792A (en) | 2018-12-26 | 2018-12-26 | Electromyography signal feature selection approach based on fuzzy C-mean algorithm and DBI |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109800792A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110516762A (en) * | 2019-10-10 | 2019-11-29 | 深圳大学 | A kind of muscular states quantization assessment method, apparatus, storage medium and intelligent terminal |
| CN110765279A (en) * | 2019-11-29 | 2020-02-07 | 南通大学 | Multi-view clustering method in clothing design resource knowledge graph construction |
| CN110956155A (en) * | 2019-12-13 | 2020-04-03 | 天地（常州）自动化股份有限公司 | Fully mechanized coal mining face operation procedure fuzzy clustering identification method based on CO data |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101732110A (en) * | 2009-11-19 | 2010-06-16 | 杭州电子科技大学 | Method for fusing brain electricity and muscle electricity signal chaos characteristics for hand motion identification |
| CN105769186A (en) * | 2016-03-23 | 2016-07-20 | 河北工业大学 | Complex network based electromyographic signal acquisition position selecting method |
-
2018
- 2018-12-26 CN CN201811606346.0A patent/CN109800792A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101732110A (en) * | 2009-11-19 | 2010-06-16 | 杭州电子科技大学 | Method for fusing brain electricity and muscle electricity signal chaos characteristics for hand motion identification |
| CN105769186A (en) * | 2016-03-23 | 2016-07-20 | 河北工业大学 | Complex network based electromyographic signal acquisition position selecting method |
Non-Patent Citations (4)
| Title |
|---|
| HANPANG HUAN ET AL: "Development of a Myoelectric Discrimination System for a Multi-Degree Prosthetic Hand", 《PROCEEDINGS OF THE 1999 IEEE INTERNATIONAL CONFERENCE ON ROBOTICS & AUTOMATION》 * |
| XUGANG XI ET AL: "Evaluation of Feature Extraction and Recognition for Activity Monitoring and Fall Detection Based on Wearable sEMG Sensors", 《SENSORS》 * |
| XUGANG XI ET AL: "Feature-Level Fusion of Surface Electromyography for Activity Monitoring", 《SENSORS》 * |
| 杨桂元: "《数学建模》", 28 February 2015 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110516762A (en) * | 2019-10-10 | 2019-11-29 | 深圳大学 | A kind of muscular states quantization assessment method, apparatus, storage medium and intelligent terminal |
| CN110516762B (en) * | 2019-10-10 | 2022-11-15 | 深圳大学 | Muscle state quantitative evaluation method and device, storage medium and intelligent terminal |
| CN110765279A (en) * | 2019-11-29 | 2020-02-07 | 南通大学 | Multi-view clustering method in clothing design resource knowledge graph construction |
| CN110956155A (en) * | 2019-12-13 | 2020-04-03 | 天地（常州）自动化股份有限公司 | Fully mechanized coal mining face operation procedure fuzzy clustering identification method based on CO data |
| CN110956155B (en) * | 2019-12-13 | 2023-05-30 | 天地（常州）自动化股份有限公司 | Fuzzy clustering identification method for fully mechanized mining face operation procedure based on CO data |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Kumar et al. | Optimization driven deep convolution neural network for brain tumor classification | |
| Hameed et al. | Multi-class skin diseases classification using deep convolutional neural network and support vector machine | |
| CN104127181B (en) | Electromyographic signal tumble detection method based on WKFDA | |
| Rezaee et al. | A new cluster validity index for the fuzzy c-mean | |
| US6745156B2 (en) | Petroleum exploration and prediction apparatus and method | |
| CN109800792A (en) | Electromyography signal feature selection approach based on fuzzy C-mean algorithm and DBI | |
| CN102961203B (en) | Method for identifying surface electromyography (sEMG) on basis of empirical mode decomposition (EMD) sample entropy | |
| CN101859377A (en) | Electromyographic signal classification method based on multi-kernel support vector machine | |
| Odeh | Using an Adaptive Neuro-Fuzzy Inference System(AnFis) Algorithm for Automatic Diagnosis of Skin Cancer | |
| CN102930284A (en) | Surface electromyogram signal pattern recognition method based on empirical mode decomposition and fractal | |
| CN110363152B (en) | Method for identifying road condition of lower limb prosthesis based on surface electromyographic signals | |
| CN110826625A (en) | Finger gesture classification method based on surface electromyographic signals | |
| Del Boca et al. | Myoelectric signal recognition using fuzzy clustering and artificial neural networks in real time | |
| Liu et al. | Improved GMM with parameter initialization for unsupervised adaptation of brain–computer interface | |
| Huang et al. | A sorting fuzzy min-max model in an embedded system for atrial fibrillation detection | |
| CN114266270A (en) | Electromyographic signal decoding method based on recurrent neural network and ensemble learning | |
| CN109948640A (en) | Electromyographic signal classification method based on two-parameter core Optimization-type extreme learning machine | |
| CN109508653A (en) | A kind of subjective and objective individual combat Emotion identification method merged based on EEG signals with psychology | |
| CN112183237A (en) | Automatic white blood cell classification method based on color space adaptive threshold segmentation | |
| Wallace et al. | Intelligent initialization of resource allocating RBF networks | |
| CN107045624A (en) | A kind of EEG signals pretreatment rolled into a ball based on maximum weighted and sorting technique | |
| CN110738093B (en) | Classification method based on improved small world echo state network electromyography | |
| CN106407706A (en) | Boruta algorithm-based multi-level old people physical state quantization level calculation method | |
| Malleswari et al. | Diverse Convolutional Neural Network Models for Feature Extraction from Brain Tumor Images | |
| CN114366116A (en) | Parameter acquisition method based on Mask R-CNN network and electrocardiogram |
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: 20190524 |
|
| RJ01 | Rejection of invention patent application after publication |