MUSCULAR CONTRACTION CLASSIFICATION USING PRINCIPAL COMPONENT ANALYSIS

1
MUSCULAR CONTRACTION CLASSIFICATIONUSING
PRINCIPAL COMPONENT ANALYSIS
D. Sueaseenak, T. Chanwimalueang, N.
Laoopugsin and C. Pintavirooj
Faculty of Medicine, Srinakharinwirot
University
Research Center for Communication and Information
Technology Department of Electronics, Faculty
of Engineering King Mongkuts Institute of
Technology Ladkrabang
2
EMG Control Prosthesis Research
Prosthesis hand
Array surface electrode
Feedback Control
Actuator Driver
Pattern Classifier
EMG Amplifier And processing
3
EMG Control Prosthesis Research Team
??.?? ???? ?????????? Faculty of
Medicine (SWU)
??.?? ?????? ??????????? Faculty
of Engineering (KMITL)
????????????? ???????????
Faculty of Engineering (KMITL)
???????? ????????? Faculty of
Medicine (SWU)
4
Outline

• EMG Overview
• EMG acquisition system
• EMG topological mapping
• Classification using PCA
• Experimental result

5
EMG Overview

• EMG Electromyography
• Electromyography measures the electrical impulses
  of muscles at rest and during contraction.
• Amplitudes of EMG signal about 10 mV
  (peak-to-peak) or 1.5 mV (rms).
• Frequency of EMG signal is between 0 to 500 Hz.
• The usable energy of EMG signal is dominant
  between 50-150 Hz.

Source http//www.delsys.com/library/papers/SEMGi
ntro.pdf
6
EMG types
Indwelling EMG
Surface EMG
7
Multi-channel EMG acquisition system
8
Final EMG acquisition board
9
PSOC MCU
RS-232
Isolate power supply
Opto isolator
16 CH AMUX
Switching power supply
EMG signal conditioning unit
Reference circuit
EMG lead wire
10
Electrode placement
Reference electrode
4.5 CM
1.5 CM
11
Modified reuseable surface electrode
1.5 CM
6 CM
12
An Experiment on CT Laboratory
13
EMG Capture Program (Single channel)
14
EMG Capture Program (Multi-channel)
15
EMG analysis
RAW EMG
Rectified EMG
Envelope of Rectified EMG
FFT
16
Process of muscular contraction classification
Multi-channel EMG data
Spectrum analysis (FFT)
Topological mapping (Spline interpolation)
Classification (PCA)
17
Process of EMG topological mapping
EMG electrode 4x4 matrix
Raw EMG channel 1-16
Spectrum analysis (FFT)
1
4
3
2
6
5
8
7
11
9
12
10
15
14
16
13
Interpolation using cubic-spline
? Area from 16 channel
EMG mapping 49x49 matrix
18
Compare interpolation methods(Wrist extension)
Linear
Nearest
Cubic
19
Topological mapping result
Hand Close
Wrist extension
Wrist flexion
Wrist pronation
20
Topological mapping result
Wrist supination
Radial flexion
Ulnar flexion
Hand open
21
Principal component analysis

• Transform each EMG mapping into a column vector
  ?i of length nx1
• size
  of F is MxN
• Mnumber of data
• N
  EMG map_width EMG map_height
• Where

22
STEP 1 Centering EMG map

• Find an average EMG map of all the training data
• Take the difference between each average EMG map
  data and the database

23
STEP 2 Whitening

• Calculate Covariance matrix
• size NxM
• size M x M
• Calculate Eigen of C (Matlab v,?eig (C))
• C is an M x M Matrix , N EMG map_width x EMG
  map_height (for EMG map_width 49, EMG
  map_height49 N 2,401)

24
Eigen data

• Use Eigenvectors to find Eigen data
• where V is matrix of which column is
  EigenVector
• (size of A is MxN, size of V is MxM, size of
  Eigen is MxN )

25
STEP 3 Projection to Eigen data

• Project EMG map data to Eigen data to get
  coefficient of each training data
• (size of Eigen data is MxN, size of A is NxM
  Hence O is of size MxM)
• where ?i is the coefficient of training data ith

26
STEP 4 Identifying the Subject

• Project test subjects EMG to Eigen data
• size of Eigen data is MxN, size of I is Nx1
• Hence ?s is of size Mx1
• Find the distance between the subject coefficient
  and the coefficient of each data in the database

27
Classification Results
28
Training Topological-Mapping Input of PCA
Eigen data
29
Ulnar flexion movement (93.3 accuracy)
30
Classifier error
31
Radial flexion movement (93.3 accuracy)
32
Classifier error
33
Guide line for future development

• For higher education
• Improve system for more efficiency.
• The classifier experiment by other technique
  ( ANN , Fuzzy ,SVM ).
• Design prosthesis hand for clinical application.

34
(No Transcript)
35
(No Transcript)
36
(No Transcript)
37
(No Transcript)
38
(No Transcript)
39
(No Transcript)
40
www.thaibme.org