1. An Eigen Based Feature on Time- Frequency Representation of EMG Direk Sueaseenak 1,3, Theerasak Chanwimalueang 2, Manas Sangworasil 1, Chuchart Pintavirooj 1 1 Department of Electronics, Faculty of Engineering, King Mongkut’s Institute of Technology Ladkrabang, Bangkok, Thailand 2 Biomedical Engineering Programme, Faculty of Engineering, Srinakharinwirot University, Nakhon-Nayok, Thailand 3 Faculty of Medicine, Srinakharinwirot University, Nakhon-Nayok, Thailand www.bmekmitl.org

2. Direk Sueaseenak, Theerasak Chanwimalueang, Manas Sangworasil, Chuchart Pintavirooj, “An Eigen Based Feature on Time-Frequency Representation of EMG ” IEEE-RIVF 2009, Danang University of Technology, VietNam, July 13-17, 2009 Publication & Present

3. www.bmekmitl.org Introduction to our research 1 Goal and objective of research 2 SEMG Acquisition System 3 Outline SEMG BSS 4 Feature Extraction 5 Conclusion 6

4. Company Logo Biomedical,Image,Signal and System ( Biosis LAB ) Assoc.Prof.Dr.Chuchart PintaviroojAssoc.Prof.Dr.Manas Sangworasil Member M.Eng 10 คน Ph.D 6 คน IS Mini CT Image reconstruction Face & fingerprint recognition UCT EMG Analysis and Recognition Infant Incubator EEG and BCI ECG monitor

5. www.bmekmitl.org EMG Control Prosthesis Research Team Faculty of Engineering (KMITL) Faculty of Engineering(SWU) Faculty of Medicine (SWU) Direk Sueaseenak (SWU+KMITL) Chuchart Pintavirooj Manas Sangworasil Niyom Laoopugsin Theerasak Chanwimalueang

6. www.bmekmitl.org Multi-channel EMG Pattern Classification (M.Eng Thesis) 4x4 EMG Sensor 16 channel EMG 16 ch Raw EMG 16 ch FFT EMG ∑ Area from 16 channel Spline Interpolation EMG Pattern

7. www.bmekmitl.org Hand Close Wrist extension Wrist flexion Wrist pronation Hand open Wrist supination Radial flexion Ulnar flexion Hand Movement and EMG Pattern

8. www.bmekmitl.org Muscular contraction% Accuracy 1.Wrist extension 100% 2.Wrist flexion 33% 3.Wrist pronation 53.3% 4.Hand closed 86.7% 5.Radial flexion 93.3% 6.Ulnar flexion 93.3% 7.Wrist supination 93.3% 8.Hand open 100% Classification Result

9. Disadvantage  System complexity  Impossible in real application www.bmekmitl.org

10. Goal of Research (Ph.D research) Portable EMG Signal Acquisition and Pre- processing EMG Feature Extraction EMG Classification  Minimum :EMG Measurement Channel  Maximum :Accuracy Rate of EMG Classification  No complexity for Real Application Mechanical Control Feedback Control

11. EMG Surface Electrode EMG Acquisition System FAST ICA Separation Time-Frequency Analysis Eigen based Feature Extraction Feature 1Feature 2 STFT ICA 1 STFT ICA 2 Object of Research www.bmekmitl.org

12. Surface EMG Acquisition And Measurement System Direk Sueaseenak, Theerasak Chanwimalueang, Manas Sangworasil, Chuchart Pintavirooj, “ PSOC-BASED MULTICHANNEL ELECTROMYOGRAM ACQUISITION SYSTEM WITH APPLICATION IN MUSCULAR FATIGUE ASSESSMENT ” Proceedings of ThaiBME2007, vol.1, pp. 110-114,2007. Publication www.bmekmitl.org

13. Surface EMG Acquisition System Surface Electrode Instrumentation Amplifier PSOC MCU (PGA,ADC,UART) EMG Recorder www.bmekmitl.org

14. Channel 1 Flexor carpi radialis Channel 2 Flexor carpi ulnaris SWAROMED Al/AgCl Electrode Surface EMG Placement www.bmekmitl.org

15. SEMG Signal Channel 1 Channel 2 www.bmekmitl.org

16. SEMG Blind Source Separation “Independent Component Analysis” Direk Sueaseenak, Theerasak Chanwimalueang, Manas Sangworasil, Chuchart Pintavirooj, “An Investigation of Robustness in Independent Component Analysis EMG” Proceedings of ECTI-CON2009, vol.2, pp. 1102-1105,2009. Publication www.bmekmitl.org

17. Blind Source Separation : Cocktail Party Problem The mathematical model of CPP : X 1 (t)=A 11 S 1 +A 12 S 2 X 2 (t)=A 21 S 1 +A 22 S 2 x = As s = Wx (1) (2) (3) www.bmekmitl.org

18. SEMG Blind Source Separation ICA Ch1 Ch2 www.bmekmitl.org The mathematical model of CPP : X 1 (t)=A 11 S 1 +A 12 S 2 X 2 (t)=A 21 S 1 +A 22 S 2 x = As s = Wx (1) (2) (3)

19. “Nongaussian is Independent”: Central Limit Theorem x = As s = Wx www.bmekmitl.org X 1 (t)=A 11 S 1 +A 12 S 2 X 2 (t)=A 21 S 1 +A 22 S 2

20. Measures of Nongaussianity  By kurtosis Subgaussian Supergaussian Subgaussian kurtosis < 0 Superguassian kurtosis > 0 Gaussian kurtosis = 0 (4) www.bmekmitl.org

21.  Initialize W (Set the weight vector to random values)  Newton 's method (until convergence)  Normalization G(u)=u 3 (5) Process of ICA s = Wx (6) www.bmekmitl.org

22. SEMG BSS Result Channel 1 Channel 2 ICA 1 ICA 2 www.bmekmitl.org

23. SEMG Time-Frequency Analysis www.bmekmitl.org

24. Short-Time Fourier Transform (7) Source: http://www.clecom.co.uk www.bmekmitl.org

25. STFT Result www.bmekmitl.org

26. Eigen based Feature Extraction www.bmekmitl.org

27. Concept of Moment (8) (9) www.bmekmitl.org

28. Concept of Moment (cont.) Where (10) (11) (12) www.bmekmitl.org

29. Concept of Moment (cont.) EMG Features = (13) (14) (15) (16) www.bmekmitl.org

30. Eigen Feature Extraction Result www.bmekmitl.org

31. ICA-applied EMG without ICA- applied EMG AVGSDAVGSD Wrist flexion2.27430.23791.96300.4652 Relaxation1.56950.32141.5300.4718 Quantitative measurement of robustness of ICA application www.bmekmitl.org

32. Conclusions  We used a multi-channel electromyogram acquisition system from previous work to acquire two channel surface electrodes on forearm muscles. and performed a blind signal separation by using an independent component analysis (ICA) technique.  We purposed the novel features extraction for the EMG contraction classification. Our features are based on Eigen-vector approach. The time-frequency analysis is applied on the time-frequency magnitude spectrum of the Independent component analysis EMG. The ratio between the two Eigen values are the novel features. www.bmekmitl.org

33. Simple EMG Robotic Control Experiment www.bmekmitl.org

35. Acknowledgment Office of the Higher Education Commission Faculty of Medicine SWU www.bmekmitl.org

36. Company Logo www.bmekmitl.org www.thaibme.org

37. LOGO