Spectral Estimation And Supervised Classification Technique For Real Time Electromyography Pattern Recognition

Burhan, Nuradebah (2018) Spectral Estimation And Supervised Classification Technique For Real Time Electromyography Pattern Recognition. Masters thesis, UTeM.

[img] Text (24 Pages)
Spectral Estimation And Supervised Classification Technique For Real Time Electromyography Pattern Recognition.pdf - Submitted Version

Download (637kB)
[img] Text (Full Text)
Spectral Estimation And Supervised Classification Technique For Real Time Electromyography Pattern Recognition.pdf - Submitted Version
Restricted to Registered users only

Download (2MB)

Abstract

Electromyography (EMG) signal is a biomedical signal which measures physical activity of human muscle.It has been acknowledged to be widely used in rehabilitation or recovery application system assisting physiotherapist to monitor a patient’s physical strength,function,motion and overall well-being by addressing the underlying physical issues.In application system associated with rehabilitation,a signal processing and classification techniques are implemented to classify EMG signal obtained.For real time application in the rehabilitation, the classification is crucial issue.The success of the signal classification depends on the selection of the features that represent a raw EMG signal in the signal processing.Therefore,a robust and resilient denoising method and spectral estimation technique have been acknowledged as necessary to distinguish and detect the EMG pattern.The present study was undertaken to determine the characteristic of EMG features using denoising method and spectral estimation technique for assessing the EMG pattern based on a supervised classification algorithm.In the study,the combination of time-frequency domain (TFD) and time domain (TD) were identified as the preferred denoising method and spectral estimation techniques.In the first part of study, the recorded EMG signal filtered the contaminated noise by using wavelet transform (WT) approach which implemented discrete wavelet transform (DWT) method of the wavelet-denoising signal. Subsequently,the filtered signal containing useful information was extracted by three methods  root mean square (RMS),mean absolute value (MAV),and autoregressive (AR) covariance,all of which are commonly used in TD.A comparative analysis of the three different techniques was performed based on the accuracy performance of the EMG pattern classification using linear vector quantization (LVQ) neural network.In the experimental work undertaken,six healthy subjects comprised of males and females were selected. Three sets of resistance band loads,namely 5 kg,9 kg,and 16 kg were used as a force during the biceps brachii muscle contraction in the rehabilitation exercise.Each of the subject was required to perform three levels of the arm angle positions (30˚, 90˚,and 150˚) for each set of resistance band load.The results of the experiment showed that Daubechies6 (db6) was the most appropriate DWT method including a 6-level decomposition,upholding soft rigrsure and heursure threshold rules,and a single-level threshold rescaling for the wavelet denoising signal analysis.From the three different techniques in extract feature vector as an input for LVQ classifier,the study concluded that the best system performance was the AR covariance method, where it obtained the average percentage of 95.56% for all classes in the EMG pattern recognition.

Item Type: Thesis (Masters)
Uncontrolled Keywords: Electromyography, Pattern recognition systems, Real Time Electromyography Pattern Recognition
Subjects: T Technology > T Technology (General)
T Technology > TK Electrical engineering. Electronics Nuclear engineering
Divisions: Library > Tesis > FKE
Depositing User: Mohd. Nazir Taib
Date Deposited: 23 Aug 2019 03:10
Last Modified: 17 Mar 2022 12:14
URI: http://eprints.utem.edu.my/id/eprint/23299
Statistic Details: View Download Statistic

Actions (login required)

View Item View Item