Electromyography signal processing based on time and time-frequency representations for prosthesis application
Electromyography (EMG) is a technique to acquire and study the signal of skeletal muscles. Skeletal muscles are attached to the bone responsible for the movements of the human body. Regarding the vast variety of EMG signal applications such as rehabilitation of people suffering from some mobility li...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2014
|
| Online Access: | http://psasir.upm.edu.my/id/eprint/60098/1/FK%202014%2073IR.pdf http://psasir.upm.edu.my/id/eprint/60098/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Electromyography (EMG) is a technique to acquire and study the signal of skeletal muscles. Skeletal muscles are attached to the bone responsible for the movements of the human body. Regarding the vast variety of EMG signal applications such as rehabilitation of people suffering from some mobility limitations, scientists have done much research on the EMG Control System (ECS). Accordingly, using EMG signal for controlling a prosthetic hand has been developed remarkably in recent years. The ECS based on pattern recognition has been improved by using new techniques in the EMG signal processing. Some of the main concerns of the ECS are the accuracy and complexity of the system. Consequently, the development of the ECS in term of accuracy and speed is the main challenge in prosthetic control. This thesis investigates the necessity of the ECS improvements by processing the EMG signal through a pattern recognition-based control system for prosthesis application. To reach this goal, different techniques in two domains of study, time and time-frequency, had been utilized to find the optimum features for EMG analysis. Mean Absolute Value (MAV), Root Mean Square (RMS), Zero Crossing (ZC) and Waveform Length (WL) were employed as feature extraction techniques in time domain and Wavelet Transform (WT) was used in time-frequency domain. Furthermore, an optimization in wavelet analysis had been investigated using twenty mother wavelets which improved the results of the EMG feature extraction. Afterwards, two discriminant analysis classifiers, Linear Discriminant Analysis (LDA) and Quadratic Discriminant Analysis (QDA) had been utilized to differentiate the five hand movements. It is worth mentioning that eighteen healthy people had participated in EMG signal recording and different wrist motions (flexion, extension, abduction, adduction and rest) had been recorded. As a result, the output of the proposed algorithm for EMG signal processing using various techniques presented an improvements in EMG signal classification in terms of accuracy. The highest classification accuracy obtained in this research was obtained by RMS feature in time domain as 98.06%. Also, the optimizing of wavelet features yielded 97.13% accuracy by applying WT+RMS (Root Mean Square of Wavelet coefficients) as the feature. On the other hand, an investigation on data segmentation before feature extraction had revealed the segment size of EMG signal plays a significant role in EMG analysis. In this study, it was presented that the techniques of segmentation and segment size affect the classification accuracy. Based on the results of this thesis, the proposed algorithm for EMG signal processing can be applied to discriminate different hand grip postures efficiently. Overall, RMS feature was demonstrated as the optimum feature for EMG classification using QDA classifier. |
|---|