Development of a New Multi-channel Electrode and Signal Processing for Surface Electromyography Signals Feature Extraction
The overall aim of the research reported in this thesis was to build a new multi-channel electrode and to investigate and develop signal processing techniques for sEMG signals, in order to enable the extractions of more useful-features. Analysis of these signal features will assist in the creation of a more effective database for diagnosing muscle ailments and conditions.
The investigation was carried out by observing the fatiguing characteristics of surface electromyography (sEMG) signals collected from the vastus lateralis muscle of the quadriceps of the dominant leg of 40 healthy participants performing an endurance (or fatiguing) task of 50% of their maximum voluntary isometric contraction (MVIC). The signals were collected using a new multi-channel electrode and analysed using an overlapping sliding window algorithm that extracted signal features of the mean frequency (MNF) and muscle fibre conduction velocity (MFCV).
The new multi-channel electrode had 11 pins and each was pre-amplified with a voltage gain of 484 and bandpass filtering from 6.8 Hz to 1.02 kHz to collect monopolar signals. The monopolar signals were then configured by the software as either linear array or Laplacian configuration. A better signal definition in terms of motor unit action potential was achieved with the Laplacian configuration.
This research also investigated a number of different signal processing techniques to extract features for classification purposes of sEMG signals during an endurance or fatiguing task. These included the use of Fast Fourier Transform, Short Time Fourier Transform and Wavelet Transform. Using the Fourier power spectrum, spectral features such as MNF, median frequency (MDF), and temporal features such as root mean square (RMS) and MFCV were determined. The results showed that, of all the signals analysed, the MNF and MDF values showed similar trends and the RMS values showed a linear relationship, which increased over the time period of the signal. The MFCV values meanwhile showed also a similar trend to those of MNF and MDF.
The MNF feature was selected over MDF as it produced a more accurate trend line that closely correlated with the measured values. Statistical analysis was performed on all 40 participants to produce the mean values for determining the range and fatigue times, and determine how much the MNF and MFCV values dropped over the contraction time. The results showed that fatigue times for the 40 participants ranged between 41.6 to 78.8 seconds when performing 50% MVIC. The mean trend lines of the MNF and MFCV features showed a drop in values in the initial and the final fatigue stage. The initial value of MNF dropped by 20.3% of the maximum value during the first 29.8% of the contraction period. The MFCV meanwhile dropped by 20.9% of the maximum value during the first 28.2% of the contraction period. The MNF value dropped by 17.4% at the final fatigue stage whereas the MFCV value dropped by 18.1%.
The findings of this research, which demonstrates new methodologies that can be used for extracting features of sEMG signals, also identify directions for future work in the field of signal classification.