Phasor represented EMG feature extraction against varying contraction level of prosthetic control. (May 2020)

• Record Type: Journal Article
• Authors: Onay, Fatih; Mert, Ahmet
• Is Part Of: Biomedical signal processing and control. Volume 59(2020)
• Publication Date: 2020-05
• Subjects: Electromyography -- Pattern recognition -- Prosthetic hand control -- Myoelectric control -- Transradial amputees; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2020.101881
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: Phasor represented EMG feature extraction is introduced. Features are formed into phasor depending on their channel positions on the residual limb. Class seperability of the EMG features is increased for prosthetic control against varying contraction level. It yields decreased processing time by 81.83% for the classification of six hand movements with three force levels belonging to nine amputees. Abstract: This paper introduces phasor representation of electromyography (EMG) feature extraction (PRE). The well-known EMG signal analysis methods, namely root mean square (RMS), and waveform length (WL) are adopted into phasor form depending electrode placement. The values of these methods are computed from 8-channel EMG signals, and their magnitudes with respect to origin are used to construct phasor represented features in this study. The class separability of the PRE is strengthened by adding difference EMG and Euclidean distanced phasor in order to obtain improved feature set against force and electrode variations. The simulations (three schemes) are performed on publicly available EMG dataset on transradial amputees, and the results are presented in terms of accuracy and processing time considering the control strategies of a prosthetic hand. Linear (LDA), and quadratic (QDA) discriminant analysis, and k -nearest neighbor ( k -NN) classifiers are trained, and tested by the PRE features. Our method outperforms previous accuracy rates in some cases, and reaches to accuracy results of the first study using this dataset without using any reduction method. In our simulations, accuracy rates up to 71.17% (PRE with QDA) for six classes hand movements with three force levels are obtained decreasing processing time by 81.83%.