Comparative Study of Segmentation and Feature Extraction Method on Finger Movement

Myoelectric control prostheses fingers are a popular developing clinical option that offers an amputee person to control their artificial fingers by recognizing the contacting muscle residual informs of electromyography (EMG) signal. Lower performance of recognition system always has been the main problem in producing the efficient prostheses finger. This is due to the inefficiency of segmentation and feature extraction in EMG recognition system. This paper aims to compare the most used overlapping segmentation scheme and time domain feature extraction method in recognition system respectively. A literature review found that a combination of Hudgins and Root Mean Square (RMS) methods is a possible way of improving feature extraction. To proof this hypothesis, an experiment was conducted by using a dataset of ten finger movements that has been pre-processed. The performance measurement considered in this study is the classification accuracy. Based on the classification accuracy results for the three common overlapping segmentation schemes, the smaller the window size with larger increment windows produce better accuracy but it will degrade the computational time. For feature extraction, the proposed Hudgins with RMS feature showed an improvement of average accuracy for ten finger movements by 0.74 and 3 per cent compared to Hudgins and RMS alone. Future study should incorporate more advance classification accuracy to improve the study.