Hand gesture recognition system using single-mixture source separation and flexible neural trees. (July 2014)

• Record Type: Journal Article
• Authors: Guo, Yina; Wang, Qinghua; Huang, Shuhua; Abraham, Ajith
• Is Part Of: Journal of vibration and control. Volume 20:Number 9(2014)
• Page Start: 1333
• Page End: 1342
• Publication Date: 2014-07
• Subjects: Hand gesture recognition -- blind source separation -- independent component analysis -- principal component analysis -- ensemble empirical mode decomposition -- flexible neural trees; Vibration -- Periodicals; Damping (Mechanics) -- Periodicals; 620.3
• DOI: 10.1177/1077546313481001
• Languages: English
• ISSNs: 1077-5463

Abstract

Surface Electromyography (sEMG) is widely used in evaluating the functional status of hands to assist in hand gesture recognition in many fields of treatment and rehabilitation. Multi-channel parallel interfaces (MCPIs) or time-division multiple access (TDMA) interfaces are the main technologies for the man–machine communication medium of sEMG recognition instruments. However, they can also result in a complex circuit connection and noise interference. A hand gesture recognition model based on sEMG signals by using single-mixture source separation and flexible neural trees (FNTs) is a breakthrough model of hand gesture recognition designed to conquer the above defects. It distinguishes itself from the traditional MCPI or TDMA interfaces by more accurate and reliable measurements of signals. Single-mixture source separation by use of ensemble empirical mode decomposition (EEMD), principal component analysis (PCA) and independent component analysis (ICA) is a novel single-input multiple-output (SIMO) blind separation method, which can simplify the two interfaces described above. The FNT model is generated and evolved based on the pre-defined simple instruction sets, which can solve the highly structure dependent problem of the artificial neural network. The testing has been conducted using several experiments conducted with five participants. The EEMD-PCA-ICA algorithm can blind separate single mixed signals with higher cross-correlation and lower relative root mean squared error. The results indicate that the model is able to classify four different hand gestures up to 97.48% accuracy.