A Novel Hybrid Approach to Pattern Recognition of Finger Movements and Grasping Gestures in Upper Limb Amputees

Abstract

The study on pattern recognition (PR) is one of the fundamental steps for prosthetic hand control to improve the life quality of amputees. However, achieving a proper and functional PR has challenges due to the limitations of bio-signals. To this end, this paper proposes a novel hybrid approach to PR of finger movements and grasping gestures simultaneously based on mechanomyography (MMG) and force-myography (FMG) signals. In this method, first, time-domain features, time-frequency-domain features, and acoustic features are extracted from signal obtained by novel hybrid MMG-FMG sensor; second, analysis of variance (ANOVA) is used to select the best features separately, and then sequential forward feature selection (SFFS) is used to select the optimal combination of the selected features by ANOVA; and finally, the selected features are classified by neural network (ANN). To evaluate the performance of the multimode PR method by hybrid MMG-FMG signals for the accurate simultaneous classification of finger motions and grasping gestures, twelve motion patterns, including seven finger motions and five grasping gestures, from twelve participants including six transracial amputees and six healthy, are tested in the experiments. The real-time experimental results show that the hybrid MMG-FMG signals lead to high resolution, with an average classification accuracy of 91.44% ±0.77 for amputees and 92.19% ±1.12 for healthies. The results illustrate that the proposed approach reduces the difference accuracy between healthy subjects and amputees in classification. The outcomes of this study have great potential to perform onto amputees with different muscle conditions.