The application of pattern recognition techniques to Electromyography (EMG) signals has shown great potential for robust, natural, prostheses control. Despite promising development in EMG pattern recognition techniques, the non-stationary properties of these signals may render these techniques ineffective after a period of time, subsequently demanding frequent recalibration during long term use. Potentially one method to reduce the impact of non-stationary traits of EMG signals is through attempting to construct a training dataset that represents this gradual change in the signal. In this paper, we investigate the potential impact of data selection schemes for inter-day motion recognition, across a period of five days of high density data recording with an LDA classifier, and present our preliminary findings. This paper proves that training a classifier with data from several spaced points of a single day can improve its inter-day performance which subsequently supports the long term use of prosthesis. Therefore the work presented here may aid in furthering our understanding of the physiological changes in EMG signals and how they may be exploited to further improve the robustness of pattern recognition methods for long term use. © Springer International Publishing AG 2017.
|Name||Lecture Notes in Computer Science|
|Conference||International Conference on Intelligent Robotics and Applications (ICIRA 2017)|
|Period||16/08/17 → 18/08/17|
- surface electromyography (sEMG)
- hand gesture recognition
- dataset optimisation
- pattern recognition