Adaptive pattern recognition of myoelectric signal towards practical multifunctional prosthesis control
Published on Oct 3, 2012 in ICIRA (International Conference on Intelligent Robotics and Applications)
· DOI :10.1007/978-3-642-33509-9_52
Towards the real-world application of multifunctional prostheses based on electromyography (EMG) signal, an unsupervised adaptive myoelectric control approach was presented in order to improve the long-time classification performance of EMG pattern recognition. The widely-used linear discriminant analysis (LDA) was improved to three new different classifiers separately termed as linear discriminant analysis with single pattern updating (SPLDA), linear discriminant analysis with multiple patterns updating (MPLDA), and linear discriminant analysis with selected data updating (SDLDA). The experimental result showed that the three new classifiers significantly outperformed the original version. MPLDA and SDLDA provided two different methods to decrease the influence of misclassification and got lower classification error rates than SPLDA. Strategies to decrease the influence of misclassification are the key to the application of unsupervised myoelectric control in the future.