Independent component analysis (ICA) was studied to examine its effect on independent component extraction and data reduction for the hand motion identification system using multichannel surface electromyogram (SEMG) and stationary wavelet transform (SWT). The results indicated that features extracted from wavelet transformed SEMGs significantly increase the classification rate. However, these extra features extracted from different wavelet scales dramatically increase the amount of computation of the artificial neural network (ANN). When using the ICA to extract independent components, the number of training epochs that were required during the ANN training period increased. This study used the unsupervised FastICA to reduce the number of SEMG channels from 7 to 4 and achieved a data reduction rate of 43%. However the hand motion identification rates using the reduced channels are significantly lower (p < 0.05) when they are compared with the traditional method. The results demonstrate that ICA can effectively reduce the size of the neural network and in turn reduce the amount of required computation with the price of reduced identification rates.
|Number of pages||6|
|Journal||Journal of Medical and Biological Engineering|
|Publication status||Published - 2006 Mar|
All Science Journal Classification (ASJC) codes
- Biomedical Engineering