Updated ICA Weight Matrix for Lower Limb Myoelectric Classification

In the recent past, several pattern recognition and computational intelligence methods have been applied for both upper and lower limb data. However, still there exist issues due to the complex nature of muscles in the arm and body. This research proposes a classification scheme using updated independent component analysis (ICA) to consider the individual characteristics from multichannel surface electromyography (sEMG) on lower limb (for healthy subjects and subjects with knee pathology) data. Firstly, multichannel sEMG data was decomposed into various independent components (ICs) using an updated ICA weight matrix. Secondly, time domain features were extracted using ICA separated sources (ICs). The feature reduction and selection were carried out using Fischer discriminant analysis (FDA) and later classified using linear discriminant analysis (LDA). The average classification accuracy is greater than 85% and 75% for healthy and knee pathology subjects respectively.