Hand Action Classification via Wavelet Reconstruction and Sub-Frame Based Feature Extraction

Classification of basic hand movements from surface electromyography (sEMG) requires extraction of important information from the signal. In this study, a very simple analysis and classification of sEMG signal are presented which includes sub-frame formation and feature extraction. At first, the signal is decomposed by wavelet transform at level 1 using db44 as the mother wavelet. Both the approximate and the detailed coefficients are then used for vector reconstruction which is then subsequently broken down into overlapping sub-frames. A feature extraction step is carried out afterward from each of these sub-frames of the reconstructed signal and also the raw sEMG data. The mean of these sub-frame features is then subjected to classification using K-nearest neighborhood (KNN) classifier in a hierarchical approach. The proposed method is tested considering 5 cross 2 cross-validation scheme on a publicly available sEMG dataset containing six different hand movements collected from two females and two males. The study includes a comparison of classification accuracy of direct feature extraction from raw data and also from wavelet coefficients before reconstruction. This research proposes a highly simplified and faster way of classification of basic hand movements by decomposition and reconstruction providing an improved accuracy compared to previous methods of similar classification.