Abstract
The hand disability can limit the integration of concerned subjects in daily life activities. The life quality of impaired persons can be augmented by using prosthetic hands. However, an effective realization of prosthetic body parts is challenging. Accurate identification of desired hand gestures is not straightforward. In this paper, an effective multirate solution is proposed for automated identification of six different hand gestures by processing the Surface Electromyogram (sEMG) signals. Firstly, the signal is divided into finite-length segments. Each segment is called as an instance. Secondly, the sEMG signal is conditioned by using the noise removal linear phase filter. Onward it is downsampled and decomposed in subbands by applying the Wavelet Decomposition scheme. Afterward, the subbands are selected based on frequency content and features are mined from the selected subbands. Finally, the extracted features sets are conveyed to the machine learning-based classifiers for automated categorization of hand gestures. The system performance is tested by using a publicly available sEMG dataset. To avoid any biasness in findings the 10-fold cross-validation (10-CV) strategy is followed with multiple evaluation measures. Results show that the devised method secures a 6.26-fold dimension reduction while attaining the average values of 97% accuracy, 98% specificity, and 99% AUC during an automated categorization of the six hand gestures of the considered subject.