A Motor Unit Action Potential Based Signal Decomposition Scheme for Myoelectric Pattern Recognition System

Abstract Electromyography (EMG) signal-based Pattern Recognition (PR) system uses EMG signals and their attributes to control external devices. Due to the complex nature of EMG signals, A PR system uses many computational steps before the final output. This paper presents an alternative and novel approach for signal treatment based on Motor Unit Action Potential (MUAP) identification before the feature extraction and classification stage. Initially, a calculated noise margin segregates the active EMG signals and removes the overlapping and low-amplitude muscle signals. This step helps in reducing noise from the active EMG signals. Next, we developed an algorithm to categorize primary and overlapping MUAPs based on their holding times. Segmentation of signals is performed based on the occurrence of repeating primary MUAPs and their correlation score. It results in a lesser segmentation width than the conventional method resulting in less computational time. Six prominent features are calculated from the obtained segmentation windows. The preliminary results show an encouraging performance of the features set obtained from the MUAP-based segments. The Kruskal-Wallis test shows a low p-value which signifies the distribution of different features from different movement classes.