Classification of EMG Signals: Using DWT Features and ANN Classifier

— This study offers a concise overview of classifying hand movements based on their kinetic and myoelectric characteristics. In this work, we propose utilizing Electromyography (EMG) signals to distinguish these movements, especially for applications like wheelchair guidance and prosthetic control. Unlike prior research on forearm-derived EMG signals, this study employs a multi-channel surface Electromyography (sEMG) signal to effectively categorize distinct movements, crucial for prosthetic control. To extract informative signal features, a two-step process is deployed, beginning with the transformation of raw EMG data using Discrete Wavelet Transform (DWT) for feature extraction. The ensuing classification task employs an Artificial Neural Network (ANN), overseen by the generation of corresponding confusion matrices (CMs) based on input features. The efficacy of our approach is validated using a human hand EMG signal dataset sourced from the UCI Machine Learning Repository. This dataset encompasses recordings from 36 subjects across 8 channels (sensors), spanning multiple days. The suggested algorithm utilizes unprocessed bipolar EMG data for both training and evaluating the performance of the neural network-based classifier. Significantly, when assessing the algorithm's performance offline, it becomes evident that the use of Frequency Domain (FD) features in sequential signal processing outperforms Standard Linear Discriminant Analysis (LDA) algorithms. The combination of the DWT and ANN results in significantly improved performance and sustained robustness of the classification algorithm. Empirical findings prove the effectiveness of this approach, achieving an accuracy of 89.9% in classifying seven distinct hand movement categories accurately. Additionally, the analysis shows an increasing classification accuracy as the dataset size increases. © (2024), (International Association of Engineers). All Rights Reserved.