Efficient EMG Signal Classification Using Domain-Optimized Features

Electromyography (EMG) has been widely employed in prosthetic control, muscle performance assessment, and movement recognition for rehabilitation. Accurate Classification of EMG signals is essential for interpreting muscle activity patterns in these applications. The study presents a comparative analysis of three feature domains; time domain (TD), frequency domain (FD), and time-frequency domain (TFD) for the classification of foot movements based on EMG signals. EMG data were collected from 40 healthy participants performing dorsiflexion and plantarflexion tasks, with signals recorded from the medial head of the gastrocnemius muscle. Advanced denoising techniques were applied to preprocess the raw EMG signals, followed by feature extraction from the three domains. Feature selection was performed using the particle swarm optimization (PSO) algorithm to enhance classification performance. The selected features were then classified using support vector machine (SVM), long short-term memory (LSTM), Convolutional Neural Network (CNN) and Temporal Convolutional Network (TCN) classifiers. Results demonstrated that all three feature sets achieved effective classification performance, with the TFD features achieving the highest accuracy of 96.40% using the LSTM classifier. These findings highlighted the superiority of TFD features and the effectiveness of the PSO-LSTM framework, offering a promising approach for EMG foot movement classification and broader applications in biomedical and rehabilitation technologies.