A Myoelectric Signal-Based Classification System for Elbow Flexion Movements

The control of prosthetics and exoskeletons has significantly advanced in recent years, improving the quality of life for individuals with amputations and motor disabilities through devices that mimic natural limb movements. This study investigates the classification of elbow flexion movements using myoelectric signals (MES) to develop efficient and reliable control systems. Utilizing classifiers such as Random Forest (RF) and Support Vector Machine (SVM), the research focuses on elbow flexion at angles of 0°, 30°, 60°, and 90°. The study collected myoelectric signals from the biceps brachii of eight able-bodied participants, with data acquisition involving 50-second cycle of rest and elbow flexion. Signal preprocessing involved digital filters to remove noise and artifacts, followed by feature extraction, including mean absolute value (MAV), zero-crossing rate (ZC), slope sign changes (SSC), integrated EMG (iEMG), and waveform length (WL). Results show that the classifiers achieve high accuracy, with RF demonstrating a more consistent performance across all angles with an overall accuracy of 65.42%, compared to SVM 59.45%. The highest individual accuracy achieved was 91%, highlighting the potential of these methods. The classifiers, however, showed reduced effectiveness when trained with combined data from all subjects and had notable difficulties in classifying movements at 30 degrees. This study confirms the efficacy of EMG signals for motion classification and highlights the importance of advanced filtering techniques and specific features. Additionally, the prediction response time of the system allows for the implementation of this classification process in real time using both classifiers.