Evaluation of multi-channel EMG for enhancing control in mobility assistive devices using neck/shoulder movement classification

Abstract

Surface electromyography (sEMG)-based control systems offer a promising hands-free interface for individuals with severe motor impairments, enabling intuitive operation of assistive mobility devices. However, generating a large number of reliable control commands using a minimal number of electrodes remains a significant challenge. This study introduces a novel dual-movement classification strategy that utilizes sEMG signals from only three neck muscles—right trapezius, left trapezius, and left sternocleidomastoid—to generate nine distinct control commands. Data were collected from 29 able-bodied participants and one polio-affected individual with severe upper and lower limb impairments. The able-bodied participants were evaluated under subject-independent conditions, while the polio-affected participant was assessed under both subject-independent and subject-dependent settings to simulate real-world usage. The raw sEMG signals were decomposed using Empirical Mode Decomposition (EMD), and a Genetic Algorithm (GA) was applied for feature selection. A k-Nearest Neighbors (k-NN) classifier was used for classification. The proposed system achieved ∼ 99 % accuracy using ∼ 20 GA-selected features, compared to ∼ 97 % using all 48 features. For the polio-affected participant, the system achieved 100 % accuracy under the subject-dependent method even without GA. With GA, it reached 100 % accuracy under both subject-dependent and independent conditions. These results demonstrate the system’s efficiency, robustness, and generalizability. By enabling nine reliable control commands using only three non-invasive, this work significantly advances the development of practical, scalable sEMG-based assistive technologies.