A Lightweight Upper Limb Intention Recognition Method Based on sEMG-IMU Fusion and PCA

Abstract

To address the demand for lightweight intention recognition algorithms in home-based portable upper limb rehabilitation robots, this paper presents a lightweight upper limb motion intention recognition scheme based on bimodal fusion of surface electromyography (sEMG) and inertial measurement unit (IMU) signals. Time-domain features are extracted from sEMG and statistical features from IMU, followed by feature-level concatenation. Principal component analysis (PCA) is then applied for dimensionality reduction, and a Support Vector Machine (SVM) is employed to classify five wrist movements. Experiments on the publicly available Ninapro DB5 dataset validate that bimodal fusion enhances recognition accuracy, while PCA effectively compresses the feature space with minimal performance degradation. Key design choices, hyperparameter configurations, and performance boundaries are reported to facilitate practical deployment on resource-constrained portable rehabilitation devices.