Feature Extraction of Multichannel EMG Signals for Shoulder Joint Movement Patterns

Abstrak

Electromyography (EMG) signals provide information about muscle activity and can support rehabilitation and prosthetic control systems. This study aims to extract and analyze features of multichannel EMG signals recorded from seven shoulder joint movement patterns. EMG data were acquired using surface electrodes placed on eight dominant muscles associated with shoulder joint motion, namely Deltoid1, Deltoid2, Infraspinatus, Supraspinatus, Teres Major, Latissimus Dorsi, Pectoralis1, and Pectoralis2. The recorded movements included resting, shoulder flexion, shoulder extension, shoulder abduction, shoulder adduction, external rotation, and internal rotation. The proposed processing procedure consisted of signal acquisition, rectification, transformation into the frequency domain using Discrete Fourier Transform, and feature extraction using Linear Envelope, Modified Mean Frequency (MMNF), and Modified Median Frequency (MMDF). The results show that Linear Envelope can describe temporal energy changes in each movement pattern, while MMNF and MMDF can identify groups of similar signal patterns and distinguish several movements through specific muscle channels. Resting movement had very small amplitude changes, while active shoulder movements produced different dominant energy patterns across subjects. MMNF and MMDF produced two main similarity groups, although the distinguishing muscles differed among subjects. These findings indicate that multichannel EMG feature extraction is useful as an initial basis for shoulder movement pattern analysis; however, further development is required to improve online acquisition, automatic gain adjustment, and classification robustness.