[Paper Review] Recognition of basic hand movements using Electromyography

Christos Sapsanis|arXiv (Cornell University)|Oct 23, 2018

Muscle activation and electromyography studies5 references17 citations

TL;DR

This study proposes a system for recognizing six basic hand movements using surface electromyography (sEMG) signals. It employs empirical mode decomposition (EMD) for noise reduction, RELIEF and PCA for feature selection and dimensionality reduction, achieving over 80% classification accuracy with a commercial Delsys system and 75% with a custom Arduino-based EMG glove.

ABSTRACT

The aim of this work was to identify six basic movements of the hand using two systems. Being an interdisciplinary topic, there has been conducted studying in the anatomy of forearm muscles, biosignals, the method of electromyography (EMG) and methods of pattern recognition. Moreover, the signal contained enough noise and had to be analyzed, using EMD, to extract features and to reduce its dimensionality, using RELIEF and PCA, to improve the success rate of classification. The first part uses an EMG system of Delsys initially for an individual and then for six people with the average successful classification, for these six movements at rates of over 80%. The second part involves the construction of an autonomous system EMG using an Arduino microcontroller, EMG sensors and electrodes, which are arranged in an elastic glove. Classification results in this case reached 75% of success.

Motivation & Objective

• To develop a reliable method for classifying six basic hand movements using surface electromyography (sEMG) signals.
• To address the challenge of high noise levels in sEMG signals through effective preprocessing and feature extraction.
• To design and evaluate a low-cost, autonomous EMG system using an Arduino microcontroller and wearable electrodes.
• To compare classification performance between a high-end commercial EMG system and a custom-built prototype.
• To improve classification accuracy through dimensionality reduction using RELIEF and PCA.

Proposed method

• Acquired sEMG signals from forearm muscles using a Delsys EMG system and a custom Arduino-based wearable glove with surface electrodes.
• Applied empirical mode decomposition (EMD) to decompose noisy sEMG signals into intrinsic mode functions (IMFs) and extract meaningful features.
• Used the RELIEF algorithm to rank and select the most discriminative features based on their ability to distinguish between hand movement classes.
• Applied principal component analysis (PCA) to reduce feature dimensionality while preserving maximum variance in the data.
• Classified hand movements using a pattern recognition approach, with performance evaluated across multiple subjects.
• Designed and implemented a low-cost, autonomous EMG system using an Arduino microcontroller, EMG sensors, and an elastic glove for wearable signal acquisition.

Experimental results

Research questions

• RQ1 Can sEMG signals reliably classify six basic hand movements using both commercial and custom-built systems?
• RQ2 How effective is empirical mode decomposition (EMD) in reducing noise and extracting features from raw sEMG signals?
• RQ3 To what extent do feature selection (RELIEF) and dimensionality reduction (PCA) improve classification accuracy?
• RQ4 What is the performance gap between a high-precision commercial EMG system and a low-cost, Arduino-based prototype?
• RQ5 Can a wearable, autonomous EMG system achieve clinically or practically useful classification accuracy for hand movement recognition?

Key findings

• The Delsys-based system achieved an average classification accuracy of over 80% across six hand movements for both a single individual and six subjects.
• The custom Arduino-based EMG system achieved a classification accuracy of 75%, demonstrating feasibility of low-cost wearable systems.
• EMD effectively reduced noise and extracted relevant signal components from raw sEMG signals, improving feature quality.
• The combination of RELIEF and PCA significantly reduced feature dimensionality while maintaining high classification performance.
• The study confirmed that sEMG is a viable modality for real-time hand movement recognition with appropriate preprocessing and feature engineering.
• The results validate the design of a low-cost, wearable EMG system suitable for applications in prosthetics, rehabilitation, and human-computer interaction.