SUBJECT-INDEPENDENT ELECTROMYOGRAM CLASSIFICATION OF HAND GESTURES

ZINVI FU, HU-JINGWEI (2022) SUBJECT-INDEPENDENT ELECTROMYOGRAM CLASSIFICATION OF HAND GESTURES. Doctoral thesis, Universiti Teknikal Malaysia Melaka.

Text SUBJECT-INDEPENDENT ELECTROMYOGRAM CLASSIFICATION OF HAND GE.pdf Restricted to Registered users only Download (11MB)

Abstract

Surface electromyography (EMG) is a potential alternative to traditional human machine interface. However, despite the benefits of EMG over traditional input methods, it is not widely used, especially in a working environment. Above all, the EMG signal is biologically unique amongst individuals. As a result, EMG based input devices are not usually cross-user compatible; the classifier requires some effort to re-train when reused or applied to other users. Hence, this research aims to investigate the cross-user compatibility of the EMG signals from the lower forearm by studying the classification accuracy in a realistic environment with a high number of subjects and possible errors due to wrist position. An EMG amplifier was designed specifically for this research. Data collection was acquired with wet electrodes from the gestures of 20 subjects. To facilitate the study, the untargeted electrode placement scheme was used and an EMG amplifier circuit was also developed. After the signals were filtered, feature extraction and classification was performed to determine the independence of the signals towards forearm rotation and also hand-exchange. Generally the subject-independence classification is high (70%) for all gestures. However when rotation was introduced, the classification accuracy dropped in the finger gestures by more than 10%. In the hand-exchange test, it was concluded that the while the wrist gestures are robust and can maintain more than 70% accuracy, the finger gestures are not suitable for between-hand exchange (<50% classification accuracy). In addition, 11 features and 3 classifiers were compared, and the linear envelope feature with discriminant analysis (LDA) classifier provided the most user and rotation-invariant results. Therefore, in a user-independent setting, the EMG signals can be classified reliably in forearm rotation but not hand-exchange. However, user-independent classification accuracy can be improved by using wrist gestures instead of finger gestures.

Item Type:	Thesis (Doctoral)
Subjects:	Technology > Manufactures
Depositing User:	ENCIK SAIFUL FADZLY JAMALUDIN
Date Deposited:	27 Jul 2026 16:19
Last Modified:	27 Jul 2026 16:19
URI:	https://repositori.mohe.gov.my/id/eprint/373

Actions (login required)

View Item