Design Of Feature Selection Methods For Hand Movement Classification Based On Electromyography Signals

• Main Author: Too, Jing Wei
• Format: Thesis
• Language: English; English
• Published: 2020
• Subjects: T Technology (General); TK Electrical engineering. Electronics Nuclear engineering
• Online Access: http://eprints.utem.edu.my/id/eprint/25426/1/Design%20Of%20Feature%20Selection%20Methods%20For%20Hand%20Movement%20Classification%20Based%20On%20Electromyography%20Signals.pdf; http://eprints.utem.edu.my/id/eprint/25426/2/Design%20Of%20Feature%20Selection%20Methods%20For%20Hand%20Movement%20Classification%20Based%20On%20Electromyography%20Signals.pdf; http://eprints.utem.edu.my/id/eprint/25426/; https://plh.utem.edu.my/cgi-bin/koha/opac-detail.pl?biblionumber=119773
• Institution: Universiti Teknikal Malaysia Melaka

Recently, electromyography (EMG) has received much attention from the researchers in rehabilitation, engineering, and clinical areas. Due to the rapid growth of technology, the multifunctional myoelectric prosthetic has become viable. Nevertheless, an increment in the number of EMG features has led to a high dimensional feature vector, which not only increases the complexity but also degrades the performance of the recognition system. Intuitively, the accuracy of hand movement classification will be reduced, and the control of myoelectric prosthetic will become highly difficult. Therefore, this thesis aims to solve the feature selection problem in EMG signals classification and improve the classification performance of EMG pattern recognition system. For this purpose, the feature selection (FS) method is applied to evaluate the best feature subset from a large available feature set. According to previous works, conventional FS methods not only minimize the number of features but also enhance the classification performance. However, the performances of conventional FS methods such as Binary Particle Swarm Optimization (BPSO) and Binary Grey Wolf Optimization (BGWO) are still far from perfect. Additionally, there are several limitations can be found within the conventional FS methods. In this regard, this thesis proposes five FS methods for efficient EMG signals classification. The first method is the Binary Tree Growth Algorithm (BTGA), which implements a hyperbolic tangent function to convert the Tree Growth Algorithm into the binary version. The second method is called the Modified Binary Tree Growth Algorithm (MBTGA) that applies swap, crossover, and mutation operators. The third method is the hybridization of BPSO and Binary Differential Evolution, namely Binary Particle Swarm Optimization Differential Evolution (BPSODE). The fourth method is Competitive Binary Grey Wolf Optimizer (CBGWO), which is an improved version of BGWO by utilizing the competition and leader enhancement strategies. The final method is called Pbest-Guide Binary Particle Swarm Optimization (PBPSO), which is an improved version of BPSO with a powerful personal best (pbest) guide strategy. The performances of proposed methods are tested using the EMG data of 10 healthy and 11 amputee subjects acquired from publicly NinaPro database 3 and 4. Initially, Short Time Fourier Transform (STFT) and Discrete Wavelet Transform (DWT) are used for signal processing. Afterward, several time-frequency features are extracted to form the STFT feature set and DWT feature set. Then, the proposed FS methods are employed to select the most informative feature subset. Five state-of-the-art FS methods are used to evaluate the effectiveness of proposed methods in this work. The experimental results show that proposed PBPSO contributed to a high classification accuracy of 99.84% and 84.05% on healthy and amputee datasets, which offered more accurate of hand movement classification and enables an excellent control on myoelectric prosthetic.