A Deep Learning Scheme for Hand Motion Classification Based on Electromyography Signal from Transradial Amputee
Triwiyanto
Department of Electromedical Engineering, POLTEKKES KEMENKES Surabaya
Abstract
Improving accuracy in the hand motion classification plays an essential role in the development of the prosthetics hand. Electromyography (EMG) signal has been used widely to control the prosthetics hand for amputee person. However, in the development of the prosthetics hand for amputee using EMG signal, generally, shows low accuracy in the hand motion classification. Therefore, the objective of this study is to evaluate hyper-parameters in the deep learning algorithm to obtain a high accuracy in the hand motion classifier based on EMG signal. Ten channels of EMG signal from amputee was extracted using root mean square (RMS) to reduce the complexity of the EMG characteristics. The EMG signal from amputee was obtained from ninapro dataset. A deep learning algorithm with a convolution neural network method was applied to classify 18 hand motion from transradial amputee person. The value of the hyper-parameters of the deep learning was investigated, in order, to obtain the best accuracy of the deep learning classifier. After the evaluation, we found that the accuracy improved significantly than conventional classifier machine learning. The highest mean accuracy is 89.6% for 18 class. In order to obtain a high accuracy in the hand motion classification, this study suggested to use, this proposed method in the development of the prosthetics hand
Keywords: Electromyography, deep learning, hand motion classification, feature extraction
Topic: Electrical and Computer Engineering