Development of sEMG-based robust oral motion classification method and its application to electric wheelchair operation
Interfaces based on surface electromyography (sEMG) signals are one of the important methods for non-invasively extracting the intention of a severely disabled person and supporting environmental control of wheelchairs and personal computers. However, sEMG-based interfaces generally have a common an...
Guardado en:
| Autores principales: | , , , |
|---|---|
| Formato: | article |
| Lenguaje: | EN |
| Publicado: |
The Japan Society of Mechanical Engineers
2019
|
| Materias: | |
| Acceso en línea: | https://doaj.org/article/c0dce470acf14d06917fb1d539f06331 |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| Sumario: | Interfaces based on surface electromyography (sEMG) signals are one of the important methods for non-invasively extracting the intention of a severely disabled person and supporting environmental control of wheelchairs and personal computers. However, sEMG-based interfaces generally have a common and maximum disadvantage of vulnerability to changes in electrode position. In this study, we aimed to develop a robust oral motion classification method that is robust to change in electrode position. Five healthy adult male subjects participated in this experiment. sEMG signals of the suprahyoid muscles during five oral motions (right, left, up tongue motion, jaw opening, and clenching) were measured using a boomerang-shaped 22-channel electrode adhered to the underside of the jaw. Oral motion classification from sEMG signals was performed using a support vector machine (SVM). When sEMG signals measured at a position different from the 22-channnel electrode position where the training data for SVM classifier was obtained were used as the test data, the classification accuracy of five oral motions sharply decreased from 92.0% to 72.8%. In contrast, when the 10 trials of sEMG signals obtained in advance at different electrode positions on different days were used as training data, the robustness against electrode position change was improved drastically and the mean classification accuracy of all subjects reached 90.4%. Furthermore, we developed an electric wheelchair control system that can operate based on classified motions and verified its usefulness for wheelchair operability and driving performance thorough the experiment. The results showed that the proposed method can omit the SVM training process required every time after the electrode is attached and can operate the wheelchair immediately after electrode attachment. Such advancement of interfaces eliminates the annoyance caused to the user who uses the interface on a daily basis and is expected to lead to an improvement in the quality of life. |
|---|