A Mode-Specific Classification Based on sEMG for User-Independent Locomotion Transition Recognition

The intelligent prostheses hold the promise of assisting disabled people for independent living. Locomotion recognition can allow the intelligent prostheses to work more efficiently. However, in previous studies, the locomotion recognition neglected different transition modes and usually required a burdensome training session before using. In this paper, the mode-specific classification strategy was used to distinguish seven locomotion modes, including four different transitions. This classifier was evaluated by classification accuracy with sEMG data from ten able-bodied participants. Additionally, the user-independent recognition was discussed. The result showed that the mode-specific classification strategy for seven-class classification improved the accuracy significantly compared to the baseline strategy for traditional three-class classification. The recognition accuracies of the proposed strategy were 99.86% with user-dependent method and 95.14% with user-independent method. Further, the user-dependent and user-independent classification strategies from the thigh achieved accuracies of over 95% and 80%, respectively. The recognition rate increased with the number of the participants included in the training session. In conclusion, the mode-specific classification obtained accurate recognition of locomotion modes, which is promising for further developed intelligent prostheses to assist disabled people.