未学習クラス推定を可能とするHMMの提案と時系列筋電パターンの識別

抄録

This paper proposes a novel pattern recognition method enabling calculation of a posteriori probability for learned and unlearned classes. In this approach, probability density functions of unlearned classes are incorporated in a hidden Markov model to classify undefined classes via model parameter estimation using given learning samples. The proposed method can be applied to unlearned motion recognition with electromyogram (EMG) signals and assistances for disease diagnosis. In the experiments, forearm motion classification from EMG signals was implemented with three subjects for eight learned/unlearned motions. The proposed method achieves higher classification performance (learned motions: 90.13%; unlearned motions: 91.25%) than previous approaches. The results demonstrated the effectiveness of the technique.