CNN Confidence Estimation for Rejection-Based Hand Gesture Classification in Myoelectric Control

Convolutional neural networks (CNNs) have been widely utilized to identify hand gestures from surface electromyography (sEMG) signals. However, due the nonstationary characteristics of sEMG, classification accuracy usually degrades significantly in daily living environment involving complex movements. To further improve reliability a classifier, unconfident classifications are expected be identified and rejected. In this study, we propose novel approach estimate probability correctness for each classification. Specifically, confidence estimation model is established generate scores (ConfScore) based on posterior probabilities CNN, an objective function designed train parameters model. addition, comprehensive metric that combines true acceptance rate (TAR) rejection (TRR) proposed evaluate performance ConfScore, so tradeoff between system security control lag could fully considered. The effectiveness ConfScore verified using data public databases our online platform. experimental results illustrate can better reflect CNN than traditional features, i.e., maximum entropy vector. Moreover, observed less sensitive variations thresholds.