One Dimensional Second Order Derivative Local Binary Pattern for Hand Gestures Classification Using sEMG Signals

Due to computational simplicity and outstanding ability of one dimensional local binary pattern (1DLBP) to capture the most representative structures of 1D signals, this operator has been recently exploited for feature extraction from biological signals. The original version of 1DLBP is obtained by first order derivative of signal and reveals its changes in time. We have improved the concept and introduced one dimensional second order derivative local binary pattern which better reveals signal changes and also exhibits convexities and concavities of the signal in time. The proposed operator has been utilized for feature extraction from EMG signals of sEMG for basic hand movement dataset and SVM has been used to classify the extracted features. The best classification accuracy of 94.9% was obtained using the combination of the first and second order derivatives. Experiments demonstrate the efficacy of the proposed feature extraction method compared to other prevalent approaches.