A Facial EMG Data Analysis for Emotion Classification based on Spectral Kurtogram and CNN

Abstract

The current digital signal processing techniques are not completely effective because of the presence of different noises, added during the acquisition of facial electromyogram (FEMG) signals, and therefore, FEMG data analysis needs strong artificial intelligent techniques along with the advanced signal processing tools. A novel approach for the emotion classification from FEMG signals is proposed in this paper which is based on the kurtogram analysis and convolutional neural network (CNN). Kurtogram is time-frequency energy density pattern gives additional information regarding frequency contents. In this work, the kurtogram considered as an input vector introduced to CNN for emotion classification. The different levels of kurtogram are considered to find the best feature set for an efficient result. Facial EMGs for five different facial emotions were recorded using two sensor wireless data acquisition device. 93% classification accuracy was achieved using our proposed method. The classification results show that the proposed method effectively differentiates the emotion class compares to other methods (SVM, ANN, and KNN).