Find in Library

Deep learning networks have a high calculation volume, which is one of their problems. To solve this defect, the data of intrinsic modes obtained from the application of empirical mode decomposition to Electroencephalograph signals were used for the first time in this study. The present paper presents a method for emotion recognition using a deep learning network and electroencephalogram signal. Based on the non-stationary nature of the electroencephalogram, the intrinsic mode functions are extracted using empirical mode decomposition before selecting the first three intrinsic mode functions. Then, electrode positions are converted into pixel positions in images using suitable mapping, and the extracted features are interpreted as pixel color components. Using a deep learning network, all generated images are input into the network to determine whether theymbelong to the high or low valence class. Similarly, the class of arousal has been determined using the same method. This method was evaluated using the DEAP database to assess its efficiency. The results show that by selecting the image with the size of 17 × 17, the proposed method can detect valence and arousal emotions with an accuracy of 82.3% and 78.4%, respectively, which is an acceptable superiority compared to previous research.