Emotion recognition in EEG signals using the continuous wavelet transform and CNNs

Emotions are mental states associated with changes that influence people’s behavior, thinking, and health. Emotional changes can also appear in the organs and tissues of the human body as electrical potential differences gathered as biosignals in datasets. This work proposes the classification of emotions in electroencephalographical signals, transforming these discrete signals into a time-scale representation by spectral analysis. Our approach uses the wavelet transform to obtain scalogram images of electroencephalographic signals, treating these images as the scaled distribution of energy associated with a sign. Feature extraction from the scalograms is performed using convolutional neural networks (CNNs), leading to the proposal of two classification models. The threshold values in primitive emotions define one model of four emotions and the second of eight. The data augmentation technique increases the dataset size to compensate for the extra classes added in the second CNN model. The classification results were evaluated using different performance metrics and compared with related works in the literature.