부산대학교 도서관

Emotions build an intrinsic and important sight of human behaviour. Facial expressions can greatly display human emotions and indicate an individual's behavioural intentions within a social situation. The detection of the state of emotion from facial expressions is becoming more diffusive with the emergence of advanced computational intelligence, In recent years, numerous types of research have been conducted on facial detection and emotion recognition using the various methods of convolutional neural networks (CNN). However, this research has been conducted to find how the DenseNet networks (DenseNet121, DenseNet169, DenseNet201) from transfer learning behave in detecting emotion from the FER2013 dataset. Our approach is to train these models separately and later build an ensemble model. Therefore, an accuracy of 71.59%, 72.01%, 72.32%, and 74.16% gain from DenseNet121, DenseNet169, DenseNet201, and ensemble model respectively. The outcome of our study is that the ensemble model (DenseNet121 + DenseNet169 + DenseNet201) can carry out better than the standalone model. This research may come in handy to solve the problem of facial expression recognition in fields 1 ike human-computer i nteraction systems and data analytics.