Deep learninig of EEG signals for emotion recognition

Abstract

Emotion recognition is an important task for computer to understand the human status in brain computer interface (BCI) systems. It is difficult to perceive the emotion of some disabled people through their facial expression, such as functional autism patient. EEG signal provides us a non-invasive way to recognize the emotion of these disable people through EEG headset electrodes placed on their scalp. In this paper, we propose a deep learning algorithm to simultaneously learn the features and classify the emotions of EEG signals. It differs from the conventional methods as we apply deep learning on the raw signal without explicit hand-crafted feature extraction. Because the EEG signal has subject dependency, it is better to train the emotion model subject-wise, while there is not much epochs available for each subject. Deep learning algorithm provides a solution with a pre-training way using three layers of restricted Boltzmann machines (RBMs). Thus, we can use epochs of all subjects to pre-training the deep network, and use back-propagation to fine tuning the network subject by subject. Experiment results show that our proposed framework achieves better recognition accuracy than conventional algorithms.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 96 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.