Related Experiment Video
Updated: Sep 27, 2025
08:31
Conscious and Non-conscious Representations of Emotional Faces in Asperger's Syndrome
Published on: July 31, 2016
13.7K
EEG Feature Extraction and Data Augmentation in Emotion Recognition
Mahsa Pourhosein Kalashami1, Mir Mohsen Pedram1, Hossein Sadr2
1Department of Electrical and Computer Engineering, Faculty of Engineering, Kharazmi University, Tehran 15719-14911, Iran.
Computational Intelligence and Neuroscience
|April 7, 2022
Summary
This study enhances emotion recognition using electroencephalogram (EEG) data by employing Conditional Wasserstein GAN (CWGAN) for data augmentation. The method significantly boosts classification accuracy for both valence and arousal.
Area of Science:
- Neuroscience
- Machine Learning
- Biomedical Engineering
Background:
- Emotion recognition is a key challenge in Brain-Computer Interaction (BCI).
- Electroencephalogram (EEG) offers unique insights into brain activity related to emotions, surpassing other modalities like facial expressions or speech.
- High dimensionality and limited availability of EEG data hinder the development of accurate emotion recognition classifiers.
Purpose of the Study:
- To address the challenges of data scarcity and high dimensionality in EEG-based emotion recognition.
- To propose and evaluate a novel data augmentation technique using deep generative models.
- To improve the accuracy of emotion recognition models by enhancing the EEG dataset.
Main Methods:
- Feature extraction techniques were employed to manage high-dimensional EEG recordings.
- A Conditional Wasserstein GAN (CWGAN) was utilized as a deep generative model for data augmentation, specifically regenerating EEG features.
- The DEAP dataset was used to validate the proposed method.
- Standard support vector machine and deep neural network classifiers were implemented to build emotion recognition models.
Main Results:
- The application of CWGAN for data augmentation successfully generated additional EEG features.
- The inclusion of augmented data led to enhanced performance in EEG-based emotion recognition models.
- A mean accuracy increase of 6.5% for valence and 3.0% for arousal was observed after data augmentation.
Conclusions:
- Data augmentation using Conditional Wasserstein GAN (CWGAN) is an effective strategy to overcome data limitations in EEG-based emotion recognition.
- The proposed method significantly improves the accuracy of emotion recognition models, demonstrating its practical utility in BCI research.
- This approach offers a promising solution for developing more robust and accurate BCI systems for emotion detection.
