EEG-Based Multi-Modal Emotion Recognition using Bag of Deep Features: An Optimal Feature Selection Approach

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Human emotion recognition is challenging due to the non-linear nature of electroencephalogram (EEG) signals.
• Machine learning approaches are increasingly used for EEG-based emotion classification.

Purpose of the Study:

• To present an advanced signal processing method using deep neural networks (DNNs) for improved EEG-based emotion recognition.
• To introduce the Bag of Deep Features (BoDF) model for reducing feature dimensionality and enhancing classification accuracy.

Main Methods:

• EEG signals are converted into 2D Spectrograms to retain spectral and temporal information.
• A pre-trained AlexNet model extracts features from the spectrograms.
• The proposed Bag of Deep Features (BoDF) model utilizes k-means clustering to create feature vocabularies, reducing dimensionality.
• Support Vector Machines (SVM) and k-nearest neighbors (k-NN) are used for final classification.

Main Results:

• The BoDF model significantly reduces feature dimensionality.
• The proposed model achieved 93.8% accuracy on the SJTU SEED dataset and 77.4% accuracy on the DEAP dataset.
• These results surpass the performance of other state-of-the-art methods for human emotion recognition.

Conclusions:

• The developed BoDF model offers a more accurate and efficient approach to EEG-based emotion recognition.
• This method holds promise for advancing the field of affective computing and brain-computer interfaces.