Related Experiment Videos
RGB-Style Input Representations for EEG: Evaluating Spatial Concatenation Versus Band-Wise Stacking in Deep Emotion
1Tianjin Key Laboratory of Wireless Mobile Communications and Power Transmission, Tianjin Normal University, Tianjin 300387, China.
Brain Sciences
|July 28, 2026
Summary
New methods transform electroencephalography (EEG) signals into images for emotion recognition. These approaches improve convolutional neural network performance, offering efficient solutions for EEG-based emotion detection.
Area of Science:
- Neuroscience
- Machine Learning
- Signal Processing
Background:
- Electroencephalography (EEG) is crucial for emotion recognition.
- Integrating diverse frequency and spatial features in EEG presents a significant challenge.
Purpose of the Study:
- To propose novel preprocessing methods for mapping EEG signals into image-style representations.
- To enhance feature extraction for improved emotion recognition using convolutional neural networks (CNNs).
Main Methods:
- Spatial Concatenation Method (SCM): Projects features onto color channels, preserving spatial topology for CNNs.
- Band-wise Stacking Method (BSM): Represents frequency bands as depth frames in a 3D tensor for CNN analysis.
- Developed dedicated CNN architectures tailored to SCM and BSM tensor structures.
Main Results:
- Both SCM and BSM achieved competitive results on DEAP and DREAMER datasets for Arousal and Valence classification.
- BSM demonstrated higher accuracy than SCM on the DREAMER dataset.
- SCM and BSM performance was comparable on the DEAP dataset.
Conclusions:
- The proposed SCM and BSM strategies provide efficient CNN-based approaches for EEG emotion recognition.
- These methods effectively leverage spatial and spectral information for enhanced emotion detection.