Computer Science · Computer Vision and Pattern Recognition
Enhancing emotion classification through signal fusion and wavelet-based feature extraction
Oleksandr Sukholeister, Rostyslav Nakonechny
See on the mapAbstract
Emotion recognition from physiological signals, such as EEG, ECG, and GSR, has shown promise for various applications in affective computing. This study introduces a wavelet-based approach that leverages continuous wavelet transform (CWT) to extract both image and numerical features from multi-modal physiological data. The extracted features are utilized to train and compare classification models, including a ResNet-50-based deep learning framework and traditional machine learning models like support vector machines (SVMs). Our results demonstrate that the combination of wavelet-based feature extraction and signal fusion significantly enhances emotion classification performance. Notably, the numerical features derived from CWT achieve comparable or superior accuracy to image-based features, offering insights into the effectiveness of different feature representations. This work highlights the potential of wavelet-based methods for emotion recognition and suggests pathways for future research in optimizing multi-modal signal processing and classification.