Olfactory Emotion Recognition Using EEG Spectral Topographic Heatmaps and CNNs

Abstract

Recognizing emotions evoked by olfactory stimuli using electroencephalogram (EEG) signals is a growing area of interest in affective computing and neuromarketing. This study proposes a lightweight framework that classifies EEG responses to scented and unscented conditions along arousal and pleasure dimensions using single-band spectral topographic heatmaps and a convolutional neural network (CNN). EEG signals were recorded from 57 participants and processed to generate frequency-specific scalp maps for delta (1-4 Hz), theta (4-8 Hz), alpha (8-14 Hz), beta (14-30 Hz), and gamma (> 30 Hz) bands. These heatmaps were used as CNN inputs, achieving up to 98.39% accuracy for arousal (Theta band) and 96.55% for pleasure (Beta and Theta band). The results demonstrate that spectral-domain features alone provide highly discriminative information for olfactory emotion recognition, while reducing computational complexity compared to multi-domain approaches. This approach shows strong potential for real-time neuromarketing, affective computing, and brain-computer interface applications.