The Intelligent Evaluation System of the Impact of Osmanthus Fragrance Olfactory Stimulation on Brainwaves Based on Deep Learning

Olfactory stimulation has been recognized as a powerful modulator of both emotional and cognitive states. However, objective neural assessments focusing on individual natural scents are still relatively limited. In this study, we propose an intelligent evaluation system that integrates deep learning to assess, in real time, the effects of the fragrance of Osmanthus fragrans (osmanthus flower) on brain activity measured through electroencephalography (EEG). The system employs a 32-channel wearable EEG device to simultaneously collect neural signals and subjective emotional ratings. After preprocessing procedures— such as band-pass filtering and artifact removal using Independent Component Analysis (ICA)—time-frequency spectrograms are extracted and input into a dual-branch CNN-Transformer model. This model is designed with a cross-subject adaptive layer and a segment-based attention mechanism, allowing for end-to-end learning along with interpretable visualizations of the neural responses. The innovations of this research are threefold. First, it introduces the first real-time, integrated EEG and deep learning framework specifically targeting the neural impact of osmanthus fragrance. Second, it incorporates a cross-subject adaptive attention mechanism to improve the generalization performance of the model across different individuals. Third, it builds a low-latency visualization platform that supports applications in aromatherapy, emotional regulation, and consumer experience evaluation. Detailed experimental design and evaluation results are presented in the subsequent sections.