AromaNet: Integrating Attention Mechanism with Convolutional Neural Network for Olfactory Perception Classification Using EEG Signals
Sagnik De, Prithwijit Mukherjee, Dipanjan Konar, Anisha Halder Roy
Abstract
Today, many people suffer from neurological and psychological issues. Neurological diseases are commonly characterized by a loss of olfactory sensation. Getting proper treatment for these neurologic diseases at an early stage increases the chances of recovery. A patient often is not able to comprehend the loss of olfactory perception in the early stages of neurological diseases such as Alzheimer's disease, Parkinson's disease, etc. In order to identify olfactory sensory loss early, deep learning frameworks can be very useful. The main objective of our study is to build a deep learning-based model for classifying five different aromatic stimuli, i.e., no smell, rose water's smell, perfume's smell, cinnamon's smell, and odonil's smell using the EEG signals of subjects. In this paper, an Attention mechanism-based 1D Convolution Neural Network (CNN) architecture has been introduced. The designed model successfully classified five types of olfactory stimuli using EEG signals with 97.92% accuracy. The study is novel in two respects, i.e., i) appending attention mechanism to the designed 1D CNN model to improve its performance, and ii) analyzing changes occurring in different brain regions in response to different olfactory stimuli.