EEG-Based Odor Recognition Using Channel-Frequency Convolutional Neural Network

Analyzing the response of the human brain to odors is critical to assess the function of olfactory and cognition. In this paper, an EEG (electroencephalograph)-based odor perception dataset is collected from ten subjects using thirteen odors stimuli. Based on the developed dataset, we employ channel-frequency convolutional neural network (CFCNN), combined with differential entropy (DE) features from different channels and frequency bands, to classify five odors that were consistently considered pleasant by the ten subjects. Meanwhile, the k-nearest neighbor (k-NN), linear discriminant analysis (LDA), support vector machine (SVM) and back propagation neural network (BPNN) are used as competing methods. The experimental results show that CFCNN is superior to the classic baselines and yields the highest accuracy in distinguishing five pleasant odors. Furthermore, compared with other four frequency bands, the gamma band presents the best classification accuracy, proving the closed relation between the olfaction and gamma band activity of the brain.