A Double-Square-Based Electrode Sequence Learning Method for Odor Concentration Identification Using EEG Signals

• Published in: IEEE transactions on instrumentation and measurement 2021, Vol.70, p.1-10
• Main Authors: Hou, Hui-Rang, Meng, Qing-Hao
• Format: Article
• Language: English
• Subjects: Biological signal processing; Classification; Datasets; double-square-based electrode sequence (DSES) learning method; Electrodes; Electroencephalography; Feature extraction; Learning; Learning systems; odor concentration stimulation; Odors; Olfactory; olfactory electroencephalogram (EEG); power-spectral-density (PSD) feature; Recognition; Signal processing; Signal processing algorithms; Teaching methods; Websites
• ISSN: 0018-9456; 1557-9662
• DOI: 10.1109/TIM.2021.3090177

The aim of this study is to improve the recognition performance of olfactory electroencephalogram (EEG) signals induced by different odor concentrations, which is a new challenge in the field of olfactory EEG research. To do this, a novel double-square-based electrode sequence (DSES) learning method is elaborately proposed. First, two square-based feature sets, each with N levels, are constructed based on N power-spectral-density (PSD) features extracted from N EEG electrodes. Subsequently, the electrode sequence (ES) codes are obtained by arranging the N feature values of each level of each square feature set in ascending order. Finally, inspired by the k -nearest neighbor ( k -NN), a minimum inconsistency classification is presented to find the odor concentration class whose ES codes are the least inconsistent with those of the testing sample. In the experiment, rose and rotten odors, each with five concentrations, were used to induce olfactory EEG signals on 13 participants. Experimental results reveal that, using the proposed DSES method, considerably high classification accuracies of 94.2% and 92.9% on five different concentrations of the rose and rotten odors were obtained, respectively, significantly outperforming ten other methods. These results verify the superiority of the proposed DSES method in identifying different odor concentrations. In addition, the EEG dataset with 4550 samples is made public through the website presented in this study. In this way, the proposed DSES method combined with the published EEG dataset may further promote the development of olfactory EEG recognition research.