A triangular hashing learning approach for olfactory EEG signal recognition

• Published in: Applied soft computing Vol. 118; p. 108471
• Main Authors: Hou, Hui-Rang, Meng, Qing-Hao, Sun, Biao
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.03.2022
• Subjects: Biological signal processing; Feature extraction; Olfactory EEG; Triangular hashing approach; Feature extraction; Olfactory EEG; Biological signal processing; Triangular hashing approach
• ISSN: 1568-4946; 1872-9681
• DOI: 10.1016/j.asoc.2022.108471

Recognition of olfactory-induced electroencephalogram (EEG) signals can provide an effective means for the research on disorder diagnostics and human–machine interaction. A novel triangular hashing (TH) approach is proposed for EEG signal recognition. The TH approach consists of a triangular feature construction and a hash inspired coding idea, which makes effective use of the feature differences between EEG electrodes. Firstly, a triangular feature set with N layers is constructed based on power-spectral density (PSD) features extracted from N electrodes for each frequency band of each olfactory EEG sample. Subsequently, the electrode orders, i.e. the TH codes for each layer of the constructed feature set are obtained by arranging the feature values in ascending order. Finally, the prediction type of the testing sample is determined by finding the most similar TH codes between EEG types and the testing sample. Experimental results reveal that for the recognition of olfactory EEG signals acquired from eleven subjects, the proposed TH recognition approach yields the considerably high accuracy of 93.0%, significantly superior to the other eight traditional methods. Besides, the EEG dataset with 5005 samples used in this study is made public through the website presented in this paper. In this way, the proposed TH method combined with the published EEG dataset may provide new perspectives for further study in olfactory EEG research. •A triangular hashing (TH) recognition approach is elaborately proposed.•The proposed TH approach is essentially a model-free method, which is easy to implement.•The TH approach is easily extended to other non-EEG recognition scenarios.•The TH approach is significantly superior to other traditional methods in terms of recognition accuracy.•The EEG dataset with 5005 samples used in this study is made public.