Odor Type Recognition And Channel Selection Based On Olfactory EEG

As the oldest sensory function in the evolutionary history of organisms,smell is closely related to human memory,learning and emotions.The cerebral cortex is the most advanced nerve center and is able to assess stimuli from various organs.The study of the brain's ability to recognize different smells is of great significance in the assessment and diagnosis of olfactory dysfunction and the regulation of mood in patients with mental diseases such as depression.In recent years,the study of smell based on Electroencephalogram(EEG)signals has attracted more and more attention from scholars all over the world.The existing researches on odor type recognition based on EEG signal usually only adopt single feature and lack of feature comparison.In addition,most studies focus on the analysis of multi-channel EEG signals,which brings a lot of inconvenience to data collection and subsequent processing,and it is difficult to meet the requirements of real time portability in practical application scenarios.In view of the above problems,this is focuses on the application of different EEG characteristics and public channel selection methods in odor recognition.The main work is as follows:(1)We designed EEG experiments induced by different types of odor stimulation and EEG signal data induced by smell were collected to construct the olfactory EEG data set.(2)The wavelet energy moment characteristics were proposed for the recognition of EEG signals induced by different smells and compared with different characteristics.Based on the constructed EEG data set,the power spectral density in frequency domain features,the wavelet energy moment features in time and frequency domain features,and the approximate entropy,sample entropy and wavelet entropy of nonlinear dynamics features were selected respectively.Support vector machine,k-nearest neighbor and random forest classifier were used to identify different odor induced EEG signals.The results show that the classification and stability of the selected frequency domain features and time-frequency domain features are superior to the nonlinear dynamical features.Among them,the wavelet energy moment feature has the highest classification accuracy(91.07%),indicating that the wavelet energy moment feature can effectively recognize olfaction EEG signals.Furthermore,the classification results of the five features in different frequency bands were compared,and it was found that the classification accuracy was the highest in the Gamma band and the lowest in the Theta band.This indicates that the Gamma band contributes the most to the recognition of olfactory EEG signal,which also indirectly indicates that the Gamma band may be closely related to the brain activity of olfactory stimulation.(3)The EEG public channel selection method based on Relief F-Pearson was proposed and validated in odor category recognition.Compared with traditional channel selection method based on Relief F algorithm,the proposed public channel selection method presented in this paper in no loss or the slight loss of classification accuracy under the conditions of effectively reduce the amount of EEG channel,and obtained public channels,mainly in the frontal,parietal and temporal lobe,consistent with the olfactory EEG neurophysiology area.The results show that the Relief F-Pearson method can effectively reduce the number of redundant EEG channels,which has practical significance for improving the real time and portability of EEG devices.