Zhai Wenpeng, Zhang Xiaonei, Hou Huirang, Meng Qinghao
School of Electrical and Information Engineering, Tianjin University, Tianjin 300072, P.R.China.
Sheng Wu Yi Xue Gong Cheng Xue Za Zhi. 2020 Jun 25;37(3):399-404. doi: 10.7507/1001-5515.201910036.
Studying the ability of the brain to recognize different odors is of great significance in the assessment and diagnosis of olfactory dysfunction. The wavelet energy moment (WEM) was proposed as a feature of olfactory electroencephalogram (EEG) signal and used for odor classification. Firstly, the olfactory evoked EEG data of 13 odors were collected by an experiment. Secondly, the WEM was extracted from olfactory evoked EEG data as the signal feature, and the power spectrum density (PSD), approximate entropy, sample entropy and wavelet entropy were used as the contrast features. Finally, -nearest neighbor ( -NN), support vector machine (SVM), random forest (RF) and decision tree classifier were used to identify different odors. The results showed that using the above four classifiers, the classification accuracy of WEM feature was higher than other features, and the -NN classifier combined with WEM feature had the highest classification accuracy (91.07%). This paper further explored the characteristics of different EEG frequency bands, and found that most of the classification accuracy based on the features of γ band was better than that of the full band and other bands, among which the WEM feature of the γ band combined with the -NN classifier had the highest classification accuracy (93.89 %). The research results of this paper could provide a new objective basis for the evaluation of olfactory function. On the other hand, it could also provide new ideas for the study of olfactory-induced emotions.
研究大脑识别不同气味的能力在嗅觉功能障碍的评估和诊断中具有重要意义。小波能量矩（WEM）被提出作为嗅觉脑电图（EEG）信号的一个特征，并用于气味分类。首先，通过实验收集了13种气味的嗅觉诱发EEG数据。其次，从嗅觉诱发EEG数据中提取WEM作为信号特征，并将功率谱密度（PSD）、近似熵、样本熵和小波熵用作对比特征。最后，使用K近邻（K-NN）、支持向量机（SVM）、随机森林（RF）和决策树分类器来识别不同的气味。结果表明，使用上述四种分类器，WEM特征的分类准确率高于其他特征，并且K-NN分类器与WEM特征相结合具有最高的分类准确率（91.07%）。本文进一步探索了不同EEG频段的特征，发现基于γ频段特征的分类准确率大多优于全频段和其他频段，其中γ频段的WEM特征与K-NN分类器相结合具有最高的分类准确率（93.89%）。本文的研究结果可为嗅觉功能评估提供新的客观依据。另一方面，也可为嗅觉诱发情绪的研究提供新思路。
