频域注意力机制下的癫痫脑电信号分类

A Classification Method of Epileptic Electroencephalograms Under Frequency-Domain Attention Mechanism

为提高癫痫脑电信号特征分类的准确率,基于残差网络结构的深度学习,提出了一种频域注意力机制下的癫痫脑电信号分类(FDAM)算法。首先分析所提取的脑电信号特征,然后根据信号特征主要分布在时频域的幅值中的特点,通过残差网络对时频域幅值特征进行二次提取,最后为了使残差网络提取的特征集中在与分类结果相关性较大的频域,设计了一种频域注意力机制,在深度学习过程中增强该类频域的幅值特征,有效提高了癫痫脑电信号的分类准确率。采用公开数据库PhysioNet中的CHB-MIT Scalp EEG Database数据库对算法的分类性能进行了验证,实验结果表明,FDAM算法对正常状态和癫痫发作状态的脑电信号分类准确率达到98.05%,特异性为99.34%,灵敏度为96.12%。

A classification method of epileptic EEG signals under frequency domain attention mechanism(FDAM)based on a deep learning of residual network structure is proposed to improve the classification accuracy of epileptic electroencephalogram(EEG)signals.Firstly,the algorithm of extracting epileptic EEG signal features is analyzed.Then,according to the characteristics of the amplitude that the signal features are mainly distributed in the time-frequency domain,the amplitude features in the time-frequency domain are extracted twice through the residual network.Finally,in order to focus the features extracted from the residual network on the frequency domain which is more relevant to the classification results,a frequency domain attention mechanism is designed to enhance the amplitude characteristics of the frequency domain in the process of deep learning and effectively improve the classification accuracy of epileptic EEG signals.The classification performance of the proposed algorithm is tested by experiments and the experimental data are obtained from the CHB-MIT Scalp EEG Database in the open PhysioNet database.The experimental results show that FDAM algorithm can classify EEG signals in normal state and epileptic state with 98.05%in accuracy,99.34%in specificity and 96.12%in sensitivity.