A Lightweight Sleep Staging Model Based on Single-channel Context Encoding
Utilizing the frequency relationships between different signals in multimodal signals to handle sleep staging tasks has become the mainstream of today.However,the difficulty of multimodal signal acquisition limits further development of sleep staging tasks.In addition,the unbalanced classification of the data set and the large magnitude of parameters are also important reasons that limit the development of sleep staging tasks.In response to the above problems,a fully convolutional network model based on single-channel Electroencephalogram(EEG)data is proposed.The waveform information of EEG signals is learned through the U-shaped structure and feature fusion module.The parallel multi-scale time features are used to instead the recurrent networks to extract time series information.The bottleneck structure is used to reduce the amount of parameters,and the focal loss function is used to reduce the impact of uneven distribution between data categories on classification accuracy.Experimental results on the public data set Sleep-EDF show that the proposed model achieves a classification accuracy of 87.5％for single-channel EEG data,alleviating the problem of imbalanced data set.It is worth mentioning that the number of parameters of the proposed method is only 2.86％of which of the Deepsleep network.
EEG signalsleep stagingdilated convolutionclass imbalancefocal loss function
国家科技期刊平台
NSTL
万方数据
