Single-Channel EEG Sleep Stage Classification Based on K-SVD Algorithm

Sleep stage classification based on visual inspection is non-automatic and subjective resulting in automatic sleep staging by computer is essential for sleep assessment. Especially, single-channel electroencephalogram (EEG) sleep staging has the particular advantage in wearable devices. Sparse representation classification (SRC) can achieve the classification with a liner combination of atoms in an over-complete dictionary and has been widely applied to pattern recognition. An important step of SRC is dictionary training that commonly used K-SVD algorithm has not been used in sleep EEG studies. In this study we introduce K-SVD dictionary training method based SRC into single-channel EEG sleep stage classification and compare the classification performance between the Pz-Oz channel and the Fpz-Cz channel. The results showed that K-SVD based SRC obtained 96.52%, 88.63%, 85.11%, 82.74% and 80.17% classification overall accuracy for 2-6 sleep stages. The assessment results showed that SRC got good performance in EEG sleep staging and Pz-Oz channel performed better than Fpz-Cz channel. Such method is beneficial to the research of sleep monitoring equipment and the study of sleep-related diseases.