| Summary: | 碩士 === 國立成功大學 === 資訊工程學系碩博士班 === 97 === People need to sleep about 7~8 hours each day. In other words, sleep takes about the 1/3 of the time one day. In such long time, the activation of the brain doesn’t stop. Contrarily, the brain carries through a series of multi-stepped and progressive activities, and produces some different types of signal in different step. Based on this, people divide the sleep into two clusters : non-rapid-eye-movement sleep (NREM), rapid-eye-movement sleep (REM), and NREM from light to deep is divided as s1, s2, SWS. On the other hand, it is found from the research that activities in different step have relation to the learning or diseases, like s2 have strong relation to memory learning. These researches prove the importance of separating sleep stages. But if doing by human beings, it consumes time and energy. Therefore, the aim of the paper is to propose a high accuracy and reliable automatic sleep staging method.
In this paper, we collect 10 records of normal subjects from each device of two, totally 20 records. Based on the analysis of the signal and the spectrum, we propose a high accuracy (88.8 % , 87.9%) and reliable automatic sleep staging method. For decreasing the effect of the individual difference on staging, we do normalization on features before the classification, and then make a decision by the hierarchical decision tree. The final result of the decision tree consists of 14 rules. Among 14 rules, wake holds 1, s1 holds 6, s2 holds 4, SWS holds 1, and REM holds 2. The staging is decided by the stage that the type of the result belongs to. For example, if the staging result is one of the types that belong to s2, the present epoch is staged as s2. Finally, according to the continuity and some other restrictions of the sleep stage, we consider the temporal contextual information and make some modifications on the proceeding results of sleep staging, getting the final answer of the automatic sleep staging.
|
|---|
