基于特征融合的ARMA短时睡眠状态分析
Sleep Level Evaluation by Feature Fusion and ARMA for Nap
-
摘要: 针对短时睡眠的特点,结合自回归-移动平均模型(Auto-Regressive and Moving Average Model,ARMA)对短时睡眠过程中的睡眠状态变化进行分析研究。以白天短时睡眠中记录的脑电信号为研究对象,首先,从脑电信号中提取了3个与短时睡眠过程相关的特征参数,采用条件概率方法对特征参数进行融合处理,计算得到一个表征睡眠状态的参数;然后,通过ARMA模型分析睡眠状态的变化趋势;最后,采用支持向量机(Support Vector Machine,SVM)方法将整个短时睡眠过程进行了睡眠状态的自动判别,并与人工判别进行比较。结果表明,基于特征融合的ARMA模型显著提高了睡眠状态分析的准确率,7组测试数据得到的平均准确率为88.7%。一方面,特征融合能够有效地提高数据处理速度,为睡眠状态实时检测提供有利的数据处理方式;另一方面,ARMA模型的预测作用,能够分析受试者的睡眠状态变化趋势,为进一步调整和控制睡眠时长提供客观评价依据。Abstract: According to the characteristic of nap,this work proposes a sleep level estimation method based on ARMA model for analyzing the sleep status varying in nap.By using the sleep data during day nap,3 relevant parameters are calculated from Electroencephalogram(EEG),which are further fused into one parameter via the conditional probability for describing different sleep levels.And then,Auto Regressive and Moving Average (ARMA) model is adopted to analyze the sleep tendency.Finally,Support Vector Machine(SVM) is utilized to classify the sleep progress automatically.Compared with the visual inspection,the proposed estimation method can raise the sleep level recognition up to the average 88.7% of all 7 subjects.On one hand,feature fusion can improve the calculation speed significantly and provide an effective method for real-time sleep level detection.On the other hand,the prediction feature of ARMA model can be utilized to analyze the sleep tendency and provide an objective evaluation for further adjusting and controlling the sleep duration.
下载: