马氏田口施密特度量学习算法研究

Research on Mahalanobis-Taguchi-Gram-Schmidt Metric Learning Algorithm

人工判别人体睡眠质量一直存在着主观性过强,步骤繁琐以及准确性不高等局限性,此外由于人体睡眠异常检测研究的对象主要是人的脑电信号,传统意义上对大数据量的脑电信号仅凭经验观察和实验的方法已很难从中提取有效信息。对此,文中将马氏田口系统判别方法引入到生物计算学领域,提出一种基于马氏田口模型的睡眠质量自动检测算法。该算法通过对人脑双通道脑电信号进行建模分析,整个人睡眠的脑电信号已被人工睡眠专家标记成了六种状态的睡眠周期,算法在不同睡眠周期下,求取各个通道的标准化向量,同时对线性独立向量组进行施密特正交化处理,运用马氏田口施密特正交化方法计算出各个睡眠周期的信噪比均值,以判别出睡眠质量正常者和异常者。实验结果表明,该算法可以有效地对睡眠质量的正常和异常进行检测。

Artificial judgment of human sleep quality is always subjective,cumbersome and inaccurate.In addition,since human sleep anomaly detection is mainly performed on human electroencephalogram (EEG),it is difficult to extract effective information from large amounts of data on electroencephalogram by empirical observation and experimental methods.Therefore,we introduce the Mahalanobis-Taguchi system method into the field of bio-computing,and propose an automatic sleep quality detection algorithm based on that.Through the modeling and analysis of human brain dual-channel EEG signals, the entire human sleep EEG signal has been labeled as six- state sleep stages by artificial sleep experts.Under different sleep stages,the normalized vector of each channel is obtained.At the same time,the linear independent vector group is subjected to Gram-Schmidt orthogonalization,and the mean value of the signal-to-noise ra- tio of each sleep phase is calculated by using the Mahalanobis-Taguchi-Gram-Schmidt method.So the normal quality of sleep and abnormal people can be identified.The experiment shows that the proposed algorithm can effectively detect the normal and abnormal quality of sleep.