Research On Feature Extraction And Staging Of Sleep EEG

Sleep stage has important research significance.At home and abroad,there are still some improvements in sleep staging algorithm and feature extraction method of sleep EEG.Based on this,this paper makes some exploration and puts forward some improvements.The experimental data comes from the sleep data of 20 healthy subjects of 10 men and 10 women in Physio Net,and 41212 experimental samples are finally extracted.This paper consists of two parts: the feature extraction of sleep EEG and the construction of sleep stage model.The wavelet threshold method is used to denoise the original EEG signals.Then TME,permutation entropy,rhythmic wave and related mathematical statistical features are extracted.Finally,Random Forest,GBDT and SVM classifiers are selected as the primary classifiers of the improved stacking fusion method,and a sleep stage model based on sleep EEG feature extraction is constructed by using the improved stacking framework.Besides a large amount of experimental work,this paper has the following contributions.(1)The improved multi-scale entropy algorithm is defined as translational multi-scale entropy algorithm,TME.It solves the problem that there is a sudden change of data "breakpoint" between different "particles" in the process of coarse-grained.Under the premise of controlling other factors,the improved model improves the prediction accuracy by 0.332%.(2)Improve the traditional stacking.Based on the cross-entropy loss criterion,stacking is improved to optimize the output feature space of the 0-th test set,and the effect of stacking fusion method is improved accurately.Compared with the traditional stacking method,the prediction accuracy of the improved model is improved by 0.107% on the premise of controlling other factors.(3)The stacking fusion method is innovatively applied to EEG sleep staging scene.The effect of the improved stacking fusion model is better than that of the single model method used in previous studies,which is 2.308% better than GBDT single model,3.132% better than random forest single model,and 6.491% better than SVM single model.Under the condition of existing experimental data,the accuracy of the improved sleep stage model is 93.62%.