Research On Sleep Staging Technology Based On Multi-sensor Data Fusion

As people's standards of health have improved,the health condiction which reflected from the sleep stages has received more and more attention.The method of traditional sleep stage classification relies mainly on the contacted sleep monitor,which can be used to determine the condition of sleep by obtaining signals such as EEG and ECG.This method requires professional operation and is expensive.The tester needs direct contact with the instrument,which can easily make the body feel unwell.With Non-contact sleep stage monitoring method,the patients do not need to contact with any instrument and the operation is simple,which can be long-term monitoring and the cost is low.Therefore,the study of non-contact sleep stage technology is of greater significance.All night sleep situation is complicated,in order to get more comprehensive sleep data,multiple sensors have been set to collect signals simultaneously.Multiple sensors can obtain multiple types of feature data,which can be mapped in different stages of sleep,thus more sleep information is provided and the accuracy of sleep stages will be improved.Based on multi-sensors data fusion,this paper studies the non-contact stage technology of sleep stages,and the main contents are as follows:1.The basic theory knowledge of multi-sensor data fusion,the algorithm of feature level fusion and decision level fusion,and the pretreatment method of heterogeneous sensors are introduced.2.A sleep stage system based on multi-sensor data fusion is proposed and the system workflow is introduced in detail.The system is composed of radar sensor and audio sensor and the standard sleep monitor PSG.The working principle of each sensor in the system,signal acquisition and extraction of characteristic parameters are introduced.Through signal processing,the respiration,heart rate,body move,snoring and other types of characteristic parameters are obtained.3.A multi-sensor feature level fusion model based on sleep stage is presented,and in the feature level fusion model,the characteristic parameters are selected by using the ReliefF algorithm in combination with related machine learning algorithms.The average accuracy of sleep staging was 82.26%,with the highest accuracy rate of 86.06%.4.A multi-sensor decision level fusion model based on sleep stage is presented and in the decision level fusion model,the Naive Bayes classifier is combined to optimize the parameters of each classifier in the system.The average accuracy rate of sleep staging is 80.67%,with the highest accuracy rate of 84.96%.