EEG-based Sleep Quality Estimation In Real Scenario

Recent years,with the development of economy and the acceleration of life rhythm,people’s working hours and loads have gradually increased,which makes it easier for people to fall into the life of insufficient sleep,thus affecting work efficiency and safety.Especially in important posts requiring high attention like high-speed train driving,the impact of sleep is far-reaching and significant,and a set of accurate and reliable sleep quality estimation system is urgently needed to control human-causing risk.Previous studies focus on EEG-based sleep quality estimation in laboratory scenarios.However,due to the reality gap constituted with device performance,subject groups,experiment settings and controlled conditions,the models trained solely on laboratory data cannot generalize well to real scenarios.In this work,we take the driver of high-speed train as the subject of real scene application,and study the sleep quality evaluation based on EEG.When using real-scenario data to model and test,domain adaptation model is used to deal with individual differences among subjects.As it is very difficult and expensive to acquire and annotate data in real scene,we use high quality data acquired in laboratory for model training,and transform the pattern knowledge acquired in simulation scene into real-scenario through domain adaptation method.In addition to the traditional domain adaptation method,we also adopt a new method called Domain Adversarial Neural Network(DANN).DANN learns domain independent features through a deep network with an adversarial network structure.The experimental results show that DANN is superior to other traditional domain adaptation methods,and improves 19.55%and 23.50% to the baseline support vector machine model in accuracy of cross-subject and cross-scene tasks,respectively.