Research On Brain Load Detection Based On Deep Learning Method In Virtual Reality Environment

In recent years,due to the progress of computer software and hardware,virtual reality technology has achieved rapid development,the excellent immersion has brought unprecedented experience,new products and new content have emerged in an endless stream,which showing a broad prospect of development in education,medical,military and entertainment industries.However,in virtual reality environment,the brain load of users are prone to increase,which accompanied by dizziness and fatigue.This phenomenon is also known as cybersickness.The mainstream view of increased brain load is caused by sensory conflict,namely sensory mismatch,which not only conceals the advantages of virtual reality technology,but may also harm the health of users.Therefore,it is very important to detect the user's brain load.Deep learning technology has made extraordinary achievements in many fields,and these substantial innovations have given other researchers the hope of breakthrough,previous attempts have shown that neural network is suitable for signal processing,and the structural characteristics of the network model make the end-to-end approach easier to implement.This paper uses deep learning method to detect brain load.The phenomenon of increased brain load in virtual reality scenes is closely related to the brain activities.People in different brain load levels will have different cerebral cortex electrical activities,so the brain load analysis can be carried out based on the EEG data.EEG signal classification belongs to the field of brain-computer interface,which is a form of human-computer interaction facing the future.Classify the brain load from the perspective of EEG,summarize the previous research,the brain load EEG data set was prepared through experiments,inspired by the baseline model,this paper proposes a network based on spatio-temporal synchronization convolution,give full consideration to time and space features of EEG data,preprocess the original EEG and input into network,which can extract meaningful characteristics from the EEG data,realize the brain load classification.And the learning ability of the network is verified by feature visualization analysis.Experimental results show that the classification accuracy of the proposed network is further improved compared with the baseline method.The aim is to learn cognitive representations of EEG data,to find and generate a decision boundary,and to classify brain load using low-dimensional features in EEG space.In addition,virtual reality content is closely related to the degree of brain load of users,different visual stimuli will induce different levels of brain load.Therefore,this paper makes prediction of brain load based on 360° video.Firstly,the brain load video data set was established,and then the salient features were extracted from the video as an experimental comparison.Aiming at the spatial features of a single image and the temporal features of image sequence,the brain load prediction model based on the three-dimensional convolutional neural network,and based on the residual network and the long short-term memory network is constructed respectively.Combined with the video data set and the framework,visual features can be extracted from video clips,and the prediction of brain load can be realized from the perspective of video.