TSAEEG-CL: A Tempo-Spatial Attention EEG Model With Contrastive Learning for VRMS Susceptibility Detection

The Virtual Reality Motion Sickness (VRMS) has hindered the development of virtual reality technology. One of the key factors determining the severity of a user's VRMS symptoms is the VRMS susceptibility. In order to explore how the susceptibility modulate the brain activity, and detect the VRMS susceptibility through the EEG, this paper designs a simulated flight mission and adaptive training experiment to obtain the different susceptibility from the same experimental conditions. And a tempo-spatial attention EEG model with contrastive learning (TSAEEG-CL) is proposed to detect the VRMS susceptibility. The model employs a temporal branch (temporal position embedding and multi-head self-attention) and a spatial branch (spatial position embedding and multi-head self-attention) to extract the tempo-spatial features from the EEG. Then a feature fusion layer is utilized to aggregate them. The contrastive learning used in the model consider the resting state EEG before and after the VRMS exposure task as the benchmark and target respectively. The model is trained to learn the difference between the target and benchmark to determine the VRMS susceptibility. Experimental results demonstrate that TSAEEG-CL attains 99.91 % accuracy surpassing the performance of several existing models. This framework provides quantitative assessment for VRMS susceptibility and offers new ideas for the development of VR system comfort. The model code is available at https://github.com/Sun-Burst39/TSAEEG-CL.