Dynamic evaluation of college students' psychological state based on multimodal physiological signal fusion and deep generation model

The dynamic assessment of the psychological state of college students is an important research direction in mental health management. In response to the problem of insufficient capture of psychological state changes by existing methods, this paper proposes a dynamic assessment method that combines multimodal physiological signal fusion and deep generation models. Firstly, collect multimodal physiological data and eliminate noise through timing synchronisation and data pre-processing techniques. Secondly, utilising a multimodal feature extraction network based on transformer structure to achieve feature fusion of physiological signals. Subsequently, an improved variational autoencoder (VAE) was designed, combined with an LSTM model, to predict the trend of psychological state changes. Technical support for real-time monitoring and tailored intervention of college students' mental health status is provided by the experimental results showing that the suggested method outperforms current methods in terms of accuracy in psychological state classification and dynamic prediction performance.