基于相频特性的稳态视觉诱发电位深度学习分类模型

A deep learning method for Steady-State Visual Evoked Potential (SSVEP) classification is proposed to solve the problem that phase and frequency information are not fully used in existing deep learning models. First, the proposed model uses complex vectors of fast Fourier transform as input and operates convolution on real and imaginary vectors to learn phase information, and then utilizes the spatial attention module to enhance discriminative frequency information. Next, two-dimensional convolution and max pooling are used to extract further spatial and frequency features. Finally, fully connected layers are utilized to classify. The accuracy of proposed model can reach 81.21% in the case of cross subject, and the accuracy can be further improved to 83.17% by adding the standard sinusoidal signal templates to the training set. The results show that the proposed model achieves better performance than canonical correlation analysis algorithm.