Depression recognition based on frequency-space domain fusion and 3D-CNN-Attention
A three-dimensional feature construction method based on spectral information is presented,in which the power values of each channel are arranged into two-dimensional feature vectors based on electrode positions.The different frequency band features are arranged into a three-dimensional integral feature tensor to extract the information in frequency domain.Meanwhile,in order to reduce the influence of volume conductor effect,functional connectivity is utilized to map the temporal electroencephalogram data to the spatial brain functional network for extracting the spatial information.By analyzing the relationship between features and target classes,a 3D-CNN-Attention network model is proposed to incorporate an Attention mechanism in 3D-CNN network to enhance the electroencephalogram feature learning capability.A series of comparative experiments on publicly available datasets show that 3D-CNN-Attention network framework outperforms other methods in depression detection,obtaining an accuracy rate of up to 96.32%.The proposed method provides an effective solution for depression detection.
万方数据
维普
