EEG-Based 3D Visual Fatigue Evaluation Using CNN

• DOI: 10.3390/electronics8111208
• Publication Date: 2019-10-25T07:20:36Z
• Authors: Kang Yue; Danli Wang
• Keywords: Morlet wavelet

Visual fatigue evaluation plays an important role in applications such as virtual reality since the visual symptoms always affect user experience seriously. Existing methods require hand-crafted features for classification, and conduct feature extraction classification a separated manner. In this paper, we designed experiment to collect electroencephalogram (EEG) signals of various levels, present multi-scale convolutional neural network (CNN) architecture named MorletInceptionNet detect using EEG input, which exploits spatial-temporal structure multichannel signals. Our adopts joint space-time-frequency scheme Morlet wavelet-like kernels are used time-frequency raw inception further extract temporal features. Then, concatenated fed fully connected layer classification. evaluation, compare our method with five state-of-the-art methods, results demonstrate that model achieve overally best performance better two widely metrics, i.e., accuracy kappa value. Furthermore, use input-perturbation network-prediction correlation maps in-depth analysis into reason why proposed outperforms other methods. The suggest is sensitive perturbation β (14–30 Hz) γ (30–40 bands. their spatial patterns high corresponding power spectral densities traditionally. This finding provides evidence hypothesis can learn time-frequency-space distinguish levels automatically.