Driver Fatigue Detection Based on Nested Residual Attention Deep Ensemble Model

Electroencephalogram (EEG) signal is one of the most reliable standards for detecting driver fatigue states. In order to achieve driver fatigue detection, this paper constructs an EEG signal classifier based on a nested residual attention deep ensemble model. Firstly, design the efficient channel attention (ECA) - convolutional neural network (CNN) - long short-term memory (LSTM) network hybrid model (ECA-CNN-LSTM) based on the nested residual attention structure. Then, train the model based on multiple feature datasets, and select sub models from the training results through F1 score. Use the weighted voting method to achieve model ensemble, obtaining a fatigue detection ensemble model with stronger robustness and detection performance. The method proposed in this paper was validated using the SEED-VIG dataset and compared with various models. The experiment results showed that the method can effectively detect driving fatigue states, with an accuracy of 96.75%. Compared with other models, it has significant advantage.