Research on EEG-Based Drowsy Driving Recognition Integrating Multi-Scale Temporal Features and Brain Functional Connectivity
Drowsy driving EEG recognition faces challenges such as signal non-stationarity, imbalanced class distribution, and insufficient utilization of spatio-temporal features. To improve fatigue state recognition performance, this paper proposes an EEG-based drowsy driving recognition method that integrates multi-scale temporal features and brain functional connectivity. The proposed method takes 32-channel EEG signals from the public MPD-DF driving fatigue dataset as the research object. The raw EEG signals are first preprocessed, and a random oversampling strategy is applied only to the training set to alleviate class imbalance. In terms of model construction, a dual-branch network architecture is designed. On the one hand, a multi-scale EEGNet is employed to extract short-term, medium-term, and long-term EEG temporal features through convolutional kernels of different scales. On the other hand, brain functional connectivity matrices are constructed based on the weighted phase lag index (WPLI), and a graph convolutional network (GCN) is used to extract spatial topological features among EEG channels. Finally, the two types of features are fused and fed into fully connected layers to classify awake and fatigued states. Experimental results show that the proposed model outperforms baseline methods such as SVM, LSTM, and EEGNet, demonstrating its effectiveness in drowsy driving recognition.