Recognizing Train Drivers’ Braking Intentions with EEG Based on a Wavelet-Based Dual-Attention Network

In railway, the rapid response capability of drivers under emergency scenarios is directly related to train operation safety. To address challenges such as limited driver reaction capacity in urgent scenarios like foreign object intrusion on tracks, this paper proposes a driver state monitoring method based on single-channel electroencephalography (EEG). A proposed Wavelet-Based Dual Attention Network (WDAnet) can extract the Event-Related Potentials (ERPs) evoked by visual stimuli in emergency scenario and recognize the EEG signals in emergency braking scenario. Initially, discrete wavelet decomposition is used to extract the time-frequency features of EEG signals and reconstruct signals across different scales. Subsequently, a dual-attention collaborative mechanism is designed, where a temporal self-attention mechanism captures the time dependency of ERPs, and a wavelet fusion attention mechanism captures the relationship between multi-scale features. Finally, a CNN architecture is used for feature classification and decision-making. The WDAnet achieves accurate identification of ERPs before emergency braking by the train driver, so it can be applied to help drivers perform emergency braking in time. Experimental results on a dataset of 12 participants show that the recognition accuracy reaches 97.67%. This study lays the foundation for realizing human-machine collaborative emergency braking in the field of railway.