Seafarer Fatigue Identification Based on PSO-ACLSTM

Abstract

The rapid development of the shipping industry and the increasingly complex maritime traffic pose potential risks, potentially leading to serious accidents and substantial economic losses. Identifying seafarer fatigue is crucial for maritime traffic safety. However, extracting discriminative electroencephalogram (EEG) features for fatigue identification remains challenging, because most methods overlook basic data in both temporal and spatial dimensions. This paper presents a novel EEG identification algorithm model, PSO-ACLSTM, which is based on the combination of particle swarm optimization (PSO), convolutional neural network (CNN), long short-term memory network (LSTM), and attention mechanism. The purpose of this method is to obtain efficient discrimination characteristics from EEGs and improve the precision of fatigue identification. Firstly, using EEG data collected from seafarers on a training vessel, we adopt a channel-wise attention mechanism to assign the weights to each channel, and then CNN is applied to extract the spatial information of the EEG signal. Then we incorporate a self-attention mechanism into LSTM to obtain temporal information from EEG samples. Finally, the spatial and temporal attention characteristics are applied to identify seafarer fatigue. Using the particle swarm optimization algorithm, we modify the hyperparameters of the neural network training and search for optimum parameters. Experiments proved that the model presented in this paper can increase the accuracy of seafarer fatigue identification compared with other models.